maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/lib/Target/X86/X86SchedSandyBridge.td

2781 lines
124 KiB
TableGen
Raw Normal View History

Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
//=- X86SchedSandyBridge.td - X86 Sandy Bridge Scheduling ----*- tablegen -*-=//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the machine model for Sandy Bridge to support instruction
// scheduling and other instruction cost heuristics.
//
//===----------------------------------------------------------------------===//
def SandyBridgeModel : SchedMachineModel {
// All x86 instructions are modeled as a single micro-op, and SB can decode 4
// instructions per cycle.
// FIXME: Identify instructions that aren't a single fused micro-op.
let IssueWidth = 4;
Update machine models. Specify buffer sizes for OOO processors. llvm-svn: 184033
2013-06-15 12:50:02 +08:00
let MicroOpBufferSize = 168; // Based on the reorder buffer.
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
let LoadLatency = 4;
let MispredictPenalty = 16;
Mark the x86 machine model as incomplete. PR17367. Ideally, the machinel model is added at the time the instructions are defined. But many instructions in X86InstrSSE.td still need a model. Without this workaround the scheduler asserts because x86 already has itinerary classes for these instructions, indicating they should be modeled by the scheduler. Since we use the new machine model for other instructions, it expects a new machine model for these too. llvm-svn: 191391
2013-09-26 02:14:12 +08:00
Move late partial-unrolling thresholds into the processor definitions The old method used by X86TTI to determine partial-unrolling thresholds was messy (because it worked by testing target features), and also would not correctly identify the target CPU if certain target features were disabled. After some discussions on IRC with Chandler et al., it was decided that the processor scheduling models were the right containers for this information (because it is often tied to special uop dispatch-buffer sizes). This does represent a small functionality change: - For generic x86-64 (which uses the SB model and, thus, will get some unrolling). - For AMD cores (because they still currently use the SB scheduling model) - For Haswell (based on benchmarking by Louis Gerbarg, it was decided to bump the default threshold to 50; we're working on a test case for this). Otherwise, nothing has changed for any other targets. The logic, however, has been moved into BasicTTI, so other targets may now also opt-in to this functionality simply by setting LoopMicroOpBufferSize in their processor model definitions. llvm-svn: 208289
2014-05-08 17:14:44 +08:00
// Based on the LSD (loop-stream detector) queue size.
let LoopMicroOpBufferSize = 28;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
// This flag is set to allow the scheduler to assign
// a default model to unrecognized opcodes.
Mark the x86 machine model as incomplete. PR17367. Ideally, the machinel model is added at the time the instructions are defined. But many instructions in X86InstrSSE.td still need a model. Without this workaround the scheduler asserts because x86 already has itinerary classes for these instructions, indicating they should be modeled by the scheduler. Since we use the new machine model for other instructions, it expects a new machine model for these too. llvm-svn: 191391
2013-09-26 02:14:12 +08:00
let CompleteModel = 0;
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
}
let SchedModel = SandyBridgeModel in {
// Sandy Bridge can issue micro-ops to 6 different ports in one cycle.
// Ports 0, 1, and 5 handle all computation.
def SBPort0 : ProcResource<1>;
def SBPort1 : ProcResource<1>;
def SBPort5 : ProcResource<1>;
// Ports 2 and 3 are identical. They handle loads and the address half of
// stores.
def SBPort23 : ProcResource<2>;
// Port 4 gets the data half of stores. Store data can be available later than
// the store address, but since we don't model the latency of stores, we can
// ignore that.
def SBPort4 : ProcResource<1>;
// Many micro-ops are capable of issuing on multiple ports.
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBPort01 : ProcResGroup<[SBPort0, SBPort1]>;
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
def SBPort05 : ProcResGroup<[SBPort0, SBPort5]>;
def SBPort15 : ProcResGroup<[SBPort1, SBPort5]>;
def SBPort015 : ProcResGroup<[SBPort0, SBPort1, SBPort5]>;
Support BufferSize on ProcResGroup for unified MOp schedulers. And add Sandybridge/Haswell resource buffers. llvm-svn: 184034
2013-06-15 12:50:06 +08:00
// 54 Entry Unified Scheduler
def SBPortAny : ProcResGroup<[SBPort0, SBPort1, SBPort23, SBPort4, SBPort5]> {
let BufferSize=54;
}
The divide unit is not pipeline, but it is still buffered. Buffered means a later divide may be executed out-of-order while a prior divide is sitting (buffered) in a reservation station. You can tell it's not pipelined, because operations that use it reserve it for more than one cycle: def : WriteRes<WriteIDiv, [HWPort0, HWDivider]> { let Latency = 25; let ResourceCycles = [1, 10]; } We don't currently distinguish between an unpipeline operation and one that is split into multiple micro-ops requiring the same unit. Except that the later may have NumMicroOps > 1 if they also consume issue/dispatch resources. llvm-svn: 178519
2013-04-02 09:58:47 +08:00
// Integer division issued on port 0.
def SBDivider : ProcResource<1>;
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
// Loads are 4 cycles, so ReadAfterLd registers needn't be available until 4
// cycles after the memory operand.
def : ReadAdvance<ReadAfterLd, 4>;
// Many SchedWrites are defined in pairs with and without a folded load.
// Instructions with folded loads are usually micro-fused, so they only appear
// as two micro-ops when queued in the reservation station.
// This multiclass defines the resource usage for variants with and without
// folded loads.
multiclass SBWriteResPair<X86FoldableSchedWrite SchedRW,
ProcResourceKind ExePort,
int Lat> {
// Register variant is using a single cycle on ExePort.
def : WriteRes<SchedRW, [ExePort]> { let Latency = Lat; }
// Memory variant also uses a cycle on port 2/3 and adds 4 cycles to the
// latency.
def : WriteRes<SchedRW.Folded, [SBPort23, ExePort]> {
let Latency = !add(Lat, 4);
}
}
// A folded store needs a cycle on port 4 for the store data, but it does not
// need an extra port 2/3 cycle to recompute the address.
def : WriteRes<WriteRMW, [SBPort4]>;
def : WriteRes<WriteStore, [SBPort23, SBPort4]>;
def : WriteRes<WriteLoad, [SBPort23]> { let Latency = 4; }
def : WriteRes<WriteMove, [SBPort015]>;
def : WriteRes<WriteZero, []>;
defm : SBWriteResPair<WriteALU, SBPort015, 1>;
defm : SBWriteResPair<WriteIMul, SBPort1, 3>;
Fix IMULX machine model. Multiple def operands require multiple SchedWrites. llvm-svn: 184566
2013-06-22 02:33:04 +08:00
def : WriteRes<WriteIMulH, []> { let Latency = 3; }
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
defm : SBWriteResPair<WriteShift, SBPort05, 1>;
defm : SBWriteResPair<WriteJump, SBPort5, 1>;
// This is for simple LEAs with one or two input operands.
// The complex ones can only execute on port 1, and they require two cycles on
// the port to read all inputs. We don't model that.
def : WriteRes<WriteLEA, [SBPort15]>;
// This is quite rough, latency depends on the dividend.
def : WriteRes<WriteIDiv, [SBPort0, SBDivider]> {
let Latency = 25;
let ResourceCycles = [1, 10];
}
def : WriteRes<WriteIDivLd, [SBPort23, SBPort0, SBDivider]> {
let Latency = 29;
let ResourceCycles = [1, 1, 10];
}
// Scalar and vector floating point.
defm : SBWriteResPair<WriteFAdd, SBPort1, 3>;
defm : SBWriteResPair<WriteFMul, SBPort0, 5>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
defm : SBWriteResPair<WriteFDiv, SBPort0, 24>;
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
defm : SBWriteResPair<WriteFRcp, SBPort0, 5>;
[X86][SchedModel] SSE reciprocal square root instruction latencies. The SSE rsqrt instruction (a fast reciprocal square root estimate) was grouped in the same scheduling IIC_SSE_SQRT* class as the accurate (but very slow) SSE sqrt instruction. For code which uses rsqrt (possibly with newton-raphson iterations) this poor scheduling was affecting performances. This patch splits off the rsqrt instruction from the sqrt instruction scheduling classes and creates new IIC_SSE_RSQER* classes with latency values based on Agner's table. Differential Revision: http://reviews.llvm.org/D5370 Patch by Simon Pilgrim. llvm-svn: 218517
2014-09-26 20:56:44 +08:00
defm : SBWriteResPair<WriteFRsqrt, SBPort0, 5>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
defm : SBWriteResPair<WriteFSqrt, SBPort0, 14>;
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
defm : SBWriteResPair<WriteCvtF2I, SBPort1, 3>;
defm : SBWriteResPair<WriteCvtI2F, SBPort1, 4>;
defm : SBWriteResPair<WriteCvtF2F, SBPort1, 3>;
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
defm : SBWriteResPair<WriteFShuffle, SBPort5, 1>;
defm : SBWriteResPair<WriteFBlend, SBPort05, 1>;
def : WriteRes<WriteFVarBlend, [SBPort0, SBPort5]> {
let Latency = 2;
let ResourceCycles = [1, 1];
}
def : WriteRes<WriteFVarBlendLd, [SBPort0, SBPort5, SBPort23]> {
let Latency = 6;
let ResourceCycles = [1, 1, 1];
}
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
// Vector integer operations.
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
defm : SBWriteResPair<WriteVecShift, SBPort5, 1>;
defm : SBWriteResPair<WriteVecLogic, SBPort5, 1>;
defm : SBWriteResPair<WriteVecALU, SBPort1, 3>;
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
defm : SBWriteResPair<WriteVecIMul, SBPort0, 5>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
defm : SBWriteResPair<WriteShuffle, SBPort5, 1>;
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
defm : SBWriteResPair<WriteBlend, SBPort15, 1>;
def : WriteRes<WriteVarBlend, [SBPort1, SBPort5]> {
let Latency = 2;
let ResourceCycles = [1, 1];
}
def : WriteRes<WriteVarBlendLd, [SBPort1, SBPort5, SBPort23]> {
let Latency = 6;
let ResourceCycles = [1, 1, 1];
}
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def : WriteRes<WriteMPSAD, [SBPort0,SBPort15]> {
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
}
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def : WriteRes<WriteMPSADLd, [SBPort0,SBPort23,SBPort15]> {
let Latency = 11;
let NumMicroOps = 4;
let ResourceCycles = [1,1,2];
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
}
Reverting commit 306414 on behalf of @gadi.haber llvm-svn: 306532
2017-06-28 19:23:31 +08:00
////////////////////////////////////////////////////////////////////////////////
// Horizontal add/sub instructions.
////////////////////////////////////////////////////////////////////////////////
// HADD, HSUB PS/PD
// x,x / v,v,v.
def : WriteRes<WriteFHAdd, [SBPort1]> {
let Latency = 3;
}
// x,m / v,v,m.
def : WriteRes<WriteFHAddLd, [SBPort1, SBPort23]> {
let Latency = 7;
let ResourceCycles = [1, 1];
}
// PHADD|PHSUB (S) W/D.
// v <- v,v.
def : WriteRes<WritePHAdd, [SBPort15]>;
// v <- v,m.
def : WriteRes<WritePHAddLd, [SBPort15, SBPort23]> {
let Latency = 5;
let ResourceCycles = [1, 1];
}
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
// String instructions.
// Packed Compare Implicit Length Strings, Return Mask
def : WriteRes<WritePCmpIStrM, [SBPort015]> {
let Latency = 11;
let ResourceCycles = [3];
}
def : WriteRes<WritePCmpIStrMLd, [SBPort015, SBPort23]> {
let Latency = 11;
let ResourceCycles = [3, 1];
}
// Packed Compare Explicit Length Strings, Return Mask
def : WriteRes<WritePCmpEStrM, [SBPort015]> {
let Latency = 11;
let ResourceCycles = [8];
}
def : WriteRes<WritePCmpEStrMLd, [SBPort015, SBPort23]> {
let Latency = 11;
let ResourceCycles = [7, 1];
}
// Packed Compare Implicit Length Strings, Return Index
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def : WriteRes<WritePCmpIStrI, [SBPort0]> {
let Latency = 11;
let NumMicroOps = 3;
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
let ResourceCycles = [3];
}
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def : WriteRes<WritePCmpIStrILd, [SBPort0,SBPort23]> {
let Latency = 17;
let NumMicroOps = 4;
let ResourceCycles = [3,1];
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
}
// Packed Compare Explicit Length Strings, Return Index
def : WriteRes<WritePCmpEStrI, [SBPort015]> {
let Latency = 4;
let ResourceCycles = [8];
}
def : WriteRes<WritePCmpEStrILd, [SBPort015, SBPort23]> {
let Latency = 4;
let ResourceCycles = [7, 1];
}
// AES Instructions.
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def : WriteRes<WriteAESDecEnc, [SBPort5,SBPort015]> {
let Latency = 7;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
}
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def : WriteRes<WriteAESDecEncLd, [SBPort5,SBPort23,SBPort015]> {
let Latency = 13;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
}
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def : WriteRes<WriteAESIMC, [SBPort5]> {
let Latency = 12;
let NumMicroOps = 2;
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
let ResourceCycles = [2];
}
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def : WriteRes<WriteAESIMCLd, [SBPort5,SBPort23]> {
let Latency = 18;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
}
def : WriteRes<WriteAESKeyGen, [SBPort015]> {
let Latency = 8;
let ResourceCycles = [11];
}
def : WriteRes<WriteAESKeyGenLd, [SBPort015, SBPort23]> {
let Latency = 8;
let ResourceCycles = [10, 1];
}
// Carry-less multiplication instructions.
def : WriteRes<WriteCLMul, [SBPort015]> {
let Latency = 14;
let ResourceCycles = [18];
}
def : WriteRes<WriteCLMulLd, [SBPort015, SBPort23]> {
let Latency = 14;
let ResourceCycles = [17, 1];
}
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
def : WriteRes<WriteSystem, [SBPort015]> { let Latency = 100; }
def : WriteRes<WriteMicrocoded, [SBPort015]> { let Latency = 100; }
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
def : WriteRes<WriteFence, [SBPort23, SBPort4]>;
def : WriteRes<WriteNop, []>;
[X86][FMA] Tag all FMA/FMA4 instructions with WriteFMA schedule class As mentioned on PR17367, many instructions are missing scheduling tags preventing us from setting 'CompleteModel = 1' for better instruction analysis. This patch deals with FMA/FMA4 which is one of the bigger offenders (along with AVX512 in general). Annoyingly all scheduler models need to define WriteFMA (now that its actually used), even for older targets without FMA/FMA4 support, but that is an existing problem shared by other schedule classes. Differential Revision: https://reviews.llvm.org/D40351 llvm-svn: 319016
2017-11-27 18:41:32 +08:00
// AVX2/FMA is not supported on that architecture, but we should define the basic
[X86][SchedModel] Add missing scheduling model for SSE related instructions. The patch defines new or refines existing generic scheduling classes to match the behavior of the SSE instructions. It also maps those scheduling classes on the related SSE instructions. <rdar://problem/15607571> llvm-svn: 202065
2014-02-25 03:33:51 +08:00
// scheduling resources anyway.
defm : SBWriteResPair<WriteFShuffle256, SBPort0, 1>;
defm : SBWriteResPair<WriteShuffle256, SBPort0, 1>;
defm : SBWriteResPair<WriteVarVecShift, SBPort0, 1>;
[X86][FMA] Tag all FMA/FMA4 instructions with WriteFMA schedule class As mentioned on PR17367, many instructions are missing scheduling tags preventing us from setting 'CompleteModel = 1' for better instruction analysis. This patch deals with FMA/FMA4 which is one of the bigger offenders (along with AVX512 in general). Annoyingly all scheduler models need to define WriteFMA (now that its actually used), even for older targets without FMA/FMA4 support, but that is an existing problem shared by other schedule classes. Differential Revision: https://reviews.llvm.org/D40351 llvm-svn: 319016
2017-11-27 18:41:32 +08:00
defm : SBWriteResPair<WriteFMA, SBPort01, 5>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
// Remaining SNB instrs.
def SBWriteResGroup0 : SchedWriteRes<[SBPort0]> {
let Latency = 1;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup0], (instregex "CVTSS2SDrr")>;
def: InstRW<[SBWriteResGroup0], (instregex "PSLLDri")>;
def: InstRW<[SBWriteResGroup0], (instregex "PSLLQri")>;
def: InstRW<[SBWriteResGroup0], (instregex "PSLLWri")>;
def: InstRW<[SBWriteResGroup0], (instregex "PSRADri")>;
def: InstRW<[SBWriteResGroup0], (instregex "PSRAWri")>;
def: InstRW<[SBWriteResGroup0], (instregex "PSRLDri")>;
def: InstRW<[SBWriteResGroup0], (instregex "PSRLQri")>;
def: InstRW<[SBWriteResGroup0], (instregex "PSRLWri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VCVTSS2SDrr")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPMOVMSKBrr")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPSLLDri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPSLLQri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPSLLWri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPSRADri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPSRAWri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPSRLDri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPSRLQri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VPSRLWri")>;
def: InstRW<[SBWriteResGroup0], (instregex "VTESTPDYrr")>;
def: InstRW<[SBWriteResGroup0], (instregex "VTESTPDrr")>;
def: InstRW<[SBWriteResGroup0], (instregex "VTESTPSYrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup0], (instregex "VTESTPSrr")>;
def SBWriteResGroup1 : SchedWriteRes<[SBPort1]> {
let Latency = 1;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup1], (instregex "COMP_FST0r")>;
def: InstRW<[SBWriteResGroup1], (instregex "COM_FST0r")>;
def: InstRW<[SBWriteResGroup1], (instregex "UCOM_FPr")>;
def: InstRW<[SBWriteResGroup1], (instregex "UCOM_Fr")>;
def SBWriteResGroup2 : SchedWriteRes<[SBPort5]> {
let Latency = 1;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup2], (instregex "ANDNPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "ANDNPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "ANDPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "ANDPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "FDECSTP")>;
def: InstRW<[SBWriteResGroup2], (instregex "FFREE")>;
def: InstRW<[SBWriteResGroup2], (instregex "FINCSTP")>;
def: InstRW<[SBWriteResGroup2], (instregex "FNOP")>;
def: InstRW<[SBWriteResGroup2], (instregex "INSERTPSrr")>;
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup2], (instregex "J(A|AE|B|BE|E|G|GE|L|LE|NE|NO|NP|NS|O|P|S)_1")>;
def: InstRW<[SBWriteResGroup2], (instregex "J(A|AE|B|BE|E|G|GE|L|LE|NE|NO|NP|NS|O|P|S)_4")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup2], (instregex "JMP64r")>;
def: InstRW<[SBWriteResGroup2], (instregex "JMP_1")>;
def: InstRW<[SBWriteResGroup2], (instregex "JMP_4")>;
def: InstRW<[SBWriteResGroup2], (instregex "LD_Frr")>;
def: InstRW<[SBWriteResGroup2], (instregex "LOOP")>;
def: InstRW<[SBWriteResGroup2], (instregex "LOOPE")>;
def: InstRW<[SBWriteResGroup2], (instregex "LOOPNE")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOV64toPQIrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVAPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVAPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVDDUPrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVDI2PDIrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVHLPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVLHPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVSDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVSHDUPrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVSLDUPrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVSSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVUPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "MOVUPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "ORPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "ORPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "RETQ")>;
def: InstRW<[SBWriteResGroup2], (instregex "SHUFPDrri")>;
def: InstRW<[SBWriteResGroup2], (instregex "SHUFPSrri")>;
def: InstRW<[SBWriteResGroup2], (instregex "ST_FPrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "ST_Frr")>;
def: InstRW<[SBWriteResGroup2], (instregex "UNPCKHPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "UNPCKHPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "UNPCKLPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "UNPCKLPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VANDNPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VANDNPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VANDNPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VANDNPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VANDPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VANDPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VANDPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VANDPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VEXTRACTF128rr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VINSERTF128rr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VINSERTPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOV64toPQIrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVAPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVAPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVAPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVAPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVDDUPYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVDDUPrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVDI2PDIrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVHLPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVHLPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVSDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVSHDUPYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVSHDUPrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVSLDUPYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVSLDUPrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVSSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVUPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVUPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVUPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VMOVUPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VORPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VORPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VORPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VORPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERM2F128rr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERMILPDYri")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERMILPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERMILPDri")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERMILPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERMILPSYri")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERMILPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERMILPSri")>;
def: InstRW<[SBWriteResGroup2], (instregex "VPERMILPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VSHUFPDYrri")>;
def: InstRW<[SBWriteResGroup2], (instregex "VSHUFPDrri")>;
def: InstRW<[SBWriteResGroup2], (instregex "VSHUFPSYrri")>;
def: InstRW<[SBWriteResGroup2], (instregex "VSHUFPSrri")>;
def: InstRW<[SBWriteResGroup2], (instregex "VUNPCKHPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VUNPCKHPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VUNPCKHPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VUNPCKHPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VUNPCKLPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VUNPCKLPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VUNPCKLPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VUNPCKLPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VXORPDYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VXORPDrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VXORPSYrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "VXORPSrr")>;
def: InstRW<[SBWriteResGroup2], (instregex "XORPDrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup2], (instregex "XORPSrr")>;
def SBWriteResGroup3 : SchedWriteRes<[SBPort01]> {
let Latency = 1;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Add LEA64_32r to scheduler models for Sandybridge,Haswell,Broadwell,Skylake llvm-svn: 320293
2017-12-10 17:14:42 +08:00
def: InstRW<[SBWriteResGroup3], (instregex "LEA(16|32|64)(_32)?r")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup4 : SchedWriteRes<[SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 1;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup4], (instregex "BLENDPDrri")>;
def: InstRW<[SBWriteResGroup4], (instregex "BLENDPSrri")>;
def: InstRW<[SBWriteResGroup4], (instregex "BT(16|32|64)ri8")>;
def: InstRW<[SBWriteResGroup4], (instregex "BT(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup4], (instregex "BTC(16|32|64)ri8")>;
def: InstRW<[SBWriteResGroup4], (instregex "BTC(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup4], (instregex "BTR(16|32|64)ri8")>;
def: InstRW<[SBWriteResGroup4], (instregex "BTR(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup4], (instregex "BTS(16|32|64)ri8")>;
def: InstRW<[SBWriteResGroup4], (instregex "BTS(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup4], (instregex "CDQ")>;
def: InstRW<[SBWriteResGroup4], (instregex "CQO")>;
def: InstRW<[SBWriteResGroup4], (instregex "LAHF")>;
def: InstRW<[SBWriteResGroup4], (instregex "SAHF")>;
def: InstRW<[SBWriteResGroup4], (instregex "SAR(16|32|64)ri")>;
def: InstRW<[SBWriteResGroup4], (instregex "SAR8ri")>;
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup4], (instregex "SET(AE|B|E|G|GE|L|LE|NE|NO|NP|NS|O|P|S)r")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup4], (instregex "SHL(16|32|64)ri")>;
def: InstRW<[SBWriteResGroup4], (instregex "SHL(16|32|64)r1")>;
def: InstRW<[SBWriteResGroup4], (instregex "SHL8r1")>;
def: InstRW<[SBWriteResGroup4], (instregex "SHL8ri")>;
def: InstRW<[SBWriteResGroup4], (instregex "SHR(16|32|64)ri")>;
def: InstRW<[SBWriteResGroup4], (instregex "SHR8ri")>;
def: InstRW<[SBWriteResGroup4], (instregex "VBLENDPDYrri")>;
def: InstRW<[SBWriteResGroup4], (instregex "VBLENDPDrri")>;
def: InstRW<[SBWriteResGroup4], (instregex "VBLENDPSYrri")>;
def: InstRW<[SBWriteResGroup4], (instregex "VBLENDPSrri")>;
def: InstRW<[SBWriteResGroup4], (instregex "VMOVDQAYrr")>;
def: InstRW<[SBWriteResGroup4], (instregex "VMOVDQArr")>;
def: InstRW<[SBWriteResGroup4], (instregex "VMOVDQUYrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup4], (instregex "VMOVDQUrr")>;
def SBWriteResGroup5 : SchedWriteRes<[SBPort15]> {
let Latency = 1;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PABSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PABSDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PABSWrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PADDQirr")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PALIGNRrri")>;
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PSHUFBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PSIGNBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PSIGNDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "MMX_PSIGNWrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "PABSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PABSDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PABSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PACKSSDWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PACKSSWBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PACKUSDWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PACKUSWBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PADDBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PADDDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PADDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PADDSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PADDSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PADDUSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PADDUSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PADDWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PALIGNRrri")>;
def: InstRW<[SBWriteResGroup5], (instregex "PAVGBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PAVGWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PBLENDWrri")>;
def: InstRW<[SBWriteResGroup5], (instregex "PCMPEQBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PCMPEQDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PCMPEQQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PCMPEQWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PCMPGTBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PCMPGTDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PCMPGTWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMAXSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMAXSDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMAXSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMAXUBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMAXUDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMAXUWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMINSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMINSDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMINSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMINUBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMINUDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMINUWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVSXBDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVSXBQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVSXBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVSXDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVSXWDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVSXWQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVZXBDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVZXBQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVZXBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVZXDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVZXWDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PMOVZXWQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSHUFBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSHUFDri")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSHUFHWri")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSHUFLWri")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "PSIGNBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSIGNDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSIGNWrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "PSLLDQri")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSRLDQri")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSUBBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSUBDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSUBQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSUBSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSUBSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSUBUSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSUBUSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PSUBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PUNPCKHBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PUNPCKHDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PUNPCKHQDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PUNPCKHWDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PUNPCKLBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PUNPCKLDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PUNPCKLQDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "PUNPCKLWDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPABSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPABSDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPABSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPACKSSDWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPACKSSWBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPACKUSDWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPACKUSWBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPADDBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPADDDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPADDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPADDSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPADDSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPADDUSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPADDUSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPADDWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPALIGNRrri")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPAVGBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPAVGWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPBLENDWrri")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPCMPEQBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPCMPEQDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPCMPEQQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPCMPEQWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPCMPGTBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPCMPGTDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPCMPGTWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMAXSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMAXSDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMAXSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMAXUBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMAXUDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMAXUWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMINSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMINSDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMINSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMINUBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMINUDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMINUWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVSXBDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVSXBQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVSXBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVSXDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVSXWDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVSXWQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVZXBDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVZXBQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVZXBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVZXDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVZXWDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPMOVZXWQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSHUFBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSHUFDri")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSHUFHWri")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSHUFLWri")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "VPSIGNBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSIGNDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSIGNWrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "VPSLLDQri")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSRLDQri")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSUBBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSUBDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSUBQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSUBSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSUBSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSUBUSBrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSUBUSWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPSUBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPUNPCKHBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPUNPCKHDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPUNPCKHQDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPUNPCKHWDrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPUNPCKLBWrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPUNPCKLDQrr")>;
def: InstRW<[SBWriteResGroup5], (instregex "VPUNPCKLQDQrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup5], (instregex "VPUNPCKLWDrr")>;
def SBWriteResGroup6 : SchedWriteRes<[SBPort015]> {
let Latency = 1;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "ADD(16|32|64)ri")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "ADD(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "ADD8i8")>;
def: InstRW<[SBWriteResGroup6], (instregex "ADD8ri")>;
def: InstRW<[SBWriteResGroup6], (instregex "ADD8rr")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "AND(16|32|64)ri")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "AND(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "AND8i8")>;
def: InstRW<[SBWriteResGroup6], (instregex "AND8ri")>;
def: InstRW<[SBWriteResGroup6], (instregex "AND8rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "CBW")>;
def: InstRW<[SBWriteResGroup6], (instregex "CMC")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "CMP(16|32|64)ri")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "CMP(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "CMP8i8")>;
def: InstRW<[SBWriteResGroup6], (instregex "CMP8ri")>;
def: InstRW<[SBWriteResGroup6], (instregex "CMP8rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "CWDE")>;
def: InstRW<[SBWriteResGroup6], (instregex "DEC(16|32|64)r")>;
def: InstRW<[SBWriteResGroup6], (instregex "DEC8r")>;
def: InstRW<[SBWriteResGroup6], (instregex "INC(16|32|64)r")>;
def: InstRW<[SBWriteResGroup6], (instregex "INC8r")>;
def: InstRW<[SBWriteResGroup6], (instregex "MMX_MOVD64from64rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "MMX_MOVQ2DQrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOV(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOV8ri")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOV8rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOVDQArr")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOVDQUrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOVPQI2QIrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOVSX(16|32|64)rr16")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOVSX(16|32|64)rr32")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOVSX(16|32|64)rr8")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOVZX(16|32|64)rr16")>;
def: InstRW<[SBWriteResGroup6], (instregex "MOVZX(16|32|64)rr8")>;
def: InstRW<[SBWriteResGroup6], (instregex "NEG(16|32|64)r")>;
def: InstRW<[SBWriteResGroup6], (instregex "NEG8r")>;
def: InstRW<[SBWriteResGroup6], (instregex "NOT(16|32|64)r")>;
def: InstRW<[SBWriteResGroup6], (instregex "NOT8r")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "OR(16|32|64)ri")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "OR(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "OR8i8")>;
def: InstRW<[SBWriteResGroup6], (instregex "OR8ri")>;
def: InstRW<[SBWriteResGroup6], (instregex "OR8rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "PANDNrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "PANDrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "PORrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "PXORrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "STC")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "SUB(16|32|64)ri")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "SUB(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "SUB8i8")>;
def: InstRW<[SBWriteResGroup6], (instregex "SUB8ri")>;
def: InstRW<[SBWriteResGroup6], (instregex "SUB8rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "TEST(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "TEST8i8")>;
def: InstRW<[SBWriteResGroup6], (instregex "TEST8ri")>;
def: InstRW<[SBWriteResGroup6], (instregex "TEST8rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "VMOVPQI2QIrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "VMOVZPQILo2PQIrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "VPANDNrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "VPANDrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "VPORrr")>;
def: InstRW<[SBWriteResGroup6], (instregex "VPXORrr")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "XOR(16|32|64)ri")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "XOR(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup6], (instregex "XOR8i8")>;
def: InstRW<[SBWriteResGroup6], (instregex "XOR8ri")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup6], (instregex "XOR8rr")>;
def SBWriteResGroup7 : SchedWriteRes<[SBPort0]> {
let Latency = 2;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup7], (instregex "MOVMSKPDrr")>;
def: InstRW<[SBWriteResGroup7], (instregex "MOVMSKPSrr")>;
def: InstRW<[SBWriteResGroup7], (instregex "MOVPDI2DIrr")>;
def: InstRW<[SBWriteResGroup7], (instregex "MOVPQIto64rr")>;
def: InstRW<[SBWriteResGroup7], (instregex "PMOVMSKBrr")>;
def: InstRW<[SBWriteResGroup7], (instregex "VMOVMSKPDYrr")>;
def: InstRW<[SBWriteResGroup7], (instregex "VMOVMSKPDrr")>;
def: InstRW<[SBWriteResGroup7], (instregex "VMOVMSKPSYrr")>;
def: InstRW<[SBWriteResGroup7], (instregex "VMOVMSKPSrr")>;
def: InstRW<[SBWriteResGroup7], (instregex "VMOVPDI2DIrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup7], (instregex "VMOVPQIto64rr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup9 : SchedWriteRes<[SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [2];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup9], (instregex "BLENDVPDrr0")>;
def: InstRW<[SBWriteResGroup9], (instregex "BLENDVPSrr0")>;
def: InstRW<[SBWriteResGroup9], (instregex "ROL(16|32|64)ri")>;
def: InstRW<[SBWriteResGroup9], (instregex "ROL8ri")>;
def: InstRW<[SBWriteResGroup9], (instregex "ROR(16|32|64)ri")>;
def: InstRW<[SBWriteResGroup9], (instregex "ROR8ri")>;
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup9], (instregex "SET(A|BE)r")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup9], (instregex "VBLENDVPDYrr")>;
def: InstRW<[SBWriteResGroup9], (instregex "VBLENDVPDrr")>;
def: InstRW<[SBWriteResGroup9], (instregex "VBLENDVPSYrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup9], (instregex "VBLENDVPSrr")>;
def SBWriteResGroup10 : SchedWriteRes<[SBPort15]> {
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [2];
}
def: InstRW<[SBWriteResGroup10], (instregex "VPBLENDVBrr")>;
def SBWriteResGroup11 : SchedWriteRes<[SBPort015]> {
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [2];
}
def: InstRW<[SBWriteResGroup11], (instregex "SCASB")>;
def: InstRW<[SBWriteResGroup11], (instregex "SCASL")>;
def: InstRW<[SBWriteResGroup11], (instregex "SCASQ")>;
def: InstRW<[SBWriteResGroup11], (instregex "SCASW")>;
def SBWriteResGroup12 : SchedWriteRes<[SBPort0,SBPort1]> {
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup12], (instregex "COMISDrr")>;
def: InstRW<[SBWriteResGroup12], (instregex "COMISSrr")>;
def: InstRW<[SBWriteResGroup12], (instregex "UCOMISDrr")>;
def: InstRW<[SBWriteResGroup12], (instregex "UCOMISSrr")>;
def: InstRW<[SBWriteResGroup12], (instregex "VCOMISDrr")>;
def: InstRW<[SBWriteResGroup12], (instregex "VCOMISSrr")>;
def: InstRW<[SBWriteResGroup12], (instregex "VUCOMISDrr")>;
def: InstRW<[SBWriteResGroup12], (instregex "VUCOMISSrr")>;
def SBWriteResGroup13 : SchedWriteRes<[SBPort0,SBPort5]> {
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup13], (instregex "CVTPS2PDrr")>;
def: InstRW<[SBWriteResGroup13], (instregex "PTESTrr")>;
def: InstRW<[SBWriteResGroup13], (instregex "VCVTPS2PDYrr")>;
def: InstRW<[SBWriteResGroup13], (instregex "VCVTPS2PDrr")>;
def: InstRW<[SBWriteResGroup13], (instregex "VPTESTYrr")>;
def: InstRW<[SBWriteResGroup13], (instregex "VPTESTrr")>;
def SBWriteResGroup14 : SchedWriteRes<[SBPort0,SBPort15]> {
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup14], (instregex "PSLLDrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "PSLLQrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "PSLLWrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "PSRADrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "PSRAWrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "PSRLDrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "PSRLQrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "PSRLWrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "VPSLLDrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "VPSLLQrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "VPSLLWrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "VPSRADrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "VPSRAWrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "VPSRLDrr")>;
def: InstRW<[SBWriteResGroup14], (instregex "VPSRLQrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup14], (instregex "VPSRLWrr")>;
def SBWriteResGroup15 : SchedWriteRes<[SBPort0,SBPort015]> {
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Add CBW/CDQ/CDQE/CQO/CWD/CWDE to WriteALU schedule class Some CPUs are already overriding these sign extension instructions but we should be able to use the WriteALU schedule class by default. Differential Revision: https://reviews.llvm.org/D39899 llvm-svn: 318308
2017-11-16 01:11:24 +08:00
def: InstRW<[SBWriteResGroup15], (instregex "CWD")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup15], (instregex "FNSTSW16r")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup16 : SchedWriteRes<[SBPort1,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup16], (instregex "BSWAP(16|32|64)r")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup17 : SchedWriteRes<[SBPort5,SBPort15]> {
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup17], (instregex "PINSRBrr")>;
def: InstRW<[SBWriteResGroup17], (instregex "PINSRDrr")>;
def: InstRW<[SBWriteResGroup17], (instregex "PINSRQrr")>;
[X86] Adjust names of PINSRW/PEXTRW intructions between MMX/SSE/AVX/AVX512 for consistency and to maybe enable more regular expression compaction in the scheduler models. NFCI llvm-svn: 323352
2018-01-25 01:58:51 +08:00
def: InstRW<[SBWriteResGroup17], (instregex "PINSRWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup17], (instregex "VPINSRBrr")>;
def: InstRW<[SBWriteResGroup17], (instregex "VPINSRDrr")>;
def: InstRW<[SBWriteResGroup17], (instregex "VPINSRQrr")>;
[X86] Adjust names of PINSRW/PEXTRW intructions between MMX/SSE/AVX/AVX512 for consistency and to maybe enable more regular expression compaction in the scheduler models. NFCI llvm-svn: 323352
2018-01-25 01:58:51 +08:00
def: InstRW<[SBWriteResGroup17], (instregex "VPINSRWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup18 : SchedWriteRes<[SBPort5,SBPort015]> {
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup18], (instregex "JRCXZ")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup18], (instregex "MMX_MOVDQ2Qrr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup19 : SchedWriteRes<[SBPort05,SBPort015]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 2;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup19], (instregex "ADC(16|32|64)ri")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup19], (instregex "ADC(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup19], (instregex "ADC8ri")>;
def: InstRW<[SBWriteResGroup19], (instregex "ADC8rr")>;
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup19], (instregex "CMOV(AE|B|E|G|GE|L|LE|NE|NO|NP|NS|O|P|S)(16|32|64)rr")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup19], (instregex "SBB(16|32|64)ri")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup19], (instregex "SBB(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup19], (instregex "SBB8ri")>;
def: InstRW<[SBWriteResGroup19], (instregex "SBB8rr")>;
def: InstRW<[SBWriteResGroup19], (instregex "SHLD(16|32|64)rri8")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup19], (instregex "SHRD(16|32|64)rri8")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup20 : SchedWriteRes<[SBPort0]> {
let Latency = 3;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup20], (instregex "MMX_PMADDUBSWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "MMX_PMULHRSWrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup20], (instregex "MMX_PMULUDQirr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMADDUBSWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMADDWDrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMULDQrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMULHRSWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMULHUWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMULHWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMULLDrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMULLWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PMULUDQrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "PSADBWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMADDUBSWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMADDWDrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMULDQrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMULHRSWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMULHUWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMULHWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMULLDrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMULLWrr")>;
def: InstRW<[SBWriteResGroup20], (instregex "VPMULUDQrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup20], (instregex "VPSADBWrr")>;
def SBWriteResGroup21 : SchedWriteRes<[SBPort1]> {
let Latency = 3;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "ADDPDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ADDPSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ADDSDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ADDSSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ADDSUBPDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ADDSUBPSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ADD_FPrST0")>;
def: InstRW<[SBWriteResGroup21], (instregex "ADD_FST0r")>;
def: InstRW<[SBWriteResGroup21], (instregex "ADD_FrST0")>;
def: InstRW<[SBWriteResGroup21], (instregex "BSF(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup21], (instregex "BSR(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup21], (instregex "CMPPDrri")>;
def: InstRW<[SBWriteResGroup21], (instregex "CMPPSrri")>;
[X86] Add CMPSDrr/rm to the scheduler models. Somehow CMPSSrr/rm was there and the VEX version was there, but this was consistently missing. llvm-svn: 320289
2017-12-10 17:14:37 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "CMPSDrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "CMPSSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "CRC32r(16|32|64)r8")>;
def: InstRW<[SBWriteResGroup21], (instregex "CRC32r(16|32|64)r64")>;
def: InstRW<[SBWriteResGroup21], (instregex "CVTDQ2PSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "CVTPS2DQrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "CVTTPS2DQrr")>;
[X86] Add IMUL scheduling info on sandybridge, fix it on >=haswell. Summary: Only IMUL16rri uses an extra P0156. IMUL32* and IMUL16rr only use P1. This was computed using https://github.com/google/EXEgesis/blob/master/exegesis/tools/compute_itineraries.cc This can easily be validated by running perf on the following code: ``` int main(int argc, char**argv) { int a = argc; int b = argc; int c = argc; int d = argc; for (int i = 0; i < LOOP_ITERATIONS; ++i) { asm volatile( R"( .rept 10000 imull $0x2, %%edx, %%eax imull $0x2, %%ecx, %%ebx imull $0x2, %%eax, %%edx imull $0x2, %%ebx, %%ecx .endr )" : "+a"(a), "+b"(b), "+c"(c), "+d"(d) : :); } return a+b+c+d; } ``` -> test.cc perf stat -x, -e cycles --pfm-events=uops_executed_port:port_0:u,uops_executed_port:port_1:u,uops_executed_port:port_2:u,uops_executed_port:port_3:u,uops_executed_port:port_4:u,uops_executed_port:port_5:u,uops_executed_port:port_6:u,uops_executed_port:port_7:u test Reviewers: craig.topper, RKSimon, gadi.haber Subscribers: llvm-commits, gchatelet, chandlerc Differential Revision: https://reviews.llvm.org/D43460 llvm-svn: 326877
2018-03-07 16:14:02 +08:00
def: InstRW<[SBWriteResGroup21], (instrs IMUL16rr, IMUL32rr, IMUL32rri, IMUL32rri8, IMUL64rr, IMUL64rri32, IMUL64rri8)>;
[X86] Add the commutable floating point min/max pseudo instructions to sandybridge,haswell,broadwell,skylakeclient scheduler models. llvm-svn: 320277
2017-12-10 09:24:05 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "MAX(C?)PDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MAX(C?)PSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MAX(C?)SDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MAX(C?)SSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MIN(C?)PDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MIN(C?)PSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MIN(C?)SDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MIN(C?)SSrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "MMX_CVTPI2PSirr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MMX_CVTPS2PIirr")>;
def: InstRW<[SBWriteResGroup21], (instregex "MMX_CVTTPS2PIirr")>;
[X86] Expand IMUL/MUL instregexs in Intel scheduler models. Add load latency to some of them in SkylakeClient model. The regular expressions and the imul names caused some instructions to be matched by multiple regexs creating unpredictable results. This changes them all to use explicit instrs instead. While doing this I also found that some instructions in Skylake were missing load latency so I fixed that too. llvm-svn: 323406
2018-01-25 14:57:42 +08:00
def: InstRW<[SBWriteResGroup21], (instrs MUL8r)>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "POPCNT(16|32|64)rr")>;
def: InstRW<[SBWriteResGroup21], (instregex "PUSHFS64")>;
def: InstRW<[SBWriteResGroup21], (instregex "ROUNDPDr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ROUNDPSr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ROUNDSDr")>;
def: InstRW<[SBWriteResGroup21], (instregex "ROUNDSSr")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUBPDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUBPSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUBR_FPrST0")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUBR_FST0r")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUBR_FrST0")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUBSDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUBSSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUB_FPrST0")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUB_FST0r")>;
def: InstRW<[SBWriteResGroup21], (instregex "SUB_FrST0")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDPDYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDPDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDPSYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDPSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDSDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDSSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDSUBPDYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDSUBPDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDSUBPSYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VADDSUBPSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCMPPDYrri")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCMPPDrri")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCMPPSYrri")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCMPPSrri")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCMPSDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCMPSSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCVTDQ2PSYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCVTDQ2PSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCVTPS2DQYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCVTPS2DQrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCVTTPS2DQYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VCVTTPS2DQrr")>;
[X86] Add the commutable floating point min/max pseudo instructions to sandybridge,haswell,broadwell,skylakeclient scheduler models. llvm-svn: 320277
2017-12-10 09:24:05 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "VMAX(C?)PDYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMAX(C?)PDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMAX(C?)PSYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMAX(C?)PSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMAX(C?)SDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMAX(C?)SSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMIN(C?)PDYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMIN(C?)PDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMIN(C?)PSYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMIN(C?)PSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMIN(C?)SDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VMIN(C?)SSrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "VROUNDPDr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VROUNDPSr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VROUNDSDr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VROUNDSSr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VROUNDYPDr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VROUNDYPSr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VSUBPDYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VSUBPDrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VSUBPSYrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VSUBPSrr")>;
def: InstRW<[SBWriteResGroup21], (instregex "VSUBSDrr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup21], (instregex "VSUBSSrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
[X86] Add IMUL scheduling info on sandybridge, fix it on >=haswell. Summary: Only IMUL16rri uses an extra P0156. IMUL32* and IMUL16rr only use P1. This was computed using https://github.com/google/EXEgesis/blob/master/exegesis/tools/compute_itineraries.cc This can easily be validated by running perf on the following code: ``` int main(int argc, char**argv) { int a = argc; int b = argc; int c = argc; int d = argc; for (int i = 0; i < LOOP_ITERATIONS; ++i) { asm volatile( R"( .rept 10000 imull $0x2, %%edx, %%eax imull $0x2, %%ecx, %%ebx imull $0x2, %%eax, %%edx imull $0x2, %%ebx, %%ecx .endr )" : "+a"(a), "+b"(b), "+c"(c), "+d"(d) : :); } return a+b+c+d; } ``` -> test.cc perf stat -x, -e cycles --pfm-events=uops_executed_port:port_0:u,uops_executed_port:port_1:u,uops_executed_port:port_2:u,uops_executed_port:port_3:u,uops_executed_port:port_4:u,uops_executed_port:port_5:u,uops_executed_port:port_6:u,uops_executed_port:port_7:u test Reviewers: craig.topper, RKSimon, gadi.haber Subscribers: llvm-commits, gchatelet, chandlerc Differential Revision: https://reviews.llvm.org/D43460 llvm-svn: 326877
2018-03-07 16:14:02 +08:00
def SBWriteResGroup21_16i : SchedWriteRes<[SBPort1, SBPort015]> {
let Latency = 3;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup21_16i], (instrs IMUL16rri, IMUL16rri8)>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup22 : SchedWriteRes<[SBPort0,SBPort5]> {
let Latency = 3;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup22], (instregex "EXTRACTPSrr")>;
def: InstRW<[SBWriteResGroup22], (instregex "VEXTRACTPSrr")>;
def SBWriteResGroup23 : SchedWriteRes<[SBPort0,SBPort15]> {
let Latency = 3;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup23], (instregex "PEXTRBrr")>;
def: InstRW<[SBWriteResGroup23], (instregex "PEXTRDrr")>;
def: InstRW<[SBWriteResGroup23], (instregex "PEXTRQrr")>;
[X86] Adjust names of PINSRW/PEXTRW intructions between MMX/SSE/AVX/AVX512 for consistency and to maybe enable more regular expression compaction in the scheduler models. NFCI llvm-svn: 323352
2018-01-25 01:58:51 +08:00
def: InstRW<[SBWriteResGroup23], (instregex "PEXTRWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup23], (instregex "VPEXTRBrr")>;
def: InstRW<[SBWriteResGroup23], (instregex "VPEXTRDrr")>;
def: InstRW<[SBWriteResGroup23], (instregex "VPEXTRQrr")>;
[X86] Adjust names of PINSRW/PEXTRW intructions between MMX/SSE/AVX/AVX512 for consistency and to maybe enable more regular expression compaction in the scheduler models. NFCI llvm-svn: 323352
2018-01-25 01:58:51 +08:00
def: InstRW<[SBWriteResGroup23], (instregex "VPEXTRWrr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def SBWriteResGroup23_2 : SchedWriteRes<[SBPort05]> {
let Latency = 3;
let NumMicroOps = 3;
let ResourceCycles = [3];
}
def: InstRW<[SBWriteResGroup23_2], (instregex "ROL(16|32|64)rCL")>;
def: InstRW<[SBWriteResGroup23_2], (instregex "ROL8rCL")>;
def: InstRW<[SBWriteResGroup23_2], (instregex "ROR(16|32|64)rCL")>;
def: InstRW<[SBWriteResGroup23_2], (instregex "ROR8rCL")>;
def: InstRW<[SBWriteResGroup23_2], (instregex "SAR(16|32|64)rCL")>;
def: InstRW<[SBWriteResGroup23_2], (instregex "SAR8rCL")>;
def: InstRW<[SBWriteResGroup23_2], (instregex "SHL(16|32|64)rCL")>;
def: InstRW<[SBWriteResGroup23_2], (instregex "SHL8rCL")>;
def: InstRW<[SBWriteResGroup23_2], (instregex "SHR(16|32|64)rCL")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup23_2], (instregex "SHR8rCL")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup24 : SchedWriteRes<[SBPort15]> {
let Latency = 3;
let NumMicroOps = 3;
let ResourceCycles = [3];
}
[X86] Name the MMX phaddd instruction with 3 Ds instead of just 2. NFC llvm-svn: 323403
2018-01-25 12:45:32 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "MMX_PHADDDrr")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "MMX_PHADDSWrr")>;
def: InstRW<[SBWriteResGroup24], (instregex "MMX_PHADDWrr")>;
def: InstRW<[SBWriteResGroup24], (instregex "MMX_PHSUBDrr")>;
def: InstRW<[SBWriteResGroup24], (instregex "MMX_PHSUBSWrr")>;
def: InstRW<[SBWriteResGroup24], (instregex "MMX_PHSUBWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "PHADDDrr")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "PHADDSWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "PHADDWrr")>;
def: InstRW<[SBWriteResGroup24], (instregex "PHSUBDrr")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "PHSUBSWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "PHSUBWrr")>;
def: InstRW<[SBWriteResGroup24], (instregex "VPHADDDrr")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "VPHADDSWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "VPHADDWrr")>;
def: InstRW<[SBWriteResGroup24], (instregex "VPHSUBDrr")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "VPHSUBSWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup24], (instregex "VPHSUBWrr")>;
def SBWriteResGroup25 : SchedWriteRes<[SBPort015]> {
let Latency = 3;
let NumMicroOps = 3;
let ResourceCycles = [3];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup25], (instregex "ADC8i8")>;
def: InstRW<[SBWriteResGroup25], (instregex "LEAVE64")>;
def: InstRW<[SBWriteResGroup25], (instregex "OUT32rr")>;
def: InstRW<[SBWriteResGroup25], (instregex "OUT8rr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup25], (instregex "SBB8i8")>;
def: InstRW<[SBWriteResGroup25], (instregex "XADD(16|32|64)rr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup25], (instregex "XADD8rr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def SBWriteResGroup25_2 : SchedWriteRes<[SBPort5,SBPort05]> {
let Latency = 3;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SBWriteResGroup25_2], (instregex "CMOVBE_F")>;
def: InstRW<[SBWriteResGroup25_2], (instregex "CMOVB_F")>;
def: InstRW<[SBWriteResGroup25_2], (instregex "CMOVE_F")>;
def: InstRW<[SBWriteResGroup25_2], (instregex "CMOVNBE_F")>;
def: InstRW<[SBWriteResGroup25_2], (instregex "CMOVNB_F")>;
def: InstRW<[SBWriteResGroup25_2], (instregex "CMOVNE_F")>;
def: InstRW<[SBWriteResGroup25_2], (instregex "CMOVNP_F")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup25_2], (instregex "CMOVP_F")>;
def SBWriteResGroup26 : SchedWriteRes<[SBPort05,SBPort015]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 3;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup26], (instregex "CMOV(A|BE)(16|32|64)rr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def SBWriteResGroup26_2 : SchedWriteRes<[SBPort0,SBPort1,SBPort5]> {
let Latency = 3;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup26_2], (instregex "COM_FIPr")>;
def: InstRW<[SBWriteResGroup26_2], (instregex "COM_FIr")>;
def: InstRW<[SBWriteResGroup26_2], (instregex "UCOM_FIPr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup26_2], (instregex "UCOM_FIr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup27 : SchedWriteRes<[SBPort0,SBPort1]> {
let Latency = 4;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Expand IMUL/MUL instregexs in Intel scheduler models. Add load latency to some of them in SkylakeClient model. The regular expressions and the imul names caused some instructions to be matched by multiple regexs creating unpredictable results. This changes them all to use explicit instrs instead. While doing this I also found that some instructions in Skylake were missing load latency so I fixed that too. llvm-svn: 323406
2018-01-25 14:57:42 +08:00
def: InstRW<[SBWriteResGroup27], (instrs MUL16r, MUL32r, MUL64r)>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup28 : SchedWriteRes<[SBPort1,SBPort5]> {
let Latency = 4;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup28], (instregex "CVTDQ2PDrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "CVTPD2DQrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "CVTPD2PSrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "CVTSD2SSrr")>;
[X86] Rename some instructions that start with Int_ to have the _Int at the end. This matches AVX512 version and is more consistent overall. And improves our scheduler models. In some cases this adds _Int to instructions that didn't have any Int_ before. It's a side effect of the adjustments made to some of the multiclasses. llvm-svn: 320325
2017-12-11 03:47:56 +08:00
def: InstRW<[SBWriteResGroup28], (instregex "CVTSI642SDrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup28], (instregex "CVTSI2SDrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "CVTTPD2DQrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "MMX_CVTPD2PIirr")>;
def: InstRW<[SBWriteResGroup28], (instregex "MMX_CVTPI2PDirr")>;
def: InstRW<[SBWriteResGroup28], (instregex "MMX_CVTTPD2PIirr")>;
def: InstRW<[SBWriteResGroup28], (instregex "VCVTDQ2PDYrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "VCVTDQ2PDrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "VCVTPD2DQYrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "VCVTPD2DQrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "VCVTPD2PSYrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "VCVTPD2PSrr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup28], (instregex "VCVTSD2SSrr")>;
[X86] Rename some instructions that start with Int_ to have the _Int at the end. This matches AVX512 version and is more consistent overall. And improves our scheduler models. In some cases this adds _Int to instructions that didn't have any Int_ before. It's a side effect of the adjustments made to some of the multiclasses. llvm-svn: 320325
2017-12-11 03:47:56 +08:00
def: InstRW<[SBWriteResGroup28], (instregex "VCVTSI642SDrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup28], (instregex "VCVTSI2SDrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "VCVTTPD2DQYrr")>;
def: InstRW<[SBWriteResGroup28], (instregex "VCVTTPD2DQrr")>;
def SBWriteResGroup29 : SchedWriteRes<[SBPort1,SBPort015]> {
let Latency = 4;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup29], (instregex "MOV64sr")>;
def SBWriteResGroup29_2 : SchedWriteRes<[SBPort5,SBPort015]> {
let Latency = 4;
let NumMicroOps = 4;
let ResourceCycles = [1,3];
}
def: InstRW<[SBWriteResGroup29_2], (instregex "OUT32ir")>;
def: InstRW<[SBWriteResGroup29_2], (instregex "OUT8ir")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup29_2], (instregex "PAUSE")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def SBWriteResGroup29_3 : SchedWriteRes<[SBPort05,SBPort015]> {
let Latency = 4;
let NumMicroOps = 4;
let ResourceCycles = [3,1];
}
def: InstRW<[SBWriteResGroup29_3], (instregex "SHLD(16|32|64)rrCL")>;
def: InstRW<[SBWriteResGroup29_3], (instregex "SHRD(16|32|64)rrCL")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup30 : SchedWriteRes<[SBPort0]> {
let Latency = 5;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup30], (instregex "MULPDrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "MULPSrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "MULSDrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "MULSSrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "MUL_FPrST0")>;
def: InstRW<[SBWriteResGroup30], (instregex "MUL_FST0r")>;
def: InstRW<[SBWriteResGroup30], (instregex "MUL_FrST0")>;
def: InstRW<[SBWriteResGroup30], (instregex "PCMPGTQrr")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup30], (instregex "PHMINPOSUWrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup30], (instregex "RCPPSr")>;
def: InstRW<[SBWriteResGroup30], (instregex "RCPSSr")>;
def: InstRW<[SBWriteResGroup30], (instregex "RSQRTPSr")>;
def: InstRW<[SBWriteResGroup30], (instregex "RSQRTSSr")>;
def: InstRW<[SBWriteResGroup30], (instregex "VMULPDYrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "VMULPDrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "VMULPSYrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "VMULPSrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "VMULSDrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "VMULSSrr")>;
def: InstRW<[SBWriteResGroup30], (instregex "VPCMPGTQrr")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup30], (instregex "VPHMINPOSUWrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup30], (instregex "VRCPPSr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup30], (instregex "VRCPSSr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup30], (instregex "VRSQRTPSr")>;
def: InstRW<[SBWriteResGroup30], (instregex "VRSQRTSSr")>;
def SBWriteResGroup31 : SchedWriteRes<[SBPort23]> {
let Latency = 5;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup31], (instregex "MOV(16|32|64)rm")>;
def: InstRW<[SBWriteResGroup31], (instregex "MOV8rm")>;
def: InstRW<[SBWriteResGroup31], (instregex "MOVSX(16|32|64)rm16")>;
def: InstRW<[SBWriteResGroup31], (instregex "MOVSX(16|32|64)rm32")>;
def: InstRW<[SBWriteResGroup31], (instregex "MOVSX(16|32|64)rm8")>;
def: InstRW<[SBWriteResGroup31], (instregex "MOVZX(16|32|64)rm16")>;
def: InstRW<[SBWriteResGroup31], (instregex "MOVZX(16|32|64)rm8")>;
def: InstRW<[SBWriteResGroup31], (instregex "PREFETCH")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup32 : SchedWriteRes<[SBPort0,SBPort1]> {
let Latency = 5;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup32], (instregex "CVTSD2SI64rr")>;
def: InstRW<[SBWriteResGroup32], (instregex "CVTSD2SIrr")>;
def: InstRW<[SBWriteResGroup32], (instregex "CVTSS2SI64rr")>;
def: InstRW<[SBWriteResGroup32], (instregex "CVTSS2SIrr")>;
def: InstRW<[SBWriteResGroup32], (instregex "CVTTSD2SI64rr")>;
def: InstRW<[SBWriteResGroup32], (instregex "CVTTSD2SIrr")>;
def: InstRW<[SBWriteResGroup32], (instregex "CVTTSS2SI64rr")>;
def: InstRW<[SBWriteResGroup32], (instregex "CVTTSS2SIrr")>;
def: InstRW<[SBWriteResGroup32], (instregex "VCVTSD2SI64rr")>;
def: InstRW<[SBWriteResGroup32], (instregex "VCVTSD2SIrr")>;
def: InstRW<[SBWriteResGroup32], (instregex "VCVTSS2SI64rr")>;
def: InstRW<[SBWriteResGroup32], (instregex "VCVTSS2SIrr")>;
def: InstRW<[SBWriteResGroup32], (instregex "VCVTTSD2SI64rr")>;
def: InstRW<[SBWriteResGroup32], (instregex "VCVTTSD2SIrr")>;
def: InstRW<[SBWriteResGroup32], (instregex "VCVTTSS2SI64rr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup32], (instregex "VCVTTSS2SIrr")>;
def SBWriteResGroup33 : SchedWriteRes<[SBPort4,SBPort23]> {
let Latency = 5;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup33], (instregex "MOV(16|32|64)mr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup33], (instregex "MOV8mr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVAPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVAPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVDQAmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVDQUmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVHPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVHPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVLPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVLPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVNTDQmr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup33], (instregex "MOVNTI_64mr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVNTImr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVNTPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVNTPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVPDI2DImr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVPQI2QImr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVPQIto64mr")>;
[X86] Add MOVQI2PQIrm, MOVSDmr, and MOVSDrm to scheduler information The VEX versions were present but not the legacy SSE versions. llvm-svn: 320294
2017-12-10 17:14:44 +08:00
def: InstRW<[SBWriteResGroup33], (instregex "MOVSDmr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup33], (instregex "MOVSSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVUPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "MOVUPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "PUSH64i8")>;
def: InstRW<[SBWriteResGroup33], (instregex "PUSH(16|32|64)r")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup33], (instregex "VEXTRACTF128mr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVAPDYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVAPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVAPSYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVAPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVDQAYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVDQAmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVDQUYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVDQUmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVHPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVHPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVLPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVLPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVNTDQYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVNTDQmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVNTPDYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVNTPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVNTPSYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVNTPSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVPDI2DImr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVPQI2QImr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVPQIto64mr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVSDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVSSmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVUPDYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVUPDmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVUPSYmr")>;
def: InstRW<[SBWriteResGroup33], (instregex "VMOVUPSmr")>;
def SBWriteResGroup34 : SchedWriteRes<[SBPort0,SBPort15]> {
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
def: InstRW<[SBWriteResGroup34], (instregex "MPSADBWrri")>;
def: InstRW<[SBWriteResGroup34], (instregex "VMPSADBWrri")>;
def SBWriteResGroup35 : SchedWriteRes<[SBPort1,SBPort5]> {
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup35], (instregex "CLI")>;
[X86] Rename some instructions that start with Int_ to have the _Int at the end. This matches AVX512 version and is more consistent overall. And improves our scheduler models. In some cases this adds _Int to instructions that didn't have any Int_ before. It's a side effect of the adjustments made to some of the multiclasses. llvm-svn: 320325
2017-12-11 03:47:56 +08:00
def: InstRW<[SBWriteResGroup35], (instregex "CVTSI642SSrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup35], (instregex "CVTSI2SSrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "HADDPDrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "HADDPSrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "HSUBPDrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "HSUBPSrr")>;
[X86] Rename some instructions that start with Int_ to have the _Int at the end. This matches AVX512 version and is more consistent overall. And improves our scheduler models. In some cases this adds _Int to instructions that didn't have any Int_ before. It's a side effect of the adjustments made to some of the multiclasses. llvm-svn: 320325
2017-12-11 03:47:56 +08:00
def: InstRW<[SBWriteResGroup35], (instregex "VCVTSI642SSrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup35], (instregex "VCVTSI2SSrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "VHADDPDYrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "VHADDPDrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "VHADDPSYrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "VHADDPSrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "VHSUBPDYrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "VHSUBPDrr")>;
def: InstRW<[SBWriteResGroup35], (instregex "VHSUBPSYrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup35], (instregex "VHSUBPSrr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def SBWriteResGroup35_2 : SchedWriteRes<[SBPort1,SBPort4,SBPort23]> {
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup35_2], (instregex "ISTT_FP16m")>;
def: InstRW<[SBWriteResGroup35_2], (instregex "ISTT_FP32m")>;
def: InstRW<[SBWriteResGroup35_2], (instregex "ISTT_FP64m")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup35_2], (instregex "PUSHGS64")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup36 : SchedWriteRes<[SBPort4,SBPort5,SBPort23]> {
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup36], (instregex "CALL64pcrel32")>;
def: InstRW<[SBWriteResGroup36], (instregex "CALL(16|32|64)r")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup36], (instregex "EXTRACTPSmr")>;
def: InstRW<[SBWriteResGroup36], (instregex "VEXTRACTPSmr")>;
def SBWriteResGroup37 : SchedWriteRes<[SBPort4,SBPort01,SBPort23]> {
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup37], (instregex "VMASKMOVPDYmr")>;
def: InstRW<[SBWriteResGroup37], (instregex "VMASKMOVPDmr")>;
def: InstRW<[SBWriteResGroup37], (instregex "VMASKMOVPSYmr")>;
def: InstRW<[SBWriteResGroup37], (instregex "VMASKMOVPSmr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup38 : SchedWriteRes<[SBPort4,SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup38], (instregex "SET(AE|B|E|G|GE|L|LE|NE|NO|NP|NS|O|P|S)m")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup39 : SchedWriteRes<[SBPort4,SBPort23,SBPort15]> {
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup39], (instregex "PEXTRBmr")>;
def: InstRW<[SBWriteResGroup39], (instregex "VPEXTRBmr")>;
def: InstRW<[SBWriteResGroup39], (instregex "VPEXTRDmr")>;
def: InstRW<[SBWriteResGroup39], (instregex "VPEXTRWmr")>;
def SBWriteResGroup40 : SchedWriteRes<[SBPort4,SBPort23,SBPort015]> {
let Latency = 5;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup40], (instregex "MOV8mi")>;
def: InstRW<[SBWriteResGroup40], (instregex "STOSB")>;
def: InstRW<[SBWriteResGroup40], (instregex "STOSL")>;
def: InstRW<[SBWriteResGroup40], (instregex "STOSQ")>;
def: InstRW<[SBWriteResGroup40], (instregex "STOSW")>;
def SBWriteResGroup41 : SchedWriteRes<[SBPort5,SBPort015]> {
let Latency = 5;
let NumMicroOps = 4;
let ResourceCycles = [1,3];
}
def: InstRW<[SBWriteResGroup41], (instregex "FNINIT")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup42 : SchedWriteRes<[SBPort05,SBPort015]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 5;
let NumMicroOps = 4;
let ResourceCycles = [1,3];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup42], (instregex "CMPXCHG(16|32|64)rr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup42], (instregex "CMPXCHG8rr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup43 : SchedWriteRes<[SBPort4,SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 5;
let NumMicroOps = 4;
let ResourceCycles = [1,1,2];
}
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup43], (instregex "SET(A|BE)m")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup44 : SchedWriteRes<[SBPort0,SBPort4,SBPort5,SBPort23]> {
let Latency = 5;
let NumMicroOps = 4;
let ResourceCycles = [1,1,1,1];
}
def: InstRW<[SBWriteResGroup44], (instregex "LDMXCSR")>;
def: InstRW<[SBWriteResGroup44], (instregex "STMXCSR")>;
def: InstRW<[SBWriteResGroup44], (instregex "VLDMXCSR")>;
def: InstRW<[SBWriteResGroup44], (instregex "VSTMXCSR")>;
def SBWriteResGroup45 : SchedWriteRes<[SBPort0,SBPort4,SBPort23,SBPort15]> {
let Latency = 5;
let NumMicroOps = 4;
let ResourceCycles = [1,1,1,1];
}
def: InstRW<[SBWriteResGroup45], (instregex "PEXTRDmr")>;
def: InstRW<[SBWriteResGroup45], (instregex "PEXTRQmr")>;
def: InstRW<[SBWriteResGroup45], (instregex "VPEXTRQmr")>;
def: InstRW<[SBWriteResGroup45], (instregex "PUSHF16")>;
def: InstRW<[SBWriteResGroup45], (instregex "PUSHF64")>;
def SBWriteResGroup46 : SchedWriteRes<[SBPort4,SBPort5,SBPort01,SBPort23]> {
let Latency = 5;
let NumMicroOps = 4;
let ResourceCycles = [1,1,1,1];
}
def: InstRW<[SBWriteResGroup46], (instregex "CLFLUSH")>;
def SBWriteResGroup47 : SchedWriteRes<[SBPort4,SBPort5,SBPort01,SBPort23]> {
let Latency = 5;
let NumMicroOps = 5;
let ResourceCycles = [1,2,1,1];
}
def: InstRW<[SBWriteResGroup47], (instregex "FXRSTOR")>;
def SBWriteResGroup48 : SchedWriteRes<[SBPort23]> {
let Latency = 6;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup48], (instregex "LDDQUrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MMX_MOVD64from64rm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOV64toPQIrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVAPDrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVAPSrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVDDUPrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVDI2PDIrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVDQArm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVDQUrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVNTDQArm")>;
[X86] Add MOVQI2PQIrm, MOVSDmr, and MOVSDrm to scheduler information The VEX versions were present but not the legacy SSE versions. llvm-svn: 320294
2017-12-10 17:14:44 +08:00
def: InstRW<[SBWriteResGroup48], (instregex "MOVQI2PQIrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVSDrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup48], (instregex "MOVSHDUPrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVSLDUPrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVSSrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVUPDrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "MOVUPSrm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup48], (instregex "POP(16|32|64)r")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup48], (instregex "VBROADCASTSSrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VLDDQUYrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VLDDQUrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOV64toPQIrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVAPDrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVAPSrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVDDUPrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVDI2PDIrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVDQArm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVDQUrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVNTDQArm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVQI2PQIrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVSDrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVSHDUPrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVSLDUPrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVSSrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVUPDrm")>;
def: InstRW<[SBWriteResGroup48], (instregex "VMOVUPSrm")>;
def SBWriteResGroup49 : SchedWriteRes<[SBPort5,SBPort23]> {
let Latency = 6;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup49], (instregex "JMP(16|32|64)m")>;
Avoid unecessary opsize byte in segment move to memory Segment moves to memory are always 16-bit. Remove invalid 32 and 64 bit variants. Recommiting with missing clang inline assembly test change. Fixes PR34478. Reviewers: rnk, craig.topper Subscribers: llvm-commits, hiraditya Differential Revision: https://reviews.llvm.org/D39847 llvm-svn: 318797
2017-11-22 03:28:13 +08:00
def: InstRW<[SBWriteResGroup49], (instregex "MOV16sm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup50 : SchedWriteRes<[SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 6;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup50], (instregex "BT(16|32|64)mi8")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup51 : SchedWriteRes<[SBPort23,SBPort15]> {
let Latency = 6;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup51], (instregex "MMX_PABSBrm")>;
def: InstRW<[SBWriteResGroup51], (instregex "MMX_PABSDrm")>;
def: InstRW<[SBWriteResGroup51], (instregex "MMX_PABSWrm")>;
def: InstRW<[SBWriteResGroup51], (instregex "MMX_PALIGNRrmi")>;
def: InstRW<[SBWriteResGroup51], (instregex "MMX_PSHUFBrm")>;
def: InstRW<[SBWriteResGroup51], (instregex "MMX_PSIGNBrm")>;
def: InstRW<[SBWriteResGroup51], (instregex "MMX_PSIGNDrm")>;
def: InstRW<[SBWriteResGroup51], (instregex "MMX_PSIGNWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup52 : SchedWriteRes<[SBPort23,SBPort015]> {
let Latency = 6;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup52], (instregex "ADD(16|32|64)rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "ADD8rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "AND(16|32|64)rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "AND8rm")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup52], (instregex "CMP(16|32|64)mi")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup52], (instregex "CMP(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup52], (instregex "CMP(16|32|64)rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "CMP8mi")>;
def: InstRW<[SBWriteResGroup52], (instregex "CMP8mr")>;
def: InstRW<[SBWriteResGroup52], (instregex "CMP8rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "LODSL")>;
def: InstRW<[SBWriteResGroup52], (instregex "LODSQ")>;
def: InstRW<[SBWriteResGroup52], (instregex "OR(16|32|64)rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "OR8rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "SUB(16|32|64)rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "SUB8rm")>;
def: InstRW<[SBWriteResGroup52], (instregex "XOR(16|32|64)rm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup52], (instregex "XOR8rm")>;
def SBWriteResGroup53 : SchedWriteRes<[SBPort4,SBPort23]> {
let Latency = 6;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
def: InstRW<[SBWriteResGroup53], (instregex "ST_F32m")>;
def: InstRW<[SBWriteResGroup53], (instregex "ST_F64m")>;
def: InstRW<[SBWriteResGroup53], (instregex "ST_FP32m")>;
def: InstRW<[SBWriteResGroup53], (instregex "ST_FP64m")>;
def: InstRW<[SBWriteResGroup53], (instregex "ST_FP80m")>;
def SBWriteResGroup54 : SchedWriteRes<[SBPort23]> {
let Latency = 7;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup54], (instregex "VBROADCASTSDYrm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup54], (instregex "VBROADCASTSSYrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup54], (instregex "VMOVAPDYrm")>;
def: InstRW<[SBWriteResGroup54], (instregex "VMOVAPSYrm")>;
def: InstRW<[SBWriteResGroup54], (instregex "VMOVDDUPYrm")>;
def: InstRW<[SBWriteResGroup54], (instregex "VMOVDQAYrm")>;
def: InstRW<[SBWriteResGroup54], (instregex "VMOVDQUYrm")>;
def: InstRW<[SBWriteResGroup54], (instregex "VMOVSHDUPYrm")>;
def: InstRW<[SBWriteResGroup54], (instregex "VMOVSLDUPYrm")>;
def: InstRW<[SBWriteResGroup54], (instregex "VMOVUPDYrm")>;
def: InstRW<[SBWriteResGroup54], (instregex "VMOVUPSYrm")>;
def SBWriteResGroup55 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 7;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup55], (instregex "CVTPS2PDrm")>;
def: InstRW<[SBWriteResGroup55], (instregex "CVTSS2SDrm")>;
def: InstRW<[SBWriteResGroup55], (instregex "VCVTPS2PDYrm")>;
def: InstRW<[SBWriteResGroup55], (instregex "VCVTPS2PDrm")>;
def: InstRW<[SBWriteResGroup55], (instregex "VCVTSS2SDrm")>;
def: InstRW<[SBWriteResGroup55], (instregex "VTESTPDrm")>;
def: InstRW<[SBWriteResGroup55], (instregex "VTESTPSrm")>;
def SBWriteResGroup56 : SchedWriteRes<[SBPort5,SBPort23]> {
let Latency = 7;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup56], (instregex "ANDNPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "ANDNPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "ANDPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "ANDPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "INSERTPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "MOVHPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "MOVHPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "MOVLPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "MOVLPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "ORPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "ORPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "SHUFPDrmi")>;
def: InstRW<[SBWriteResGroup56], (instregex "SHUFPSrmi")>;
def: InstRW<[SBWriteResGroup56], (instregex "UNPCKHPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "UNPCKHPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "UNPCKLPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "UNPCKLPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VANDNPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VANDNPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VANDPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VANDPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VBROADCASTF128")>;
def: InstRW<[SBWriteResGroup56], (instregex "VINSERTPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VMOVHPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VMOVHPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VMOVLPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VMOVLPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VORPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VORPSrm")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup56], (instregex "VPERMILPDmi")>;
def: InstRW<[SBWriteResGroup56], (instregex "VPERMILPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VPERMILPSmi")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup56], (instregex "VPERMILPSrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup56], (instregex "VSHUFPDrmi")>;
def: InstRW<[SBWriteResGroup56], (instregex "VSHUFPSrmi")>;
def: InstRW<[SBWriteResGroup56], (instregex "VUNPCKHPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VUNPCKHPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VUNPCKLPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VUNPCKLPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VXORPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "VXORPSrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "XORPDrm")>;
def: InstRW<[SBWriteResGroup56], (instregex "XORPSrm")>;
def SBWriteResGroup57 : SchedWriteRes<[SBPort5,SBPort015]> {
let Latency = 7;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup57], (instregex "AESDECLASTrr")>;
def: InstRW<[SBWriteResGroup57], (instregex "AESDECrr")>;
def: InstRW<[SBWriteResGroup57], (instregex "AESENCLASTrr")>;
def: InstRW<[SBWriteResGroup57], (instregex "AESENCrr")>;
def: InstRW<[SBWriteResGroup57], (instregex "VAESDECLASTrr")>;
def: InstRW<[SBWriteResGroup57], (instregex "VAESDECrr")>;
def: InstRW<[SBWriteResGroup57], (instregex "VAESENCLASTrr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup57], (instregex "VAESENCrr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup58 : SchedWriteRes<[SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 7;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup58], (instregex "BLENDPDrmi")>;
def: InstRW<[SBWriteResGroup58], (instregex "BLENDPSrmi")>;
def: InstRW<[SBWriteResGroup58], (instregex "VBLENDPDrmi")>;
def: InstRW<[SBWriteResGroup58], (instregex "VBLENDPSrmi")>;
def: InstRW<[SBWriteResGroup58], (instregex "VINSERTF128rm")>;
def SBWriteResGroup59 : SchedWriteRes<[SBPort23,SBPort15]> {
let Latency = 7;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup59], (instregex "MMX_PADDQirm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PABSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PABSDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PABSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PACKSSDWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PACKSSWBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PACKUSDWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PACKUSWBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PADDBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PADDDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PADDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PADDSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PADDSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PADDUSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PADDUSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PADDWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PALIGNRrmi")>;
def: InstRW<[SBWriteResGroup59], (instregex "PAVGBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PAVGWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PBLENDWrmi")>;
def: InstRW<[SBWriteResGroup59], (instregex "PCMPEQBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PCMPEQDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PCMPEQQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PCMPEQWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PCMPGTBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PCMPGTDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PCMPGTWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PINSRBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PINSRDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PINSRQrm")>;
[X86] Adjust names of PINSRW/PEXTRW intructions between MMX/SSE/AVX/AVX512 for consistency and to maybe enable more regular expression compaction in the scheduler models. NFCI llvm-svn: 323352
2018-01-25 01:58:51 +08:00
def: InstRW<[SBWriteResGroup59], (instregex "PINSRWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup59], (instregex "PMAXSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMAXSDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMAXSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMAXUBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMAXUDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMAXUWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMINSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMINSDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMINSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMINUBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMINUDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMINUWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVSXBDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVSXBQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVSXBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVSXDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVSXWDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVSXWQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVZXBDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVZXBQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVZXBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVZXDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVZXWDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PMOVZXWQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSHUFBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSHUFDmi")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSHUFHWmi")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSHUFLWmi")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup59], (instregex "PSIGNBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSIGNDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSIGNWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup59], (instregex "PSUBBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSUBDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSUBQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSUBSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSUBSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSUBUSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSUBUSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PSUBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PUNPCKHBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PUNPCKHDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PUNPCKHQDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PUNPCKHWDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PUNPCKLBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PUNPCKLDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PUNPCKLQDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "PUNPCKLWDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPABSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPABSDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPABSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPACKSSDWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPACKSSWBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPACKUSDWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPACKUSWBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPADDBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPADDDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPADDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPADDSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPADDSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPADDUSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPADDUSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPADDWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPALIGNRrmi")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPAVGBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPAVGWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPBLENDWrmi")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPCMPEQBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPCMPEQDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPCMPEQQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPCMPEQWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPCMPGTBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPCMPGTDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPCMPGTWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPINSRBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPINSRDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPINSRQrm")>;
[X86] Adjust names of PINSRW/PEXTRW intructions between MMX/SSE/AVX/AVX512 for consistency and to maybe enable more regular expression compaction in the scheduler models. NFCI llvm-svn: 323352
2018-01-25 01:58:51 +08:00
def: InstRW<[SBWriteResGroup59], (instregex "VPINSRWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup59], (instregex "VPMAXSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMAXSDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMAXSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMAXUBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMAXUDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMAXUWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMINSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMINSDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMINSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMINUBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMINUDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMINUWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVSXBDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVSXBQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVSXBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVSXDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVSXWDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVSXWQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVZXBDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVZXBQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVZXBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVZXDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVZXWDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPMOVZXWQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSHUFBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSHUFDmi")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSHUFHWmi")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSHUFLWmi")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup59], (instregex "VPSIGNBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSIGNDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSIGNWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup59], (instregex "VPSUBBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSUBDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSUBQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSUBSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSUBSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSUBUSBrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSUBUSWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPSUBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPUNPCKHBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPUNPCKHDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPUNPCKHQDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPUNPCKHWDrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPUNPCKLBWrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPUNPCKLDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPUNPCKLQDQrm")>;
def: InstRW<[SBWriteResGroup59], (instregex "VPUNPCKLWDrm")>;
def SBWriteResGroup60 : SchedWriteRes<[SBPort23,SBPort015]> {
let Latency = 7;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup60], (instregex "PANDNrm")>;
def: InstRW<[SBWriteResGroup60], (instregex "PANDrm")>;
def: InstRW<[SBWriteResGroup60], (instregex "PORrm")>;
def: InstRW<[SBWriteResGroup60], (instregex "PXORrm")>;
def: InstRW<[SBWriteResGroup60], (instregex "VPANDNrm")>;
def: InstRW<[SBWriteResGroup60], (instregex "VPANDrm")>;
def: InstRW<[SBWriteResGroup60], (instregex "VPORrm")>;
def: InstRW<[SBWriteResGroup60], (instregex "VPXORrm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup61 : SchedWriteRes<[SBPort0,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 7;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup61], (instregex "VRCPPSYr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup61], (instregex "VRSQRTPSYr")>;
def SBWriteResGroup62 : SchedWriteRes<[SBPort5,SBPort23]> {
let Latency = 7;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SBWriteResGroup62], (instregex "VERRm")>;
def: InstRW<[SBWriteResGroup62], (instregex "VERWm")>;
def SBWriteResGroup63 : SchedWriteRes<[SBPort23,SBPort015]> {
let Latency = 7;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
def: InstRW<[SBWriteResGroup63], (instregex "LODSB")>;
def: InstRW<[SBWriteResGroup63], (instregex "LODSW")>;
def SBWriteResGroup64 : SchedWriteRes<[SBPort5,SBPort01,SBPort23]> {
let Latency = 7;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup64], (instregex "FARJMP64")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup65 : SchedWriteRes<[SBPort23,SBPort05,SBPort015]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 7;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup65], (instregex "ADC(16|32|64)rm")>;
def: InstRW<[SBWriteResGroup65], (instregex "ADC8rm")>;
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup65], (instregex "CMOV(AE|B|E|G|GE|L|LE|NE|NO|NP|NS|O|P|S)(16|32|64)rm")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup65], (instregex "SBB(16|32|64)rm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup65], (instregex "SBB8rm")>;
def SBWriteResGroup66 : SchedWriteRes<[SBPort0,SBPort4,SBPort23]> {
let Latency = 7;
let NumMicroOps = 4;
let ResourceCycles = [1,1,2];
}
def: InstRW<[SBWriteResGroup66], (instregex "FNSTSWm")>;
def SBWriteResGroup67 : SchedWriteRes<[SBPort1,SBPort5,SBPort015]> {
let Latency = 7;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup67], (instregex "SLDT(16|32|64)r")>;
def: InstRW<[SBWriteResGroup67], (instregex "STR(16|32|64)r")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup68 : SchedWriteRes<[SBPort4,SBPort5,SBPort23]> {
let Latency = 7;
let NumMicroOps = 4;
let ResourceCycles = [1,1,2];
}
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup68], (instregex "CALL(16|32|64)m")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup68], (instregex "FNSTCW16m")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup69 : SchedWriteRes<[SBPort4,SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 7;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup69], (instregex "BTC(16|32|64)mi8")>;
def: InstRW<[SBWriteResGroup69], (instregex "BTR(16|32|64)mi8")>;
def: InstRW<[SBWriteResGroup69], (instregex "BTS(16|32|64)mi8")>;
def: InstRW<[SBWriteResGroup69], (instregex "SAR(16|32|64)mi")>;
def: InstRW<[SBWriteResGroup69], (instregex "SAR8mi")>;
def: InstRW<[SBWriteResGroup69], (instregex "SHL(16|32|64)m1")>;
def: InstRW<[SBWriteResGroup69], (instregex "SHL(16|32|64)mi")>;
def: InstRW<[SBWriteResGroup69], (instregex "SHL8m1")>;
def: InstRW<[SBWriteResGroup69], (instregex "SHL8mi")>;
def: InstRW<[SBWriteResGroup69], (instregex "SHR(16|32|64)mi")>;
def: InstRW<[SBWriteResGroup69], (instregex "SHR8mi")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup70 : SchedWriteRes<[SBPort4,SBPort23,SBPort015]> {
let Latency = 7;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "ADD(16|32|64)mi")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "ADD(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup70], (instregex "ADD8mi")>;
def: InstRW<[SBWriteResGroup70], (instregex "ADD8mr")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "AND(16|32|64)mi")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "AND(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup70], (instregex "AND8mi")>;
def: InstRW<[SBWriteResGroup70], (instregex "AND8mr")>;
def: InstRW<[SBWriteResGroup70], (instregex "DEC(16|32|64)m")>;
def: InstRW<[SBWriteResGroup70], (instregex "DEC8m")>;
def: InstRW<[SBWriteResGroup70], (instregex "INC(16|32|64)m")>;
def: InstRW<[SBWriteResGroup70], (instregex "INC8m")>;
def: InstRW<[SBWriteResGroup70], (instregex "NEG(16|32|64)m")>;
def: InstRW<[SBWriteResGroup70], (instregex "NEG8m")>;
def: InstRW<[SBWriteResGroup70], (instregex "NOT(16|32|64)m")>;
def: InstRW<[SBWriteResGroup70], (instregex "NOT8m")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "OR(16|32|64)mi")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "OR(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup70], (instregex "OR8mi")>;
def: InstRW<[SBWriteResGroup70], (instregex "OR8mr")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "SUB(16|32|64)mi")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "SUB(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup70], (instregex "SUB8mi")>;
def: InstRW<[SBWriteResGroup70], (instregex "SUB8mr")>;
[X86] Change register&memory TEST instructions from MRMSrcMem to MRMDstMem Summary: Intel documentation shows the memory operand as the first operand. But we currently treat it as the second operand. Conceptually the order doesn't matter since it doesn't write memory. We have aliases to parse with the operands in either order and the isel matching is commutable. For the register&register form order does matter for the assembly parser. PR22995 was previously filed and fixed by changing the register&register form from MRMSrcReg to MRMDestReg to match gas. Ideally the memory form should match by using MRMDestMem. I believe this supercedes D38025 which was trying to switch the register&register form back to pre-PR22995. Reviewers: aymanmus, RKSimon, zvi Reviewed By: aymanmus Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38120 llvm-svn: 314639
2017-10-02 07:53:53 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "TEST(16|32|64)mr")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "TEST8mi")>;
[X86] Change register&memory TEST instructions from MRMSrcMem to MRMDstMem Summary: Intel documentation shows the memory operand as the first operand. But we currently treat it as the second operand. Conceptually the order doesn't matter since it doesn't write memory. We have aliases to parse with the operands in either order and the isel matching is commutable. For the register&register form order does matter for the assembly parser. PR22995 was previously filed and fixed by changing the register&register form from MRMSrcReg to MRMDestReg to match gas. Ideally the memory form should match by using MRMDestMem. I believe this supercedes D38025 which was trying to switch the register&register form back to pre-PR22995. Reviewers: aymanmus, RKSimon, zvi Reviewed By: aymanmus Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38120 llvm-svn: 314639
2017-10-02 07:53:53 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "TEST8mr")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "XOR(16|32|64)mi")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "XOR(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup70], (instregex "XOR8mi")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup70], (instregex "XOR8mr")>;
def SBWriteResGroup71 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 8;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup71], (instregex "MMX_PMADDUBSWrm")>;
def: InstRW<[SBWriteResGroup71], (instregex "MMX_PMULHRSWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup71], (instregex "VTESTPDYrm")>;
def: InstRW<[SBWriteResGroup71], (instregex "VTESTPSYrm")>;
def SBWriteResGroup72 : SchedWriteRes<[SBPort1,SBPort23]> {
let Latency = 8;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup72], (instregex "BSF(16|32|64)rm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup72], (instregex "BSR(16|32|64)rm")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup72], (instregex "CRC32r(16|32|64)m64")>;
def: InstRW<[SBWriteResGroup72], (instregex "CRC32r(16|32|64)m8")>;
def: InstRW<[SBWriteResGroup72], (instregex "FCOM32m")>;
def: InstRW<[SBWriteResGroup72], (instregex "FCOM64m")>;
def: InstRW<[SBWriteResGroup72], (instregex "FCOMP32m")>;
def: InstRW<[SBWriteResGroup72], (instregex "FCOMP64m")>;
[X86] Expand IMUL/MUL instregexs in Intel scheduler models. Add load latency to some of them in SkylakeClient model. The regular expressions and the imul names caused some instructions to be matched by multiple regexs creating unpredictable results. This changes them all to use explicit instrs instead. While doing this I also found that some instructions in Skylake were missing load latency so I fixed that too. llvm-svn: 323406
2018-01-25 14:57:42 +08:00
def: InstRW<[SBWriteResGroup72], (instrs MUL8m)>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup73 : SchedWriteRes<[SBPort5,SBPort23]> {
let Latency = 8;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup73], (instregex "VANDNPDYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VANDNPSYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VANDPDYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VANDPSYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VORPDYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VORPSYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VPERM2F128rm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VPERMILPDYmi")>;
def: InstRW<[SBWriteResGroup73], (instregex "VPERMILPDYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VPERMILPSYmi")>;
def: InstRW<[SBWriteResGroup73], (instregex "VPERMILPSYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VSHUFPDYrmi")>;
def: InstRW<[SBWriteResGroup73], (instregex "VSHUFPSYrmi")>;
def: InstRW<[SBWriteResGroup73], (instregex "VUNPCKHPDYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VUNPCKHPSYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VUNPCKLPDYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VUNPCKLPSYrm")>;
def: InstRW<[SBWriteResGroup73], (instregex "VXORPDYrm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup73], (instregex "VXORPSYrm")>;
def SBWriteResGroup74 : SchedWriteRes<[SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 8;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup74], (instregex "VBLENDPDYrmi")>;
def: InstRW<[SBWriteResGroup74], (instregex "VBLENDPSYrmi")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup75 : SchedWriteRes<[SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 8;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
def: InstRW<[SBWriteResGroup75], (instregex "BLENDVPDrm0")>;
def: InstRW<[SBWriteResGroup75], (instregex "BLENDVPSrm0")>;
def: InstRW<[SBWriteResGroup75], (instregex "VBLENDVPDrm")>;
def: InstRW<[SBWriteResGroup75], (instregex "VBLENDVPSrm")>;
def: InstRW<[SBWriteResGroup75], (instregex "VMASKMOVPDrm")>;
def: InstRW<[SBWriteResGroup75], (instregex "VMASKMOVPSrm")>;
def SBWriteResGroup76 : SchedWriteRes<[SBPort23,SBPort15]> {
let Latency = 8;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
def: InstRW<[SBWriteResGroup76], (instregex "PBLENDVBrr0")>;
def: InstRW<[SBWriteResGroup76], (instregex "VPBLENDVBrm")>;
def SBWriteResGroup77 : SchedWriteRes<[SBPort0,SBPort1,SBPort23]> {
let Latency = 8;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup77], (instregex "COMISDrm")>;
def: InstRW<[SBWriteResGroup77], (instregex "COMISSrm")>;
def: InstRW<[SBWriteResGroup77], (instregex "UCOMISDrm")>;
def: InstRW<[SBWriteResGroup77], (instregex "UCOMISSrm")>;
def: InstRW<[SBWriteResGroup77], (instregex "VCOMISDrm")>;
def: InstRW<[SBWriteResGroup77], (instregex "VCOMISSrm")>;
def: InstRW<[SBWriteResGroup77], (instregex "VUCOMISDrm")>;
def: InstRW<[SBWriteResGroup77], (instregex "VUCOMISSrm")>;
def SBWriteResGroup78 : SchedWriteRes<[SBPort0,SBPort5,SBPort23]> {
let Latency = 8;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup78], (instregex "PTESTrm")>;
def: InstRW<[SBWriteResGroup78], (instregex "VPTESTrm")>;
def SBWriteResGroup79 : SchedWriteRes<[SBPort0,SBPort23,SBPort15]> {
let Latency = 8;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup79], (instregex "PSLLDrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "PSLLQrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "PSLLWrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "PSRADrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "PSRAWrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "PSRLDrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "PSRLQrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "PSRLWrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "VPSLLDrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "VPSLLQrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "VPSLLWrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "VPSRADrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "VPSRAWrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "VPSRLDrm")>;
def: InstRW<[SBWriteResGroup79], (instregex "VPSRLQrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup79], (instregex "VPSRLWrm")>;
def SBWriteResGroup80 : SchedWriteRes<[SBPort23,SBPort15]> {
let Latency = 8;
let NumMicroOps = 4;
let ResourceCycles = [1,3];
}
[X86] Name the MMX phaddd instruction with 3 Ds instead of just 2. NFC llvm-svn: 323403
2018-01-25 12:45:32 +08:00
def: InstRW<[SBWriteResGroup80], (instregex "MMX_PHADDDrm")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup80], (instregex "MMX_PHADDSWrm")>;
def: InstRW<[SBWriteResGroup80], (instregex "MMX_PHADDWrm")>;
def: InstRW<[SBWriteResGroup80], (instregex "MMX_PHSUBDrm")>;
def: InstRW<[SBWriteResGroup80], (instregex "MMX_PHSUBSWrm")>;
def: InstRW<[SBWriteResGroup80], (instregex "MMX_PHSUBWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup81 : SchedWriteRes<[SBPort23,SBPort015]> {
let Latency = 8;
let NumMicroOps = 4;
let ResourceCycles = [1,3];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup81], (instregex "CMPXCHG(16|32|64)rm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup81], (instregex "CMPXCHG8rm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup82 : SchedWriteRes<[SBPort23,SBPort05,SBPort015]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 8;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
[X86] Make better use of instregex for cmovcc/setcc/jcc instructions in the Intel scheduler models. Combine all the separate condition codes into a singular expression when possible. llvm-svn: 322924
2018-01-19 13:47:32 +08:00
def: InstRW<[SBWriteResGroup82], (instregex "CMOV(A|BE)(16|32|64)rm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup83 : SchedWriteRes<[SBPort23,SBPort015]> {
let Latency = 8;
let NumMicroOps = 5;
let ResourceCycles = [2,3];
}
def: InstRW<[SBWriteResGroup83], (instregex "CMPSB")>;
def: InstRW<[SBWriteResGroup83], (instregex "CMPSL")>;
def: InstRW<[SBWriteResGroup83], (instregex "CMPSQ")>;
def: InstRW<[SBWriteResGroup83], (instregex "CMPSW")>;
def SBWriteResGroup84 : SchedWriteRes<[SBPort4,SBPort5,SBPort23]> {
let Latency = 8;
let NumMicroOps = 5;
let ResourceCycles = [1,2,2];
}
def: InstRW<[SBWriteResGroup84], (instregex "FLDCW16m")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup85 : SchedWriteRes<[SBPort4,SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 8;
let NumMicroOps = 5;
let ResourceCycles = [1,2,2];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup85], (instregex "ROL(16|32|64)mi")>;
def: InstRW<[SBWriteResGroup85], (instregex "ROL8mi")>;
def: InstRW<[SBWriteResGroup85], (instregex "ROR(16|32|64)mi")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup85], (instregex "ROR8mi")>;
def SBWriteResGroup86 : SchedWriteRes<[SBPort4,SBPort23,SBPort015]> {
let Latency = 8;
let NumMicroOps = 5;
let ResourceCycles = [1,2,2];
}
def: InstRW<[SBWriteResGroup86], (instregex "MOVSB")>;
def: InstRW<[SBWriteResGroup86], (instregex "MOVSL")>;
def: InstRW<[SBWriteResGroup86], (instregex "MOVSQ")>;
def: InstRW<[SBWriteResGroup86], (instregex "MOVSW")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup86], (instregex "XADD(16|32|64)rm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup86], (instregex "XADD8rm")>;
def SBWriteResGroup87 : SchedWriteRes<[SBPort4,SBPort5,SBPort01,SBPort23]> {
let Latency = 8;
let NumMicroOps = 5;
let ResourceCycles = [1,1,1,2];
}
def: InstRW<[SBWriteResGroup87], (instregex "FARCALL64")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup88 : SchedWriteRes<[SBPort4,SBPort23,SBPort05,SBPort015]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 8;
let NumMicroOps = 5;
let ResourceCycles = [1,2,1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup88], (instregex "SHLD(16|32|64)mri8")>;
def: InstRW<[SBWriteResGroup88], (instregex "SHRD(16|32|64)mri8")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup89 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 9;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup89], (instregex "MMX_PMULUDQirm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMADDUBSWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMADDWDrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMULDQrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMULHRSWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMULHUWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMULHWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMULLDrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMULLWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PMULUDQrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "PSADBWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMADDUBSWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMADDWDrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMULDQrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMULHRSWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMULHUWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMULHWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMULLDrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMULLWrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPMULUDQrm")>;
def: InstRW<[SBWriteResGroup89], (instregex "VPSADBWrm")>;
def SBWriteResGroup90 : SchedWriteRes<[SBPort1,SBPort23]> {
let Latency = 9;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup90], (instregex "ADDPDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "ADDPSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "ADDSDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "ADDSSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "ADDSUBPDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "ADDSUBPSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "CMPPDrmi")>;
def: InstRW<[SBWriteResGroup90], (instregex "CMPPSrmi")>;
[X86] Add CMPSDrr/rm to the scheduler models. Somehow CMPSSrr/rm was there and the VEX version was there, but this was consistently missing. llvm-svn: 320289
2017-12-10 17:14:37 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "CMPSDrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "CMPSSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "CVTDQ2PSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "CVTPS2DQrm")>;
[X86] Rename some instructions that start with Int_ to have the _Int at the end. This matches AVX512 version and is more consistent overall. And improves our scheduler models. In some cases this adds _Int to instructions that didn't have any Int_ before. It's a side effect of the adjustments made to some of the multiclasses. llvm-svn: 320325
2017-12-11 03:47:56 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "CVTSI642SDrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "CVTSI2SDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "CVTTPS2DQrm")>;
[X86] Add the commutable floating point min/max pseudo instructions to sandybridge,haswell,broadwell,skylakeclient scheduler models. llvm-svn: 320277
2017-12-10 09:24:05 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "MAX(C?)PDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MAX(C?)PSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MAX(C?)SDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MAX(C?)SSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MIN(C?)PDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MIN(C?)PSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MIN(C?)SDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MIN(C?)SSrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "MMX_CVTPI2PSirm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MMX_CVTPS2PIirm")>;
def: InstRW<[SBWriteResGroup90], (instregex "MMX_CVTTPS2PIirm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "POPCNT(16|32|64)rm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "ROUNDPDm")>;
def: InstRW<[SBWriteResGroup90], (instregex "ROUNDPSm")>;
def: InstRW<[SBWriteResGroup90], (instregex "ROUNDSDm")>;
def: InstRW<[SBWriteResGroup90], (instregex "ROUNDSSm")>;
def: InstRW<[SBWriteResGroup90], (instregex "SUBPDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "SUBPSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "SUBSDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "SUBSSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VADDPDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VADDPSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VADDSDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VADDSSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VADDSUBPDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VADDSUBPSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VCMPPDrmi")>;
def: InstRW<[SBWriteResGroup90], (instregex "VCMPPSrmi")>;
def: InstRW<[SBWriteResGroup90], (instregex "VCMPSDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VCMPSSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VCVTDQ2PSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VCVTPS2DQrm")>;
[X86] Rename some instructions that start with Int_ to have the _Int at the end. This matches AVX512 version and is more consistent overall. And improves our scheduler models. In some cases this adds _Int to instructions that didn't have any Int_ before. It's a side effect of the adjustments made to some of the multiclasses. llvm-svn: 320325
2017-12-11 03:47:56 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "VCVTSI642SDrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "VCVTSI2SDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VCVTTPS2DQrm")>;
[X86] Add the commutable floating point min/max pseudo instructions to sandybridge,haswell,broadwell,skylakeclient scheduler models. llvm-svn: 320277
2017-12-10 09:24:05 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "VMAX(C?)PDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VMAX(C?)PSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VMAX(C?)SDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VMAX(C?)SSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VMIN(C?)PDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VMIN(C?)PSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VMIN(C?)SDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VMIN(C?)SSrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup90], (instregex "VROUNDPDm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VROUNDPSm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VROUNDSDm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VROUNDSSm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VSUBPDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VSUBPSrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VSUBSDrm")>;
def: InstRW<[SBWriteResGroup90], (instregex "VSUBSSrm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup91 : SchedWriteRes<[SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 9;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup91], (instregex "VBLENDVPDYrm")>;
def: InstRW<[SBWriteResGroup91], (instregex "VBLENDVPSYrm")>;
def: InstRW<[SBWriteResGroup91], (instregex "VMASKMOVPDYrm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup91], (instregex "VMASKMOVPSYrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup92 : SchedWriteRes<[SBPort0,SBPort1,SBPort5]> {
let Latency = 9;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup92], (instregex "DPPDrri")>;
def: InstRW<[SBWriteResGroup92], (instregex "VDPPDrri")>;
def SBWriteResGroup93 : SchedWriteRes<[SBPort0,SBPort1,SBPort23]> {
let Latency = 9;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup93], (instregex "CVTSD2SI64rm")>;
def: InstRW<[SBWriteResGroup93], (instregex "CVTSD2SIrm")>;
def: InstRW<[SBWriteResGroup93], (instregex "CVTSS2SI64rm")>;
def: InstRW<[SBWriteResGroup93], (instregex "CVTSS2SIrm")>;
def: InstRW<[SBWriteResGroup93], (instregex "CVTTSD2SI64rm")>;
def: InstRW<[SBWriteResGroup93], (instregex "CVTTSD2SIrm")>;
def: InstRW<[SBWriteResGroup93], (instregex "CVTTSS2SI64rm")>;
def: InstRW<[SBWriteResGroup93], (instregex "CVTTSS2SIrm")>;
[X86] Expand IMUL/MUL instregexs in Intel scheduler models. Add load latency to some of them in SkylakeClient model. The regular expressions and the imul names caused some instructions to be matched by multiple regexs creating unpredictable results. This changes them all to use explicit instrs instead. While doing this I also found that some instructions in Skylake were missing load latency so I fixed that too. llvm-svn: 323406
2018-01-25 14:57:42 +08:00
def: InstRW<[SBWriteResGroup93], (instrs MUL16m, MUL32m, MUL64m)>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup94 : SchedWriteRes<[SBPort0,SBPort5,SBPort23]> {
let Latency = 9;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup94], (instregex "VPTESTYrm")>;
def SBWriteResGroup95 : SchedWriteRes<[SBPort5,SBPort01,SBPort23]> {
let Latency = 9;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup95], (instregex "LD_F32m")>;
def: InstRW<[SBWriteResGroup95], (instregex "LD_F64m")>;
def: InstRW<[SBWriteResGroup95], (instregex "LD_F80m")>;
def SBWriteResGroup96 : SchedWriteRes<[SBPort23,SBPort15]> {
let Latency = 9;
let NumMicroOps = 4;
let ResourceCycles = [1,3];
}
def: InstRW<[SBWriteResGroup96], (instregex "PHADDDrm")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup96], (instregex "PHADDSWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup96], (instregex "PHADDWrm")>;
def: InstRW<[SBWriteResGroup96], (instregex "PHSUBDrm")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup96], (instregex "PHSUBSWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup96], (instregex "PHSUBWrm")>;
def: InstRW<[SBWriteResGroup96], (instregex "VPHADDDrm")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup96], (instregex "VPHADDSWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup96], (instregex "VPHADDWrm")>;
def: InstRW<[SBWriteResGroup96], (instregex "VPHSUBDrm")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup96], (instregex "VPHSUBSWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup96], (instregex "VPHSUBWrm")>;
def SBWriteResGroup97 : SchedWriteRes<[SBPort1,SBPort4,SBPort23]> {
let Latency = 9;
let NumMicroOps = 4;
let ResourceCycles = [1,1,2];
}
def: InstRW<[SBWriteResGroup97], (instregex "IST_F16m")>;
def: InstRW<[SBWriteResGroup97], (instregex "IST_F32m")>;
def: InstRW<[SBWriteResGroup97], (instregex "IST_FP16m")>;
def: InstRW<[SBWriteResGroup97], (instregex "IST_FP32m")>;
def: InstRW<[SBWriteResGroup97], (instregex "IST_FP64m")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def SBWriteResGroup97_2 : SchedWriteRes<[SBPort4,SBPort23,SBPort05]> {
let Latency = 9;
let NumMicroOps = 6;
let ResourceCycles = [1,2,3];
}
def: InstRW<[SBWriteResGroup97_2], (instregex "ROL(16|32|64)mCL")>;
def: InstRW<[SBWriteResGroup97_2], (instregex "ROL8mCL")>;
def: InstRW<[SBWriteResGroup97_2], (instregex "ROR(16|32|64)mCL")>;
def: InstRW<[SBWriteResGroup97_2], (instregex "ROR8mCL")>;
def: InstRW<[SBWriteResGroup97_2], (instregex "SAR(16|32|64)mCL")>;
def: InstRW<[SBWriteResGroup97_2], (instregex "SAR8mCL")>;
def: InstRW<[SBWriteResGroup97_2], (instregex "SHL(16|32|64)mCL")>;
def: InstRW<[SBWriteResGroup97_2], (instregex "SHL8mCL")>;
def: InstRW<[SBWriteResGroup97_2], (instregex "SHR(16|32|64)mCL")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup97_2], (instregex "SHR8mCL")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup98 : SchedWriteRes<[SBPort4,SBPort23,SBPort015]> {
let Latency = 9;
let NumMicroOps = 6;
let ResourceCycles = [1,2,3];
}
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup98], (instregex "ADC(16|32|64)mi")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup98], (instregex "ADC8mi")>;
[X86] Fix scheduler models to support ADD32ri in addition to ADD32ri8. Similar for all sizes of AND/OR/XOR/SUB/ADC/SBB/CMP. llvm-svn: 320291
2017-12-10 17:14:39 +08:00
def: InstRW<[SBWriteResGroup98], (instregex "SBB(16|32|64)mi")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup98], (instregex "SBB8mi")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup99 : SchedWriteRes<[SBPort4,SBPort23,SBPort05,SBPort015]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 9;
let NumMicroOps = 6;
let ResourceCycles = [1,2,2,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup99], (instregex "ADC(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup99], (instregex "ADC8mr")>;
def: InstRW<[SBWriteResGroup99], (instregex "SBB(16|32|64)mr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup99], (instregex "SBB8mr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup100 : SchedWriteRes<[SBPort4,SBPort5,SBPort23,SBPort05,SBPort015]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 9;
let NumMicroOps = 6;
let ResourceCycles = [1,1,2,1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup100], (instregex "BT(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup100], (instregex "BTC(16|32|64)mr")>;
def: InstRW<[SBWriteResGroup100], (instregex "BTR(16|32|64)mr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup100], (instregex "BTS(16|32|64)mr")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup101 : SchedWriteRes<[SBPort1,SBPort23]> {
let Latency = 10;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup101], (instregex "ADD_F32m")>;
def: InstRW<[SBWriteResGroup101], (instregex "ADD_F64m")>;
def: InstRW<[SBWriteResGroup101], (instregex "ILD_F16m")>;
def: InstRW<[SBWriteResGroup101], (instregex "ILD_F32m")>;
def: InstRW<[SBWriteResGroup101], (instregex "ILD_F64m")>;
def: InstRW<[SBWriteResGroup101], (instregex "SUBR_F32m")>;
def: InstRW<[SBWriteResGroup101], (instregex "SUBR_F64m")>;
def: InstRW<[SBWriteResGroup101], (instregex "SUB_F32m")>;
def: InstRW<[SBWriteResGroup101], (instregex "SUB_F64m")>;
def: InstRW<[SBWriteResGroup101], (instregex "VADDPDYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VADDPSYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VADDSUBPDYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VADDSUBPSYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VCMPPDYrmi")>;
def: InstRW<[SBWriteResGroup101], (instregex "VCMPPSYrmi")>;
def: InstRW<[SBWriteResGroup101], (instregex "VCVTDQ2PSYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VCVTPS2DQYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VCVTTPS2DQYrm")>;
[X86] Add the commutable floating point min/max pseudo instructions to sandybridge,haswell,broadwell,skylakeclient scheduler models. llvm-svn: 320277
2017-12-10 09:24:05 +08:00
def: InstRW<[SBWriteResGroup101], (instregex "VMAX(C?)PDYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VMAX(C?)PSYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VMIN(C?)PDYrm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VMIN(C?)PSYrm")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup101], (instregex "VROUNDYPDm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VROUNDYPSm")>;
def: InstRW<[SBWriteResGroup101], (instregex "VSUBPDYrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup101], (instregex "VSUBPSYrm")>;
def SBWriteResGroup102 : SchedWriteRes<[SBPort0,SBPort1,SBPort23]> {
let Latency = 10;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def: InstRW<[SBWriteResGroup102], (instregex "VCVTSD2SI64rm")>;
def: InstRW<[SBWriteResGroup102], (instregex "VCVTSD2SIrm")>;
def: InstRW<[SBWriteResGroup102], (instregex "VCVTSS2SI64rm")>;
def: InstRW<[SBWriteResGroup102], (instregex "VCVTSS2SIrm")>;
def: InstRW<[SBWriteResGroup102], (instregex "VCVTTSD2SI64rm")>;
def: InstRW<[SBWriteResGroup102], (instregex "VCVTTSD2SIrm")>;
def: InstRW<[SBWriteResGroup102], (instregex "VCVTTSS2SI64rm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup102], (instregex "VCVTTSS2SIrm")>;
def SBWriteResGroup103 : SchedWriteRes<[SBPort1,SBPort5,SBPort23]> {
let Latency = 10;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup103], (instregex "CVTDQ2PDrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "CVTPD2DQrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "CVTPD2PSrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "CVTSD2SSrm")>;
[X86] Rename some instructions that start with Int_ to have the _Int at the end. This matches AVX512 version and is more consistent overall. And improves our scheduler models. In some cases this adds _Int to instructions that didn't have any Int_ before. It's a side effect of the adjustments made to some of the multiclasses. llvm-svn: 320325
2017-12-11 03:47:56 +08:00
def: InstRW<[SBWriteResGroup103], (instregex "CVTSI642SSrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup103], (instregex "CVTSI2SSrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "CVTTPD2DQrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "MMX_CVTPD2PIirm")>;
def: InstRW<[SBWriteResGroup103], (instregex "MMX_CVTPI2PDirm")>;
def: InstRW<[SBWriteResGroup103], (instregex "MMX_CVTTPD2PIirm")>;
def: InstRW<[SBWriteResGroup103], (instregex "VCVTDQ2PDYrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "VCVTDQ2PDrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "VCVTPD2DQrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "VCVTPD2PSrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "VCVTSD2SSrm")>;
[X86] Rename some instructions that start with Int_ to have the _Int at the end. This matches AVX512 version and is more consistent overall. And improves our scheduler models. In some cases this adds _Int to instructions that didn't have any Int_ before. It's a side effect of the adjustments made to some of the multiclasses. llvm-svn: 320325
2017-12-11 03:47:56 +08:00
def: InstRW<[SBWriteResGroup103], (instregex "VCVTSI642SSrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup103], (instregex "VCVTSI2SSrm")>;
def: InstRW<[SBWriteResGroup103], (instregex "VCVTTPD2DQrm")>;
[X86] Remove windows line endings. llvm-svn: 313862
2017-09-21 12:55:07 +08:00
def SBWriteResGroup103_2 : SchedWriteRes<[SBPort4,SBPort23,SBPort05,SBPort015]> {
let Latency = 10;
let NumMicroOps = 7;
let ResourceCycles = [1,2,3,1];
}
def: InstRW<[SBWriteResGroup103_2], (instregex "SHLD(16|32|64)mrCL")>;
def: InstRW<[SBWriteResGroup103_2], (instregex "SHRD(16|32|64)mrCL")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def SBWriteResGroup104 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 11;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup104], (instregex "MULPDrm")>;
def: InstRW<[SBWriteResGroup104], (instregex "MULPSrm")>;
def: InstRW<[SBWriteResGroup104], (instregex "MULSDrm")>;
def: InstRW<[SBWriteResGroup104], (instregex "MULSSrm")>;
def: InstRW<[SBWriteResGroup104], (instregex "PCMPGTQrm")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup104], (instregex "PHMINPOSUWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup104], (instregex "RCPPSm")>;
def: InstRW<[SBWriteResGroup104], (instregex "RCPSSm")>;
def: InstRW<[SBWriteResGroup104], (instregex "RSQRTPSm")>;
def: InstRW<[SBWriteResGroup104], (instregex "RSQRTSSm")>;
def: InstRW<[SBWriteResGroup104], (instregex "VMULPDrm")>;
def: InstRW<[SBWriteResGroup104], (instregex "VMULPSrm")>;
def: InstRW<[SBWriteResGroup104], (instregex "VMULSDrm")>;
def: InstRW<[SBWriteResGroup104], (instregex "VMULSSrm")>;
def: InstRW<[SBWriteResGroup104], (instregex "VPCMPGTQrm")>;
[X86] Remove 64/128/256 from MMX/SSE/AVX instruction names for overall consistency. NFC MMX instrutions all start with MMX_ so the 64 isn't needed for disambigutation. SSE/AVX1 instructions are assumed 128-bit so we don't need to say 128. AVX2 instructions should use a Y to indicate 256-bits. llvm-svn: 323402
2018-01-25 12:45:30 +08:00
def: InstRW<[SBWriteResGroup104], (instregex "VPHMINPOSUWrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup104], (instregex "VRCPPSm")>;
def: InstRW<[SBWriteResGroup104], (instregex "VRCPSSm")>;
def: InstRW<[SBWriteResGroup104], (instregex "VRSQRTPSm")>;
def: InstRW<[SBWriteResGroup104], (instregex "VRSQRTSSm")>;
def SBWriteResGroup105 : SchedWriteRes<[SBPort0]> {
let Latency = 11;
let NumMicroOps = 3;
let ResourceCycles = [3];
}
def: InstRW<[SBWriteResGroup105], (instregex "PCMPISTRIrr")>;
def: InstRW<[SBWriteResGroup105], (instregex "PCMPISTRM128rr")>;
def: InstRW<[SBWriteResGroup105], (instregex "VPCMPISTRIrr")>;
def: InstRW<[SBWriteResGroup105], (instregex "VPCMPISTRM128rr")>;
def SBWriteResGroup106 : SchedWriteRes<[SBPort1,SBPort23]> {
let Latency = 11;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SBWriteResGroup106], (instregex "FICOM16m")>;
def: InstRW<[SBWriteResGroup106], (instregex "FICOM32m")>;
def: InstRW<[SBWriteResGroup106], (instregex "FICOMP16m")>;
def: InstRW<[SBWriteResGroup106], (instregex "FICOMP32m")>;
def SBWriteResGroup107 : SchedWriteRes<[SBPort1,SBPort5,SBPort23]> {
let Latency = 11;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup107], (instregex "VCVTPD2DQYrm")>;
def: InstRW<[SBWriteResGroup107], (instregex "VCVTPD2PSYrm")>;
def: InstRW<[SBWriteResGroup107], (instregex "VCVTTPD2DQYrm")>;
def SBWriteResGroup108 : SchedWriteRes<[SBPort0,SBPort23,SBPort15]> {
let Latency = 11;
let NumMicroOps = 4;
let ResourceCycles = [1,1,2];
}
def: InstRW<[SBWriteResGroup108], (instregex "MPSADBWrmi")>;
def: InstRW<[SBWriteResGroup108], (instregex "VMPSADBWrmi")>;
def SBWriteResGroup109 : SchedWriteRes<[SBPort1,SBPort5,SBPort23]> {
let Latency = 11;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
def: InstRW<[SBWriteResGroup109], (instregex "HADDPDrm")>;
def: InstRW<[SBWriteResGroup109], (instregex "HADDPSrm")>;
def: InstRW<[SBWriteResGroup109], (instregex "HSUBPDrm")>;
def: InstRW<[SBWriteResGroup109], (instregex "HSUBPSrm")>;
def: InstRW<[SBWriteResGroup109], (instregex "VHADDPDrm")>;
def: InstRW<[SBWriteResGroup109], (instregex "VHADDPSrm")>;
def: InstRW<[SBWriteResGroup109], (instregex "VHSUBPDrm")>;
def: InstRW<[SBWriteResGroup109], (instregex "VHSUBPSrm")>;
def SBWriteResGroup110 : SchedWriteRes<[SBPort5]> {
let Latency = 12;
let NumMicroOps = 2;
let ResourceCycles = [2];
}
def: InstRW<[SBWriteResGroup110], (instregex "AESIMCrr")>;
def: InstRW<[SBWriteResGroup110], (instregex "VAESIMCrr")>;
def SBWriteResGroup111 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 12;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup111], (instregex "MUL_F32m")>;
def: InstRW<[SBWriteResGroup111], (instregex "MUL_F64m")>;
def: InstRW<[SBWriteResGroup111], (instregex "VMULPDYrm")>;
def: InstRW<[SBWriteResGroup111], (instregex "VMULPSYrm")>;
def SBWriteResGroup112 : SchedWriteRes<[SBPort0,SBPort1,SBPort5]> {
let Latency = 12;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
def: InstRW<[SBWriteResGroup112], (instregex "DPPSrri")>;
def: InstRW<[SBWriteResGroup112], (instregex "VDPPSYrri")>;
def: InstRW<[SBWriteResGroup112], (instregex "VDPPSrri")>;
def SBWriteResGroup113 : SchedWriteRes<[SBPort1,SBPort5,SBPort23]> {
let Latency = 12;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup113], (instregex "VHADDPDYrm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup113], (instregex "VHADDPSYrm")>;
def: InstRW<[SBWriteResGroup113], (instregex "VHSUBPDYrm")>;
def: InstRW<[SBWriteResGroup113], (instregex "VHSUBPSYrm")>;
def SBWriteResGroup114 : SchedWriteRes<[SBPort1,SBPort23]> {
let Latency = 13;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SBWriteResGroup114], (instregex "ADD_FI16m")>;
def: InstRW<[SBWriteResGroup114], (instregex "ADD_FI32m")>;
def: InstRW<[SBWriteResGroup114], (instregex "SUBR_FI16m")>;
def: InstRW<[SBWriteResGroup114], (instregex "SUBR_FI32m")>;
def: InstRW<[SBWriteResGroup114], (instregex "SUB_FI16m")>;
def: InstRW<[SBWriteResGroup114], (instregex "SUB_FI32m")>;
def SBWriteResGroup115 : SchedWriteRes<[SBPort5,SBPort23,SBPort015]> {
let Latency = 13;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup115], (instregex "AESDECLASTrm")>;
def: InstRW<[SBWriteResGroup115], (instregex "AESDECrm")>;
def: InstRW<[SBWriteResGroup115], (instregex "AESENCLASTrm")>;
def: InstRW<[SBWriteResGroup115], (instregex "AESENCrm")>;
def: InstRW<[SBWriteResGroup115], (instregex "VAESDECLASTrm")>;
def: InstRW<[SBWriteResGroup115], (instregex "VAESDECrm")>;
def: InstRW<[SBWriteResGroup115], (instregex "VAESENCLASTrm")>;
def: InstRW<[SBWriteResGroup115], (instregex "VAESENCrm")>;
def SBWriteResGroup116 : SchedWriteRes<[SBPort0]> {
let Latency = 14;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup116], (instregex "DIVPSrr")>;
def: InstRW<[SBWriteResGroup116], (instregex "DIVSSrr")>;
def: InstRW<[SBWriteResGroup116], (instregex "SQRTPSr")>;
def: InstRW<[SBWriteResGroup116], (instregex "SQRTSSr")>;
def: InstRW<[SBWriteResGroup116], (instregex "VDIVPSrr")>;
def: InstRW<[SBWriteResGroup116], (instregex "VDIVSSrr")>;
def: InstRW<[SBWriteResGroup116], (instregex "VSQRTPSr")>;
def SBWriteResGroup117 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 14;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup117], (instregex "VSQRTSSm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup118 : SchedWriteRes<[SBPort0,SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 14;
let NumMicroOps = 4;
let ResourceCycles = [2,1,1];
}
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def: InstRW<[SBWriteResGroup118], (instregex "VRCPPSYm")>;
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
def: InstRW<[SBWriteResGroup118], (instregex "VRSQRTPSYm")>;
def SBWriteResGroup119 : SchedWriteRes<[SBPort0,SBPort1,SBPort23]> {
let Latency = 15;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup119], (instregex "MUL_FI16m")>;
def: InstRW<[SBWriteResGroup119], (instregex "MUL_FI32m")>;
def SBWriteResGroup120 : SchedWriteRes<[SBPort0,SBPort1,SBPort5,SBPort23]> {
let Latency = 15;
let NumMicroOps = 4;
let ResourceCycles = [1,1,1,1];
}
def: InstRW<[SBWriteResGroup120], (instregex "DPPDrmi")>;
def: InstRW<[SBWriteResGroup120], (instregex "VDPPDrmi")>;
def SBWriteResGroup121 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 17;
let NumMicroOps = 4;
let ResourceCycles = [3,1];
}
def: InstRW<[SBWriteResGroup121], (instregex "PCMPISTRIrm")>;
def: InstRW<[SBWriteResGroup121], (instregex "PCMPISTRM128rm")>;
def: InstRW<[SBWriteResGroup121], (instregex "VPCMPISTRIrm")>;
def: InstRW<[SBWriteResGroup121], (instregex "VPCMPISTRM128rm")>;
def SBWriteResGroup122 : SchedWriteRes<[SBPort5,SBPort23]> {
let Latency = 18;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SBWriteResGroup122], (instregex "AESIMCrm")>;
def: InstRW<[SBWriteResGroup122], (instregex "VAESIMCrm")>;
def SBWriteResGroup123 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 20;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup123], (instregex "DIVPSrm")>;
def: InstRW<[SBWriteResGroup123], (instregex "DIVSSrm")>;
def: InstRW<[SBWriteResGroup123], (instregex "SQRTPSm")>;
def: InstRW<[SBWriteResGroup123], (instregex "SQRTSSm")>;
def: InstRW<[SBWriteResGroup123], (instregex "VDIVPSrm")>;
def: InstRW<[SBWriteResGroup123], (instregex "VDIVSSrm")>;
def: InstRW<[SBWriteResGroup123], (instregex "VSQRTPSm")>;
def SBWriteResGroup124 : SchedWriteRes<[SBPort0]> {
let Latency = 21;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup124], (instregex "VSQRTSDr")>;
def SBWriteResGroup125 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 21;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup125], (instregex "VSQRTSDm")>;
def SBWriteResGroup126 : SchedWriteRes<[SBPort0]> {
let Latency = 22;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup126], (instregex "DIVPDrr")>;
def: InstRW<[SBWriteResGroup126], (instregex "DIVSDrr")>;
def: InstRW<[SBWriteResGroup126], (instregex "SQRTPDr")>;
def: InstRW<[SBWriteResGroup126], (instregex "SQRTSDr")>;
def: InstRW<[SBWriteResGroup126], (instregex "VDIVPDrr")>;
def: InstRW<[SBWriteResGroup126], (instregex "VDIVSDrr")>;
def: InstRW<[SBWriteResGroup126], (instregex "VSQRTPDr")>;
def SBWriteResGroup127 : SchedWriteRes<[SBPort0]> {
let Latency = 24;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup127], (instregex "DIVR_FPrST0")>;
def: InstRW<[SBWriteResGroup127], (instregex "DIVR_FST0r")>;
def: InstRW<[SBWriteResGroup127], (instregex "DIVR_FrST0")>;
def: InstRW<[SBWriteResGroup127], (instregex "DIV_FPrST0")>;
def: InstRW<[SBWriteResGroup127], (instregex "DIV_FST0r")>;
def: InstRW<[SBWriteResGroup127], (instregex "DIV_FrST0")>;
def SBWriteResGroup128 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 28;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup128], (instregex "DIVPDrm")>;
def: InstRW<[SBWriteResGroup128], (instregex "DIVSDrm")>;
def: InstRW<[SBWriteResGroup128], (instregex "SQRTPDm")>;
def: InstRW<[SBWriteResGroup128], (instregex "SQRTSDm")>;
def: InstRW<[SBWriteResGroup128], (instregex "VDIVPDrm")>;
def: InstRW<[SBWriteResGroup128], (instregex "VDIVSDrm")>;
def: InstRW<[SBWriteResGroup128], (instregex "VSQRTPDm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup129 : SchedWriteRes<[SBPort0,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 29;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SBWriteResGroup129], (instregex "VDIVPSYrr")>;
def: InstRW<[SBWriteResGroup129], (instregex "VSQRTPSYr")>;
def SBWriteResGroup130 : SchedWriteRes<[SBPort0,SBPort23]> {
let Latency = 31;
let NumMicroOps = 2;
let ResourceCycles = [1,1];
}
def: InstRW<[SBWriteResGroup130], (instregex "DIVR_F32m")>;
def: InstRW<[SBWriteResGroup130], (instregex "DIVR_F64m")>;
def: InstRW<[SBWriteResGroup130], (instregex "DIV_F32m")>;
def: InstRW<[SBWriteResGroup130], (instregex "DIV_F64m")>;
def SBWriteResGroup131 : SchedWriteRes<[SBPort0,SBPort1,SBPort23]> {
let Latency = 34;
let NumMicroOps = 3;
let ResourceCycles = [1,1,1];
}
def: InstRW<[SBWriteResGroup131], (instregex "DIVR_FI16m")>;
def: InstRW<[SBWriteResGroup131], (instregex "DIVR_FI32m")>;
def: InstRW<[SBWriteResGroup131], (instregex "DIV_FI16m")>;
def: InstRW<[SBWriteResGroup131], (instregex "DIV_FI32m")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup132 : SchedWriteRes<[SBPort0,SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 36;
let NumMicroOps = 4;
let ResourceCycles = [2,1,1];
}
def: InstRW<[SBWriteResGroup132], (instregex "VDIVPSYrm")>;
def: InstRW<[SBWriteResGroup132], (instregex "VSQRTPSYm")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup133 : SchedWriteRes<[SBPort0,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 45;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SBWriteResGroup133], (instregex "VDIVPDYrr")>;
def: InstRW<[SBWriteResGroup133], (instregex "VSQRTPDYr")>;
[X86][SandyBridge] Additional updates to the SNB instructions scheduling information This is a continuation patch for commit r307529 which completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target (see also https://reviews.llvm.org/D35019). In this patch we added the scheduling information of additional SNB instructions that were missing from the patch commit r307529, fixed the scheduling of several resource groups that include only port0 instead of port05 (i.e., port0 OR port5) and fixed several incorrect instructions' scheduling in the r307529 commit. The patch also includes the X87 instructions which were missing in previous patch commit r307529 as reported in bugzilla bug 34080. Reviewers: zvi, RKSimon, chandlerc, igorb, m_zuckerman, craig.topper, aymanmus, dim Differential Revision: https://reviews.llvm.org/D36388 llvm-svn: 310792
2017-08-13 21:59:24 +08:00
def SBWriteResGroup134 : SchedWriteRes<[SBPort0,SBPort23,SBPort05]> {
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target. The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information. Please note that the patch extensively affects the X86 MC instr scheduling for SNB. Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX. The updated and extended information about each instruction includes the following details: •static latency of the instruction •number of uOps from which the instruction consists of •all ports used by the instruction's' uOPs For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5: def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> { let Latency = 9; let NumMicroOps = 6; let ResourceCycles = [1,2,2,1]; } def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>; def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>; Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script. Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb Differential Revision: https://reviews.llvm.org/D35019#inline-304691 llvm-svn: 307529
2017-07-10 17:53:16 +08:00
let Latency = 52;
let NumMicroOps = 4;
let ResourceCycles = [2,1,1];
}
def: InstRW<[SBWriteResGroup134], (instregex "VDIVPDYrm")>;
def: InstRW<[SBWriteResGroup134], (instregex "VSQRTPDYm")>;
def SBWriteResGroup135 : SchedWriteRes<[SBPort0]> {
let Latency = 114;
let NumMicroOps = 1;
let ResourceCycles = [1];
}
def: InstRW<[SBWriteResGroup135], (instregex "VSQRTSSr")>;
Add a scheduling model for Intel Sandy Bridge microarchitecture. The model isn't hooked up by this patch because the instruction set isn't fully annotated yet. llvm-svn: 177942
2013-03-26 07:37:17 +08:00
} // SchedModel