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/SystemZ/SystemZScheduleZ196.td

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1241 lines
56 KiB
TableGen
Raw Normal View History

[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//=- SystemZScheduleZ196.td - SystemZ Scheduling Definitions ---*- tblgen -*-=//
//
Update the file headers across all of the LLVM projects in the monorepo to reflect the new license. We understand that people may be surprised that we're moving the header entirely to discuss the new license. We checked this carefully with the Foundation's lawyer and we believe this is the correct approach. Essentially, all code in the project is now made available by the LLVM project under our new license, so you will see that the license headers include that license only. Some of our contributors have contributed code under our old license, and accordingly, we have retained a copy of our old license notice in the top-level files in each project and repository. llvm-svn: 351636
2019-01-19 16:50:56 +08:00
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//
//===----------------------------------------------------------------------===//
//
// This file defines the machine model for Z196 to support instruction
// scheduling and other instruction cost heuristics.
//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
// Pseudos expanded right after isel do not need to be modelled here.
//
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
def Z196Model : SchedMachineModel {
[SystemZ] Rework processor feature definitions and add -mcpu=archX support This patch implements two changes: - Move processor feature definition into a new file SystemZFeatures.td, and provide explicit lists of supported and unsupported features for each level of the z/Architecture. This allows specifying unsupported features in the scheduler definition files for each processor. - Add optional aliases for the -mcpu processor names according to the level of the z/Architecture, for compatibility with other compilers on the platform. The supported aliases are: -mcpu=arch8 equals -mcpu=z10 -mcpu=arch9 equals -mcpu=z196 -mcpu=arch10 equals -mcpu=zEC12 -mcpu=arch11 equals -mcpu=z13 llvm-svn: 285577
2016-10-31 22:33:29 +08:00
let UnsupportedFeatures = Arch9UnsupportedFeatures.List;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
let IssueWidth = 3;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
let MicroOpBufferSize = 40; // Issue queues
let LoadLatency = 1; // Optimistic load latency.
let PostRAScheduler = 1;
// Extra cycles for a mispredicted branch.
[SystemZ] Improve use of conditional instructions This patch moves formation of LOC-type instructions from (late) IfConversion to the early if-conversion pass, and in some cases additionally creates them directly from select instructions during DAG instruction selection. To make early if-conversion work, the patch implements the canInsertSelect / insertSelect callbacks. It also implements the commuteInstructionImpl and FoldImmediate callbacks to enable generation of the full range of LOC instructions. Finally, the patch adds support for all instructions of the load-store-on-condition-2 facility, which allows using LOC instructions also for high registers. Due to the use of the GRX32 register class to enable high registers, we now also have to handle the cases where there are still no single hardware instructions (conditional move from a low register to a high register or vice versa). These are converted back to a branch sequence after register allocation. Since the expandRAPseudos callback is not allowed to create new basic blocks, this requires a simple new pass, modelled after the ARM/AArch64 ExpandPseudos pass. Overall, this patch causes significantly more LOC-type instructions to be used, and results in a measurable performance improvement. llvm-svn: 288028
2016-11-28 21:34:08 +08:00
let MispredictPenalty = 16;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
}
let SchedModel = Z196Model in {
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
// These definitions need the SchedModel value. They could be put in a
// subtarget common include file, but it seems the include system in Tablegen
// currently (2016) rejects multiple includes of same file.
// Decoder grouping rules
let NumMicroOps = 1 in {
def : WriteRes<NormalGr, []>;
def : WriteRes<BeginGroup, []> { let BeginGroup = 1; }
def : WriteRes<EndGroup, []> { let EndGroup = 1; }
}
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
def : WriteRes<GroupAlone, []> {
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
let NumMicroOps = 3;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
let BeginGroup = 1;
let EndGroup = 1;
}
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : WriteRes<GroupAlone2, []> {
let NumMicroOps = 6;
let BeginGroup = 1;
let EndGroup = 1;
}
def : WriteRes<GroupAlone3, []> {
let NumMicroOps = 9;
let BeginGroup = 1;
let EndGroup = 1;
}
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
// Incoming latency removed from the register operand which is used together
// with a memory operand by the instruction.
def : ReadAdvance<RegReadAdv, 4>;
// LoadLatency (above) is not used for instructions in this file. This is
// instead the role of LSULatency, which is the latency value added to the
// result of loads and instructions with folded memory operands.
def : WriteRes<LSULatency, []> { let Latency = 4; let NumMicroOps = 0; }
let NumMicroOps = 0 in {
[SystemZ] Use tablegen loops in SchedModels NFC changes to make scheduler TableGen files more readable, by using loops instead of a lot of similar defs with just e.g. a latency value that changes. https://reviews.llvm.org/D49598 Review: Ulrich Weigand, Javed Abshar llvm-svn: 337909
2018-07-25 19:42:55 +08:00
foreach L = 1-30 in {
def : WriteRes<!cast<SchedWrite>("WLat"#L), []> { let Latency = L; }
}
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
}
// Execution units.
[SystemZ] Fixes in SchedModels for older subtargets. IssueWidth updated to reflect the capacity of the issue unit correctly. Correct number of FX and LS units modelled (2, was 1). Review: Ulrich Weigand llvm-svn: 286109
2016-11-07 22:47:25 +08:00
def Z196_FXUnit : ProcResource<2>;
def Z196_LSUnit : ProcResource<2>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
def Z196_FPUnit : ProcResource<1>;
[SystemZ] Add decimal integer instructions This adds the set of decimal integer (BCD) instructions for assembler / disassembler use. llvm-svn: 302646
2017-05-10 20:42:45 +08:00
def Z196_DFUnit : ProcResource<1>;
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def Z196_MCD : ProcResource<1>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Subtarget specific definitions of scheduling resources.
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
let NumMicroOps = 0 in {
[SystemZ] Use tablegen loops in SchedModels NFC changes to make scheduler TableGen files more readable, by using loops instead of a lot of similar defs with just e.g. a latency value that changes. https://reviews.llvm.org/D49598 Review: Ulrich Weigand, Javed Abshar llvm-svn: 337909
2018-07-25 19:42:55 +08:00
def : WriteRes<FXU, [Z196_FXUnit]>;
def : WriteRes<LSU, [Z196_LSUnit]>;
def : WriteRes<FPU, [Z196_FPUnit]>;
def : WriteRes<DFU, [Z196_DFUnit]>;
foreach Num = 2-6 in { let ResourceCycles = [Num] in {
def : WriteRes<!cast<SchedWrite>("FXU"#Num), [Z196_FXUnit]>;
def : WriteRes<!cast<SchedWrite>("LSU"#Num), [Z196_LSUnit]>;
def : WriteRes<!cast<SchedWrite>("FPU"#Num), [Z196_FPUnit]>;
def : WriteRes<!cast<SchedWrite>("DFU"#Num), [Z196_DFUnit]>;
}}
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
}
def : WriteRes<MCD, [Z196_MCD]> { let NumMicroOps = 3;
let BeginGroup = 1;
let EndGroup = 1; }
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// -------------------------- INSTRUCTIONS ---------------------------------- //
// InstRW constructs have been used in order to preserve the
// readability of the InstrInfo files.
// For each instruction, as matched by a regexp, provide a list of
// resources that it needs. These will be combined into a SchedClass.
//===----------------------------------------------------------------------===//
// Stack allocation
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "ADJDYNALLOC$")>; // Pseudo -> LA / LAY
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
[SystemZ] Refactor branch and conditional instruction patterns Rework patterns for branches, call & return instructions, compare-and-branch, compare-and-trap, and conditional move instructions. In particular, simplify creation of patterns for the extended opcodes of instructions that take a CC mask. Also, use semantical instruction classes for all the instructions instead of open-coding them in SystemZInstrInfo.td. Adds a couple of the basic branch instructions (that are unused for codegen) for the assembler/disassembler. llvm-svn: 286263
2016-11-09 02:30:50 +08:00
// Branch instructions
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Branch
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU, EndGroup], (instregex "(Call)?BRC(L)?(Asm.*)?$")>;
def : InstRW<[WLat1, LSU, EndGroup], (instregex "(Call)?J(G)?(Asm.*)?$")>;
def : InstRW<[WLat1, LSU, EndGroup], (instregex "(Call)?BC(R)?(Asm.*)?$")>;
def : InstRW<[WLat1, LSU, EndGroup], (instregex "(Call)?B(R)?(Asm.*)?$")>;
def : InstRW<[WLat1, FXU, LSU, GroupAlone], (instregex "BRCT(G|H)?$")>;
def : InstRW<[WLat1, FXU, LSU, GroupAlone], (instregex "BCT(G)?(R)?$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat1, FXU3, LSU, GroupAlone2],
[SystemZ] Add remaining branch instructions This patch adds assembler support for the remaining branch instructions: the non-relative branch on count variants, and all variants of branch on index. The only one of those that can be readily exploited for code generation is BRCTH (branch on count using a high 32-bit register as count). Do use it, however, it is necessary to also introduce a hew CHIMux pseudo to allow comparisons of a 32-bit value agains a short immediate to go into a high register as well (implemented via CHI/CIH). This causes a bit of codegen changes overall, but those have proven to be neutral (or even beneficial) in performance measurements. llvm-svn: 288029
2016-11-28 21:40:08 +08:00
(instregex "B(R)?X(H|L).*$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Refactor branch and conditional instruction patterns Rework patterns for branches, call & return instructions, compare-and-branch, compare-and-trap, and conditional move instructions. In particular, simplify creation of patterns for the extended opcodes of instructions that take a CC mask. Also, use semantical instruction classes for all the instructions instead of open-coding them in SystemZInstrInfo.td. Adds a couple of the basic branch instructions (that are unused for codegen) for the assembler/disassembler. llvm-svn: 286263
2016-11-09 02:30:50 +08:00
// Compare and branch
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, LSU, GroupAlone],
[SystemZ] Refactor branch and conditional instruction patterns Rework patterns for branches, call & return instructions, compare-and-branch, compare-and-trap, and conditional move instructions. In particular, simplify creation of patterns for the extended opcodes of instructions that take a CC mask. Also, use semantical instruction classes for all the instructions instead of open-coding them in SystemZInstrInfo.td. Adds a couple of the basic branch instructions (that are unused for codegen) for the assembler/disassembler. llvm-svn: 286263
2016-11-09 02:30:50 +08:00
(instregex "C(L)?(G)?(I|R)J(Asm.*)?$")>;
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, LSU, GroupAlone],
[SystemZ] Refactor branch and conditional instruction patterns Rework patterns for branches, call & return instructions, compare-and-branch, compare-and-trap, and conditional move instructions. In particular, simplify creation of patterns for the extended opcodes of instructions that take a CC mask. Also, use semantical instruction classes for all the instructions instead of open-coding them in SystemZInstrInfo.td. Adds a couple of the basic branch instructions (that are unused for codegen) for the assembler/disassembler. llvm-svn: 286263
2016-11-09 02:30:50 +08:00
(instregex "C(L)?(G)?(I|R)B(Call|Return|Asm.*)?$")>;
//===----------------------------------------------------------------------===//
// Trap instructions
//===----------------------------------------------------------------------===//
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Trap
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU, EndGroup], (instregex "(Cond)?Trap$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Compare and trap
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "C(G)?(I|R)T(Asm.*)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CL(G)?RT(Asm.*)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CL(F|G)IT(Asm.*)?$")>;
[SystemZ] Refactor branch and conditional instruction patterns Rework patterns for branches, call & return instructions, compare-and-branch, compare-and-trap, and conditional move instructions. In particular, simplify creation of patterns for the extended opcodes of instructions that take a CC mask. Also, use semantical instruction classes for all the instructions instead of open-coding them in SystemZInstrInfo.td. Adds a couple of the basic branch instructions (that are unused for codegen) for the assembler/disassembler. llvm-svn: 286263
2016-11-09 02:30:50 +08:00
//===----------------------------------------------------------------------===//
// Call and return instructions
//===----------------------------------------------------------------------===//
// Call
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU, FXU2, GroupAlone], (instregex "(Call)?BRAS$")>;
def : InstRW<[WLat1, LSU, FXU2, GroupAlone], (instregex "(Call)?BRASL$")>;
def : InstRW<[WLat1, LSU, FXU2, GroupAlone], (instregex "(Call)?BAS(R)?$")>;
def : InstRW<[WLat1, LSU, FXU2, GroupAlone], (instregex "TLS_(G|L)DCALL$")>;
[SystemZ] Refactor branch and conditional instruction patterns Rework patterns for branches, call & return instructions, compare-and-branch, compare-and-trap, and conditional move instructions. In particular, simplify creation of patterns for the extended opcodes of instructions that take a CC mask. Also, use semantical instruction classes for all the instructions instead of open-coding them in SystemZInstrInfo.td. Adds a couple of the basic branch instructions (that are unused for codegen) for the assembler/disassembler. llvm-svn: 286263
2016-11-09 02:30:50 +08:00
// Return
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU, EndGroup], (instregex "Return$")>;
def : InstRW<[WLat1, LSU, EndGroup], (instregex "CondReturn$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Move instructions
//===----------------------------------------------------------------------===//
// Moves
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "MV(G|H)?HI$")>;
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "MVI(Y)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Move character
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, LSU3, GroupAlone], (instregex "MVC$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "MVCL(E|U)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Pseudo -> reg move
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "COPY(_TO_REGCLASS)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "EXTRACT_SUBREG$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "INSERT_SUBREG$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "REG_SEQUENCE$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Loads
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "L(Y|FH|RL|Mux)?$")>;
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "LG(RL)?$")>;
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "L128$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LLIH(F|H|L)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LLIL(F|H|L)$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LG(F|H)I$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LHI(Mux)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LR(Mux)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Load and test
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, WLat1LSU, LSU, FXU, NormalGr], (instregex "LT(G)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LT(G)?R$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Stores
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "STG(RL)?$")>;
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "ST128$")>;
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "ST(Y|FH|RL|Mux)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// String moves.
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "MVST$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Refactor branch and conditional instruction patterns Rework patterns for branches, call & return instructions, compare-and-branch, compare-and-trap, and conditional move instructions. In particular, simplify creation of patterns for the extended opcodes of instructions that take a CC mask. Also, use semantical instruction classes for all the instructions instead of open-coding them in SystemZInstrInfo.td. Adds a couple of the basic branch instructions (that are unused for codegen) for the assembler/disassembler. llvm-svn: 286263
2016-11-09 02:30:50 +08:00
//===----------------------------------------------------------------------===//
// Conditional move instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2, FXU, EndGroup], (instregex "LOC(G)?R(Asm.*)?$")>;
def : InstRW<[WLat2LSU, RegReadAdv, FXU, LSU, EndGroup],
(instregex "LOC(G)?(Asm.*)?$")>;
def : InstRW<[WLat1, FXU, LSU, EndGroup], (instregex "STOC(G)?(Asm.*)?$")>;
[SystemZ] Refactor branch and conditional instruction patterns Rework patterns for branches, call & return instructions, compare-and-branch, compare-and-trap, and conditional move instructions. In particular, simplify creation of patterns for the extended opcodes of instructions that take a CC mask. Also, use semantical instruction classes for all the instructions instead of open-coding them in SystemZInstrInfo.td. Adds a couple of the basic branch instructions (that are unused for codegen) for the assembler/disassembler. llvm-svn: 286263
2016-11-09 02:30:50 +08:00
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Sign extensions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "L(B|H|G)R$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LG(B|H|F)R$")>;
def : InstRW<[WLat1LSU, WLat1LSU, FXU, LSU, NormalGr], (instregex "LTGF$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LTGFR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "LB(H|Mux)?$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "LH(Y)?$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "LH(H|Mux|RL)$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "LG(B|H|F)$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "LG(H|F)RL$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Zero extensions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LLCR(Mux)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LLHR(Mux)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LLG(C|H|F|T)R$")>;
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "LLC(Mux)?$")>;
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "LLH(Mux)?$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "LL(C|H)H$")>;
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "LLHRL$")>;
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "LLG(C|H|F|T|HRL|FRL)$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Truncations
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "STC(H|Y|Mux)?$")>;
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "STH(H|Y|RL|Mux)?$")>;
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "STCM(H|Y)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Multi-register moves
//===----------------------------------------------------------------------===//
// Load multiple (estimated average of 5 ops)
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat10, WLat10, LSU5, GroupAlone], (instregex "LM(H|Y|G)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Add missing arithmetic instructions This adds the remaining general arithmetic instructions for assembler / disassembler use. Most of these are not useful for codegen; a few might be, and those are listed in the README.txt for future improvements. llvm-svn: 302665
2017-05-10 22:18:47 +08:00
// Load multiple disjoint
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, WLat30, MCD], (instregex "LMD$")>;
[SystemZ] Add missing arithmetic instructions This adds the remaining general arithmetic instructions for assembler / disassembler use. Most of these are not useful for codegen; a few might be, and those are listed in the README.txt for future improvements. llvm-svn: 302665
2017-05-10 22:18:47 +08:00
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Store multiple (estimated average of 3 ops)
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU2, FXU5, GroupAlone], (instregex "STM(H|Y|G)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Byte swaps
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LRV(G)?R$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "LRV(G|H)?$")>;
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "STRV(G|H)?$")>;
def : InstRW<[WLat30, MCD], (instregex "MVCIN$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Load address instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LA(Y|RL)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Load the Global Offset Table address
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "GOT$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Absolute and Negation
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2, WLat2, FXU, NormalGr], (instregex "LP(G)?R$")>;
def : InstRW<[WLat3, WLat3, FXU2, GroupAlone], (instregex "L(N|P)GFR$")>;
def : InstRW<[WLat2, WLat2, FXU, NormalGr], (instregex "LN(R|GR)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LC(R|GR)$")>;
def : InstRW<[WLat2, WLat2, FXU2, GroupAlone], (instregex "LCGFR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Insertion
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, RegReadAdv, FXU, LSU, NormalGr], (instregex "IC(Y)?$")>;
def : InstRW<[WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "IC32(Y)?$")>;
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "ICM(H|Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "II(F|H|L)Mux$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "IIHF(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "IIHH(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "IIHL(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "IILF(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "IILH(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "IILL(64)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Addition
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "A(L)?(Y)?$")>;
def : InstRW<[WLat1LSU, WLat1LSU, FXU, LSU, NormalGr], (instregex "A(L)?SI$")>;
def : InstRW<[WLat2LSU, WLat2LSU, RegReadAdv, FXU2, LSU, GroupAlone],
(instregex "AH(Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AIH$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AFI(Mux)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AGFI$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AGHI(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AGR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AHI(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AHIMux(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AL(FI|HSIK)$")>;
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "ALGF$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "ALGHSIK$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "ALGF(I|R)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "ALGR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "ALR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "AR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "A(L)?HHHR$")>;
def : InstRW<[WLat2, WLat2, FXU2, GroupAlone], (instregex "A(L)?HHLR$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "ALSIH(N)?$")>;
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "A(L)?G$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "A(L)?GSI$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Logical addition with carry
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, WLat2LSU, RegReadAdv, FXU, LSU, GroupAlone],
(instregex "ALC(G)?$")>;
def : InstRW<[WLat2, WLat2, FXU, GroupAlone], (instregex "ALC(G)?R$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Add with sign extension (32 -> 64)
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, WLat2LSU, RegReadAdv, FXU2, LSU, GroupAlone],
(instregex "AGF$")>;
def : InstRW<[WLat2, WLat2, FXU2, GroupAlone], (instregex "AGFR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Subtraction
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "S(G|Y)?$")>;
def : InstRW<[WLat2LSU, WLat2LSU, RegReadAdv, FXU2, LSU, GroupAlone],
(instregex "SH(Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SGR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SLFI$")>;
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "SL(G|GF|Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SLGF(I|R)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SLGR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SLR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "S(L)?HHHR$")>;
def : InstRW<[WLat2, WLat2, FXU2, GroupAlone], (instregex "S(L)?HHLR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Subtraction with borrow
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, WLat2LSU, RegReadAdv, FXU, LSU, GroupAlone],
(instregex "SLB(G)?$")>;
def : InstRW<[WLat2, WLat2, FXU, GroupAlone], (instregex "SLB(G)?R$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Subtraction with sign extension (32 -> 64)
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, WLat2LSU, RegReadAdv, FXU2, LSU, GroupAlone],
(instregex "SGF$")>;
def : InstRW<[WLat2, WLat2, FXU2, GroupAlone], (instregex "SGFR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// AND
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "N(G|Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NGR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NI(FMux|HMux|LMux)$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "NI(Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NIHF(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NIHH(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NIHL(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NILF(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NILH(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NILL(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "NR(K)?$")>;
def : InstRW<[WLat5LSU, LSU2, FXU, GroupAlone], (instregex "NC$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// OR
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "O(G|Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OGR(K)?$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "OI(Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OI(FMux|HMux|LMux)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OIHF(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OIHH(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OIHL(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OILF(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OILH(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OILL(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "OR(K)?$")>;
def : InstRW<[WLat5LSU, LSU2, FXU, GroupAlone], (instregex "OC$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// XOR
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "X(G|Y)?$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "XI(Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "XIFMux$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "XGR(K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "XIHF(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "XILF(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "XR(K)?$")>;
def : InstRW<[WLat5LSU, LSU2, FXU, GroupAlone], (instregex "XC$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Multiplication
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat6LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "MS(GF|Y)?$")>;
def : InstRW<[WLat6, FXU, NormalGr], (instregex "MS(R|FI)$")>;
def : InstRW<[WLat8LSU, RegReadAdv, FXU, LSU, NormalGr], (instregex "MSG$")>;
def : InstRW<[WLat8, FXU, NormalGr], (instregex "MSGR$")>;
def : InstRW<[WLat6, FXU, NormalGr], (instregex "MSGF(I|R)$")>;
def : InstRW<[WLat11LSU, RegReadAdv, FXU2, LSU, GroupAlone],
(instregex "MLG$")>;
def : InstRW<[WLat9, FXU2, GroupAlone], (instregex "MLGR$")>;
def : InstRW<[WLat5, FXU, NormalGr], (instregex "MGHI$")>;
def : InstRW<[WLat5, FXU, NormalGr], (instregex "MHI$")>;
def : InstRW<[WLat5LSU, RegReadAdv, FXU, LSU, NormalGr], (instregex "MH(Y)?$")>;
def : InstRW<[WLat7, FXU2, GroupAlone], (instregex "M(L)?R$")>;
def : InstRW<[WLat7LSU, RegReadAdv, FXU2, LSU, GroupAlone],
(instregex "M(FY|L)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Division and remainder
//===----------------------------------------------------------------------===//
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat30, FPU4, FXU5, GroupAlone3], (instregex "DR$")>;
def : InstRW<[WLat30, RegReadAdv, FPU4, LSU, FXU4, GroupAlone3],
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
(instregex "D$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat30, FPU4, FXU4, GroupAlone3], (instregex "DSG(F)?R$")>;
def : InstRW<[WLat30, RegReadAdv, FPU4, LSU, FXU3, GroupAlone3],
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
(instregex "DSG(F)?$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat30, FPU4, FXU5, GroupAlone3], (instregex "DL(G)?R$")>;
def : InstRW<[WLat30, RegReadAdv, FPU4, LSU, FXU4, GroupAlone3],
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
(instregex "DL(G)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Shifts
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SLL(G|K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SRL(G|K)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "SRA(G|K)?$")>;
def : InstRW<[WLat2, WLat2, FXU, NormalGr], (instregex "SLA(G|K)?$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat5LSU, WLat5LSU, FXU4, LSU, GroupAlone2],
[SystemZ] Minor fixing in SystemZScheduleZ196.td Some minor corrections for the recently added instructions. Review: Ulrich Weigand llvm-svn: 308028
2017-07-14 22:30:46 +08:00
(instregex "S(L|R)D(A|L)$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Rotate
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, FXU, LSU, NormalGr], (instregex "RLL(G)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Rotate and insert
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "RISBG(32)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "RISBH(G|H|L)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "RISBL(G|H|L)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "RISBMux$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Rotate and Select
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat3, WLat3, FXU2, GroupAlone], (instregex "R(N|O|X)SBG$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Comparison
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, RegReadAdv, FXU, LSU, NormalGr], (instregex "C(G|Y|Mux|RL)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "C(F|H)I(Mux)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CG(F|H)I$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CG(HSI|RL)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "C(G)?R$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CIH$")>;
def : InstRW<[WLat1LSU, RegReadAdv, FXU, LSU, NormalGr], (instregex "CHF$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CHSI$")>;
def : InstRW<[WLat1LSU, RegReadAdv, FXU, LSU, NormalGr],
(instregex "CL(Y|Mux)?$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CLFHSI$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CLFI(Mux)?$")>;
def : InstRW<[WLat1LSU, RegReadAdv, FXU, LSU, NormalGr], (instregex "CLG$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CLG(HRL|HSI)$")>;
def : InstRW<[WLat1LSU, RegReadAdv, FXU, LSU, NormalGr], (instregex "CLGF$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CLGFRL$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CLGF(I|R)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CLGR$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CLGRL$")>;
def : InstRW<[WLat1LSU, RegReadAdv, FXU, LSU, NormalGr], (instregex "CLHF$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CLH(RL|HSI)$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CLIH$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CLI(Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "CLR$")>;
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "CLRL$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "C(L)?HHR$")>;
def : InstRW<[WLat2, FXU2, GroupAlone], (instregex "C(L)?HLR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Compare halfword
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, RegReadAdv, FXU2, LSU, GroupAlone],
(instregex "CH(Y)?$")>;
def : InstRW<[WLat2LSU, FXU2, LSU, GroupAlone], (instregex "CHRL$")>;
def : InstRW<[WLat2LSU, RegReadAdv, FXU2, LSU, GroupAlone], (instregex "CGH$")>;
def : InstRW<[WLat2LSU, FXU2, LSU, GroupAlone], (instregex "CGHRL$")>;
def : InstRW<[WLat2LSU, FXU2, LSU, GroupAlone], (instregex "CHHSI$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Compare with sign extension (32 -> 64)
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, RegReadAdv, FXU2, LSU, GroupAlone], (instregex "CGF$")>;
def : InstRW<[WLat2LSU, FXU2, LSU, GroupAlone], (instregex "CGFRL$")>;
def : InstRW<[WLat2, FXU2, GroupAlone], (instregex "CGFR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Compare logical character
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, FXU, LSU2, GroupAlone], (instregex "CLC$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "CLCL(E|U)?$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "CLST$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Test under mask
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, FXU, LSU, NormalGr], (instregex "TM(Y)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "TM(H|L)Mux$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "TMHH(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "TMHL(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "TMLH(64)?$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "TMLL(64)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Add missing arithmetic instructions This adds the remaining general arithmetic instructions for assembler / disassembler use. Most of these are not useful for codegen; a few might be, and those are listed in the README.txt for future improvements. llvm-svn: 302665
2017-05-10 22:18:47 +08:00
// Compare logical characters under mask
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, RegReadAdv, FXU2, LSU, GroupAlone],
(instregex "CLM(H|Y)?$")>;
[SystemZ] Add missing arithmetic instructions This adds the remaining general arithmetic instructions for assembler / disassembler use. Most of these are not useful for codegen; a few might be, and those are listed in the README.txt for future improvements. llvm-svn: 302665
2017-05-10 22:18:47 +08:00
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Prefetch
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU, GroupAlone], (instregex "PFD(RL)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Atomic operations
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU, EndGroup], (instregex "Serialize$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, WLat1LSU, FXU, LSU, NormalGr], (instregex "LAA(G)?$")>;
def : InstRW<[WLat1LSU, WLat1LSU, FXU, LSU, NormalGr], (instregex "LAAL(G)?$")>;
def : InstRW<[WLat1LSU, WLat1LSU, FXU, LSU, NormalGr], (instregex "LAN(G)?$")>;
def : InstRW<[WLat1LSU, WLat1LSU, FXU, LSU, NormalGr], (instregex "LAO(G)?$")>;
def : InstRW<[WLat1LSU, WLat1LSU, FXU, LSU, NormalGr], (instregex "LAX(G)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
// Test and set
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1LSU, FXU, LSU, EndGroup], (instregex "TS$")>;
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Compare and swap
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, WLat2LSU, FXU2, LSU, GroupAlone],
(instregex "CS(G|Y)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
// Compare double and swap
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat5LSU, WLat5LSU, FXU5, LSU, GroupAlone2],
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
(instregex "CDS(Y)?$")>;
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat12, WLat12, FXU6, LSU2, GroupAlone],
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
(instregex "CDSG$")>;
// Compare and swap and store
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "CSST$")>;
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
// Perform locked operation
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "PLO$")>;
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
// Load/store pair from/to quadword
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat4LSU, LSU2, GroupAlone], (instregex "LPQ$")>;
def : InstRW<[WLat1, FXU2, LSU2, GroupAlone], (instregex "STPQ$")>;
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
// Load pair disjoint
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat2LSU, WLat2LSU, LSU2, GroupAlone], (instregex "LPD(G)?$")>;
[SystemZ] Support remaining atomic instructions Add assembler support for all atomic instructions that weren't already supported. Some of those could be used to implement codegen for 128-bit atomic operations, but this isn't done here yet. llvm-svn: 288526
2016-12-03 02:24:16 +08:00
[SystemZ] Add translate/convert instructions This adds the set of character-set translate and convert instructions for assembler / disassembler use. llvm-svn: 302644
2017-05-10 20:41:12 +08:00
//===----------------------------------------------------------------------===//
// Translate and convert
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "TR$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "TRT$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "TRTR$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "TRE$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "TRT(R)?E(Opt)?$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "TR(T|O)(T|O)(Opt)?$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD],
(instregex "CU(12|14|21|24|41|42)(Opt)?$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "(CUUTF|CUTFU)(Opt)?$")>;
[SystemZ] Add translate/convert instructions This adds the set of character-set translate and convert instructions for assembler / disassembler use. llvm-svn: 302644
2017-05-10 20:41:12 +08:00
[SystemZ] Add crypto instructions This adds the set of message-security assist instructions for assembler / disassembler use. llvm-svn: 302645
2017-05-10 20:42:00 +08:00
//===----------------------------------------------------------------------===//
// Message-security assist
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, WLat30, WLat30, WLat30, MCD],
(instregex "KM(C|F|O|CTR)?$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "(KIMD|KLMD|KMAC|PCC)$")>;
[SystemZ] Add crypto instructions This adds the set of message-security assist instructions for assembler / disassembler use. llvm-svn: 302645
2017-05-10 20:42:00 +08:00
[SystemZ] Add decimal integer instructions This adds the set of decimal integer (BCD) instructions for assembler / disassembler use. llvm-svn: 302646
2017-05-10 20:42:45 +08:00
//===----------------------------------------------------------------------===//
// Decimal arithmetic
//===----------------------------------------------------------------------===//
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat30, RegReadAdv, FXU, DFU2, LSU2, GroupAlone2],
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
(instregex "CVBG$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat20, RegReadAdv, FXU, DFU, LSU, GroupAlone2],
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
(instregex "CVB(Y)?$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat1, FXU3, DFU4, LSU, GroupAlone3], (instregex "CVDG$")>;
def : InstRW<[WLat1, FXU2, DFU, LSU, GroupAlone3], (instregex "CVD(Y)?$")>;
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU5, GroupAlone], (instregex "MV(N|O|Z)$")>;
def : InstRW<[WLat1, LSU5, GroupAlone], (instregex "(PACK|PKA|PKU)$")>;
def : InstRW<[WLat10, LSU5, GroupAlone], (instregex "UNPK(A|U)$")>;
def : InstRW<[WLat1, LSU5, GroupAlone], (instregex "UNPK$")>;
[SystemZ] Add decimal integer instructions This adds the set of decimal integer (BCD) instructions for assembler / disassembler use. llvm-svn: 302646
2017-05-10 20:42:45 +08:00
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat11LSU, FXU, DFU4, LSU2, GroupAlone],
[SystemZ] Add decimal integer instructions This adds the set of decimal integer (BCD) instructions for assembler / disassembler use. llvm-svn: 302646
2017-05-10 20:42:45 +08:00
(instregex "(A|S|ZA)P$")>;
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, DFU4, LSU2, GroupAlone], (instregex "(M|D)P$")>;
def : InstRW<[WLat15, FXU2, DFU4, LSU3, GroupAlone], (instregex "SRP$")>;
def : InstRW<[WLat11, DFU4, LSU2, GroupAlone], (instregex "CP$")>;
def : InstRW<[WLat5LSU, DFU2, LSU2, GroupAlone], (instregex "TP$")>;
def : InstRW<[WLat30, MCD], (instregex "ED(MK)?$")>;
[SystemZ] Add decimal integer instructions This adds the set of decimal integer (BCD) instructions for assembler / disassembler use. llvm-svn: 302646
2017-05-10 20:42:45 +08:00
[SystemZ] Model access registers as LLVM registers Add the 16 access registers as LLVM registers. This allows removing a lot of special cases in the assembler and disassembler where we were handling access registers; this can all just use the generic register code now. Also add a bunch of instructions to operate on access registers, for assembler/disassembler use only. No change in code generation intended. llvm-svn: 286283
2016-11-09 04:15:26 +08:00
//===----------------------------------------------------------------------===//
// Access registers
//===----------------------------------------------------------------------===//
// Extract/set/copy access register
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat3, LSU, NormalGr], (instregex "(EAR|SAR|CPYA)$")>;
[SystemZ] Model access registers as LLVM registers Add the 16 access registers as LLVM registers. This allows removing a lot of special cases in the assembler and disassembler where we were handling access registers; this can all just use the generic register code now. Also add a bunch of instructions to operate on access registers, for assembler/disassembler use only. No change in code generation intended. llvm-svn: 286283
2016-11-09 04:15:26 +08:00
// Load address extended
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat5, LSU, FXU, GroupAlone], (instregex "LAE(Y)?$")>;
[SystemZ] Model access registers as LLVM registers Add the 16 access registers as LLVM registers. This allows removing a lot of special cases in the assembler and disassembler where we were handling access registers; this can all just use the generic register code now. Also add a bunch of instructions to operate on access registers, for assembler/disassembler use only. No change in code generation intended. llvm-svn: 286283
2016-11-09 04:15:26 +08:00
// Load/store access multiple (not modeled precisely)
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat10, WLat10, LSU5, GroupAlone], (instregex "LAM(Y)?$")>;
def : InstRW<[WLat1, FXU5, LSU5, GroupAlone], (instregex "STAM(Y)?$")>;
[SystemZ] Model access registers as LLVM registers Add the 16 access registers as LLVM registers. This allows removing a lot of special cases in the assembler and disassembler where we were handling access registers; this can all just use the generic register code now. Also add a bunch of instructions to operate on access registers, for assembler/disassembler use only. No change in code generation intended. llvm-svn: 286283
2016-11-09 04:15:26 +08:00
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
[SystemZ] Add program mask and addressing mode instructions Add several instructions that operate on the program mask or the addressing mode. These are not really needed for code generation under Linux, but are provided for completeness for the assembler/disassembler. llvm-svn: 286284
2016-11-09 04:17:02 +08:00
// Program mask and addressing mode
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// Insert Program Mask
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat3, FXU, EndGroup], (instregex "IPM$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Add program mask and addressing mode instructions Add several instructions that operate on the program mask or the addressing mode. These are not really needed for code generation under Linux, but are provided for completeness for the assembler/disassembler. llvm-svn: 286284
2016-11-09 04:17:02 +08:00
// Set Program Mask
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat3, LSU, EndGroup], (instregex "SPM$")>;
[SystemZ] Add program mask and addressing mode instructions Add several instructions that operate on the program mask or the addressing mode. These are not really needed for code generation under Linux, but are provided for completeness for the assembler/disassembler. llvm-svn: 286284
2016-11-09 04:17:02 +08:00
// Branch and link
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU2, LSU, GroupAlone], (instregex "BAL(R)?$")>;
[SystemZ] Add program mask and addressing mode instructions Add several instructions that operate on the program mask or the addressing mode. These are not really needed for code generation under Linux, but are provided for completeness for the assembler/disassembler. llvm-svn: 286284
2016-11-09 04:17:02 +08:00
// Test addressing mode
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "TAM$")>;
[SystemZ] Add program mask and addressing mode instructions Add several instructions that operate on the program mask or the addressing mode. These are not really needed for code generation under Linux, but are provided for completeness for the assembler/disassembler. llvm-svn: 286284
2016-11-09 04:17:02 +08:00
// Set addressing mode
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, LSU, EndGroup], (instregex "SAM(24|31|64)$")>;
[SystemZ] Add program mask and addressing mode instructions Add several instructions that operate on the program mask or the addressing mode. These are not really needed for code generation under Linux, but are provided for completeness for the assembler/disassembler. llvm-svn: 286284
2016-11-09 04:17:02 +08:00
// Branch (and save) and set mode.
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, LSU, GroupAlone], (instregex "BSM$")>;
def : InstRW<[WLat1, FXU2, LSU, GroupAlone], (instregex "BASSM$")>;
[SystemZ] Add program mask and addressing mode instructions Add several instructions that operate on the program mask or the addressing mode. These are not really needed for code generation under Linux, but are provided for completeness for the assembler/disassembler. llvm-svn: 286284
2016-11-09 04:17:02 +08:00
//===----------------------------------------------------------------------===//
// Miscellaneous Instructions.
//===----------------------------------------------------------------------===//
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Find leftmost one
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, WLat7, FXU2, GroupAlone], (instregex "FLOGR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Population count
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat3, WLat3, FXU, NormalGr], (instregex "POPCNT$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// String instructions
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "SRST(U)?$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "CUSE$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Add miscellaneous instructions This adds a few missing instructions for the assembler and disassembler. Those should be the last missing general- purpose (Chapter 7) instructions for the z10 ISA. llvm-svn: 302667
2017-05-10 22:20:15 +08:00
// Various complex instructions
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, WLat30, WLat30, WLat30, MCD], (instregex "CFC$")>;
def : InstRW<[WLat30, WLat30, WLat30, WLat30, WLat30, WLat30, MCD],
(instregex "UPT$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "CKSM$")>;
def : InstRW<[WLat30, WLat30, WLat30, WLat30, MCD], (instregex "CMPSC$")>;
[SystemZ] Add miscellaneous instructions This adds a few missing instructions for the assembler and disassembler. Those should be the last missing general- purpose (Chapter 7) instructions for the z10 ISA. llvm-svn: 302667
2017-05-10 22:20:15 +08:00
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Execute
def : InstRW<[LSU, GroupAlone], (instregex "EX(RL)?$")>;
//===----------------------------------------------------------------------===//
// .insn directive instructions
//===----------------------------------------------------------------------===//
// An "empty" sched-class will be assigned instead of the "invalid sched-class".
// getNumDecoderSlots() will then return 1 instead of 0.
def : InstRW<[], (instregex "Insn.*")>;
// ----------------------------- Floating point ----------------------------- //
//===----------------------------------------------------------------------===//
// FP: Move instructions
//===----------------------------------------------------------------------===//
// Load zero
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LZ(DR|ER)$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat2, FXU2, GroupAlone2], (instregex "LZXR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Load
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LER$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LD(R|R32|GR)$")>;
def : InstRW<[WLat3, FXU, NormalGr], (instregex "LGDR$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat2, FXU2, GroupAlone2], (instregex "LXR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Load and Test
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, WLat9, FPU, NormalGr], (instregex "LT(E|D)BR$")>;
def : InstRW<[WLat9, FPU, NormalGr], (instregex "LT(E|D)BRCompare$")>;
def : InstRW<[WLat10, WLat10, FPU4, GroupAlone], (instregex "LTXBR(Compare)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Copy sign
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat5, FXU2, GroupAlone], (instregex "CPSDR(d|s)(d|s)$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// FP: Load instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "L(E|D)(Y|E32)?$")>;
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "LX$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// FP: Store instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "ST(E|D)(Y)?$")>;
def : InstRW<[WLat1, FXU, LSU, NormalGr], (instregex "STX$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// FP: Conversion instructions
//===----------------------------------------------------------------------===//
// Load rounded
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, FPU, NormalGr], (instregex "LEDBR(A)?$")>;
def : InstRW<[WLat9, FPU2, NormalGr], (instregex "L(E|D)XBR(A)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Load lengthened
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, FPU, LSU, NormalGr], (instregex "LDEB$")>;
def : InstRW<[WLat7, FPU, NormalGr], (instregex "LDEBR$")>;
def : InstRW<[WLat11LSU, FPU4, LSU, GroupAlone], (instregex "LX(E|D)B$")>;
def : InstRW<[WLat10, FPU4, GroupAlone], (instregex "LX(E|D)BR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Convert from fixed / logical
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat8, FXU, FPU, GroupAlone], (instregex "C(E|D)(F|G)BR(A)?$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat11, FXU, FPU4, GroupAlone2], (instregex "CX(F|G)BR(A?)$")>;
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat8, FXU, FPU, GroupAlone], (instregex "CEL(F|G)BR$")>;
def : InstRW<[WLat8, FXU, FPU, GroupAlone], (instregex "CDL(F|G)BR$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat11, FXU, FPU4, GroupAlone2], (instregex "CXL(F|G)BR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Convert to fixed / logical
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat12, WLat12, FXU, FPU, GroupAlone],
(instregex "C(F|G)(E|D)BR(A?)$")>;
def : InstRW<[WLat12, WLat12, FXU, FPU2, GroupAlone],
(instregex "C(F|G)XBR(A?)$")>;
def : InstRW<[WLat12, WLat12, FXU, FPU, GroupAlone],
(instregex "CL(F|G)(E|D)BR$")>;
def : InstRW<[WLat12, WLat12, FXU, FPU2, GroupAlone], (instregex "CL(F|G)XBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// FP: Unary arithmetic
//===----------------------------------------------------------------------===//
// Load Complement / Negative / Positive
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, WLat7, FPU, NormalGr], (instregex "L(C|N|P)(E|D)BR$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "L(C|N|P)DFR(_32)?$")>;
def : InstRW<[WLat10, WLat10, FPU4, GroupAlone], (instregex "L(C|N|P)XBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Square root
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, FPU, LSU, NormalGr], (instregex "SQ(E|D)B$")>;
def : InstRW<[WLat30, FPU, NormalGr], (instregex "SQ(E|D)BR$")>;
def : InstRW<[WLat30, FPU4, GroupAlone], (instregex "SQXBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Load FP integer
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, FPU, NormalGr], (instregex "FI(E|D)BR(A)?$")>;
def : InstRW<[WLat15, FPU4, GroupAlone], (instregex "FIXBR(A)?$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// FP: Binary arithmetic
//===----------------------------------------------------------------------===//
// Addition
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, WLat7LSU, RegReadAdv, FPU, LSU, NormalGr],
(instregex "A(E|D)B$")>;
def : InstRW<[WLat7, WLat7, FPU, NormalGr], (instregex "A(E|D)BR$")>;
def : InstRW<[WLat20, WLat20, FPU4, GroupAlone], (instregex "AXBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Subtraction
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, WLat7LSU, RegReadAdv, FPU, LSU, NormalGr],
(instregex "S(E|D)B$")>;
def : InstRW<[WLat7, WLat7, FPU, NormalGr], (instregex "S(E|D)BR$")>;
def : InstRW<[WLat20, WLat20, FPU4, GroupAlone], (instregex "SXBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Multiply
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, RegReadAdv, FPU, LSU, NormalGr],
(instregex "M(D|DE|EE)B$")>;
def : InstRW<[WLat7, FPU, NormalGr], (instregex "M(D|DE|EE)BR$")>;
def : InstRW<[WLat11LSU, RegReadAdv, FPU4, LSU, GroupAlone],
(instregex "MXDB$")>;
def : InstRW<[WLat10, FPU4, GroupAlone], (instregex "MXDBR$")>;
def : InstRW<[WLat30, FPU4, GroupAlone], (instregex "MXBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Multiply and add / subtract
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, RegReadAdv, RegReadAdv, FPU2, LSU, GroupAlone],
(instregex "M(A|S)EB$")>;
def : InstRW<[WLat7, FPU, GroupAlone], (instregex "M(A|S)EBR$")>;
def : InstRW<[WLat7LSU, RegReadAdv, RegReadAdv, FPU2, LSU, GroupAlone],
(instregex "M(A|S)DB$")>;
def : InstRW<[WLat7, FPU, GroupAlone], (instregex "M(A|S)DBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Division
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, RegReadAdv, FPU, LSU, NormalGr], (instregex "D(E|D)B$")>;
def : InstRW<[WLat30, FPU, NormalGr], (instregex "D(E|D)BR$")>;
def : InstRW<[WLat30, FPU4, GroupAlone], (instregex "DXBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Add missing arithmetic instructions This adds the remaining general arithmetic instructions for assembler / disassembler use. Most of these are not useful for codegen; a few might be, and those are listed in the README.txt for future improvements. llvm-svn: 302665
2017-05-10 22:18:47 +08:00
// Divide to integer
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "DI(E|D)BR$")>;
[SystemZ] Add missing arithmetic instructions This adds the remaining general arithmetic instructions for assembler / disassembler use. Most of these are not useful for codegen; a few might be, and those are listed in the README.txt for future improvements. llvm-svn: 302665
2017-05-10 22:18:47 +08:00
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
//===----------------------------------------------------------------------===//
// FP: Comparisons
//===----------------------------------------------------------------------===//
// Compare
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat11LSU, RegReadAdv, FPU, LSU, NormalGr],
(instregex "(K|C)(E|D)B$")>;
def : InstRW<[WLat9, FPU, NormalGr], (instregex "(K|C)(E|D)BR$")>;
def : InstRW<[WLat30, FPU2, NormalGr], (instregex "(K|C)XBR$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
// Test Data Class
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat15, FPU, LSU, NormalGr], (instregex "TC(E|D)B$")>;
def : InstRW<[WLat15, FPU4, LSU, GroupAlone], (instregex "TCXB$")>;
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
[SystemZ] Support floating-point control register instructions Add assembler support for instructions manipulating the FPC. Also add codegen support via the GCC compatibility builtins: __builtin_s390_sfpc __builtin_s390_efpc llvm-svn: 288525
2016-12-03 02:21:53 +08:00
//===----------------------------------------------------------------------===//
// FP: Floating-point control register instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat4, FXU, LSU, GroupAlone], (instregex "EFPC$")>;
def : InstRW<[WLat1, FXU, LSU, GroupAlone], (instregex "STFPC$")>;
def : InstRW<[WLat1, LSU, GroupAlone], (instregex "SFPC$")>;
def : InstRW<[WLat1, LSU2, GroupAlone], (instregex "LFPC$")>;
def : InstRW<[WLat30, MCD], (instregex "SFASR$")>;
def : InstRW<[WLat30, MCD], (instregex "LFAS$")>;
def : InstRW<[WLat2, FXU, GroupAlone], (instregex "SRNM(B|T)?$")>;
[SystemZ] Support floating-point control register instructions Add assembler support for instructions manipulating the FPC. Also add codegen support via the GCC compatibility builtins: __builtin_s390_sfpc __builtin_s390_efpc llvm-svn: 288525
2016-12-03 02:21:53 +08:00
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// --------------------- Hexadecimal floating point ------------------------- //
//===----------------------------------------------------------------------===//
// HFP: Move instructions
//===----------------------------------------------------------------------===//
// Load and Test
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, WLat9, FPU, NormalGr], (instregex "LT(E|D)R$")>;
def : InstRW<[WLat9, WLat9, FPU4, GroupAlone], (instregex "LTXR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
//===----------------------------------------------------------------------===//
// HFP: Conversion instructions
//===----------------------------------------------------------------------===//
// Load rounded
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, FPU, NormalGr], (instregex "(LEDR|LRER)$")>;
def : InstRW<[WLat7, FPU, NormalGr], (instregex "LEXR$")>;
def : InstRW<[WLat9, FPU, NormalGr], (instregex "(LDXR|LRDR)$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Load lengthened
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "LDE$")>;
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LDER$")>;
def : InstRW<[WLat11LSU, FPU4, LSU, GroupAlone], (instregex "LX(E|D)$")>;
def : InstRW<[WLat9, FPU4, GroupAlone], (instregex "LX(E|D)R$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Convert from fixed
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat8, FXU, FPU, GroupAlone], (instregex "C(E|D)(F|G)R$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat10, FXU, FPU4, GroupAlone2], (instregex "CX(F|G)R$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Convert to fixed
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat12, WLat12, FXU, FPU, GroupAlone],
(instregex "C(F|G)(E|D)R$")>;
def : InstRW<[WLat30, WLat30, FXU, FPU2, GroupAlone], (instregex "C(F|G)XR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Convert BFP to HFP / HFP to BFP.
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, WLat7, FPU, NormalGr], (instregex "THD(E)?R$")>;
def : InstRW<[WLat7, WLat7, FPU, NormalGr], (instregex "TB(E)?DR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
//===----------------------------------------------------------------------===//
// HFP: Unary arithmetic
//===----------------------------------------------------------------------===//
// Load Complement / Negative / Positive
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, WLat7, FPU, NormalGr], (instregex "L(C|N|P)(E|D)R$")>;
def : InstRW<[WLat9, WLat9, FPU4, GroupAlone], (instregex "L(C|N|P)XR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Halve
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, FPU, NormalGr], (instregex "H(E|D)R$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Square root
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, FPU, LSU, NormalGr], (instregex "SQ(E|D)$")>;
def : InstRW<[WLat30, FPU, NormalGr], (instregex "SQ(E|D)R$")>;
def : InstRW<[WLat30, FPU4, GroupAlone], (instregex "SQXR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Load FP integer
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, FPU, NormalGr], (instregex "FI(E|D)R$")>;
def : InstRW<[WLat15, FPU4, GroupAlone], (instregex "FIXR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
//===----------------------------------------------------------------------===//
// HFP: Binary arithmetic
//===----------------------------------------------------------------------===//
// Addition
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, WLat7LSU, RegReadAdv, FPU, LSU, NormalGr],
(instregex "A(E|D|U|W)$")>;
def : InstRW<[WLat7, WLat7, FPU, NormalGr], (instregex "A(E|D|U|W)R$")>;
def : InstRW<[WLat15, WLat15, FPU4, GroupAlone], (instregex "AXR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Subtraction
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, WLat7LSU, RegReadAdv, FPU, LSU, NormalGr],
(instregex "S(E|D|U|W)$")>;
def : InstRW<[WLat7, WLat7, FPU, NormalGr], (instregex "S(E|D|U|W)R$")>;
def : InstRW<[WLat15, WLat15, FPU4, GroupAlone], (instregex "SXR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Multiply
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, RegReadAdv, FPU, LSU, NormalGr], (instregex "M(D|EE)$")>;
def : InstRW<[WLat8LSU, RegReadAdv, FPU, LSU, NormalGr], (instregex "M(DE|E)$")>;
def : InstRW<[WLat7, FPU, NormalGr], (instregex "M(D|EE)R$")>;
def : InstRW<[WLat8, FPU, NormalGr], (instregex "M(DE|E)R$")>;
def : InstRW<[WLat11LSU, RegReadAdv, FPU4, LSU, GroupAlone], (instregex "MXD$")>;
def : InstRW<[WLat10, FPU4, GroupAlone], (instregex "MXDR$")>;
def : InstRW<[WLat30, FPU4, GroupAlone], (instregex "MXR$")>;
def : InstRW<[WLat11LSU, RegReadAdv, FPU4, LSU, GroupAlone], (instregex "MY$")>;
def : InstRW<[WLat7LSU, RegReadAdv, FPU2, LSU, GroupAlone],
(instregex "MY(H|L)$")>;
def : InstRW<[WLat10, FPU4, GroupAlone], (instregex "MYR$")>;
def : InstRW<[WLat7, FPU, GroupAlone], (instregex "MY(H|L)R$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Multiply and add / subtract
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, RegReadAdv, RegReadAdv, FPU2, LSU, GroupAlone],
(instregex "M(A|S)(E|D)$")>;
def : InstRW<[WLat7, FPU, GroupAlone], (instregex "M(A|S)(E|D)R$")>;
def : InstRW<[WLat11LSU, RegReadAdv, RegReadAdv, FPU4, LSU, GroupAlone],
(instregex "MAY$")>;
def : InstRW<[WLat7LSU, RegReadAdv, RegReadAdv, FPU2, LSU, GroupAlone],
(instregex "MAY(H|L)$")>;
def : InstRW<[WLat10, FPU4, GroupAlone], (instregex "MAYR$")>;
def : InstRW<[WLat7, FPU, GroupAlone], (instregex "MAY(H|L)R$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
// Division
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, RegReadAdv, FPU, LSU, NormalGr], (instregex "D(E|D)$")>;
def : InstRW<[WLat30, FPU, NormalGr], (instregex "D(E|D)R$")>;
def : InstRW<[WLat30, FPU4, GroupAlone], (instregex "DXR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
//===----------------------------------------------------------------------===//
// HFP: Comparisons
//===----------------------------------------------------------------------===//
// Compare
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat11LSU, RegReadAdv, FPU, LSU, NormalGr], (instregex "C(E|D)$")>;
def : InstRW<[WLat9, FPU, NormalGr], (instregex "C(E|D)R$")>;
def : InstRW<[WLat15, FPU2, NormalGr], (instregex "CXR$")>;
[SystemZ] Add hexadecimal floating-point instructions This adds assembler / disassembler support for the hexadecimal floating-point instructions. Since the Linux ABI does not use any hex float data types, these are not useful for codegen. llvm-svn: 304202
2017-05-30 18:13:23 +08:00
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// ------------------------ Decimal floating point -------------------------- //
//===----------------------------------------------------------------------===//
// DFP: Move instructions
//===----------------------------------------------------------------------===//
// Load and Test
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat4, WLat4, DFU, NormalGr], (instregex "LTDTR$")>;
def : InstRW<[WLat6, WLat6, DFU4, GroupAlone], (instregex "LTXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
//===----------------------------------------------------------------------===//
// DFP: Conversion instructions
//===----------------------------------------------------------------------===//
// Load rounded
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, DFU, NormalGr], (instregex "LEDTR$")>;
def : InstRW<[WLat30, DFU2, NormalGr], (instregex "LDXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Load lengthened
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, DFU, NormalGr], (instregex "LDETR$")>;
def : InstRW<[WLat6, DFU4, GroupAlone], (instregex "LXDTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Convert from fixed / logical
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, FXU, DFU, GroupAlone], (instregex "CDFTR$")>;
def : InstRW<[WLat30, FXU, DFU, GroupAlone], (instregex "CDGTR(A)?$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat5, FXU, DFU4, GroupAlone2], (instregex "CXFTR(A)?$")>;
def : InstRW<[WLat30, FXU, DFU4, GroupAlone2], (instregex "CXGTR(A)?$")>;
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, FXU, DFU, GroupAlone], (instregex "CDL(F|G)TR$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat9, FXU, DFU4, GroupAlone2], (instregex "CXLFTR$")>;
def : InstRW<[WLat5, FXU, DFU4, GroupAlone2], (instregex "CXLGTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Convert to fixed / logical
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat11, WLat11, FXU, DFU, GroupAlone], (instregex "CFDTR(A)?$")>;
def : InstRW<[WLat30, WLat30, FXU, DFU, GroupAlone], (instregex "CGDTR(A)?$")>;
def : InstRW<[WLat7, WLat7, FXU, DFU2, GroupAlone], (instregex "CFXTR$")>;
def : InstRW<[WLat30, WLat30, FXU, DFU2, GroupAlone], (instregex "CGXTR(A)?$")>;
def : InstRW<[WLat11, WLat11, FXU, DFU, GroupAlone], (instregex "CL(F|G)DTR$")>;
def : InstRW<[WLat7, WLat7, FXU, DFU2, GroupAlone], (instregex "CL(F|G)XTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Convert from / to signed / unsigned packed
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat5, FXU, DFU, GroupAlone], (instregex "CD(S|U)TR$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat8, FXU2, DFU4, GroupAlone2], (instregex "CX(S|U)TR$")>;
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, FXU, DFU, GroupAlone], (instregex "C(S|U)DTR$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat12, FXU2, DFU4, GroupAlone2], (instregex "C(S|U)XTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Perform floating-point operation
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "PFPO$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
//===----------------------------------------------------------------------===//
// DFP: Unary arithmetic
//===----------------------------------------------------------------------===//
// Load FP integer
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat8, DFU, NormalGr], (instregex "FIDTR$")>;
def : InstRW<[WLat10, DFU4, GroupAlone], (instregex "FIXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Extract biased exponent
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, FXU, DFU, GroupAlone], (instregex "EEDTR$")>;
def : InstRW<[WLat8, FXU, DFU2, GroupAlone], (instregex "EEXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Extract significance
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7, FXU, DFU, GroupAlone], (instregex "ESDTR$")>;
def : InstRW<[WLat8, FXU, DFU2, GroupAlone], (instregex "ESXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
//===----------------------------------------------------------------------===//
// DFP: Binary arithmetic
//===----------------------------------------------------------------------===//
// Addition
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, WLat9, DFU, NormalGr], (instregex "ADTR(A)?$")>;
def : InstRW<[WLat30, WLat30, DFU4, GroupAlone], (instregex "AXTR(A)?$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Subtraction
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, WLat9, DFU, NormalGr], (instregex "SDTR(A)?$")>;
def : InstRW<[WLat30, WLat30, DFU4, GroupAlone], (instregex "SXTR(A)?$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Multiply
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, DFU, NormalGr], (instregex "MDTR(A)?$")>;
def : InstRW<[WLat30, DFU4, GroupAlone], (instregex "MXTR(A)?$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Division
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, DFU, NormalGr], (instregex "DDTR(A)?$")>;
def : InstRW<[WLat30, DFU4, GroupAlone], (instregex "DXTR(A)?$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Quantize
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat8, WLat8, DFU, NormalGr], (instregex "QADTR$")>;
def : InstRW<[WLat10, WLat10, DFU4, GroupAlone], (instregex "QAXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Reround
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat11, WLat11, FXU, DFU, GroupAlone], (instregex "RRDTR$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat30, WLat30, FXU, DFU4, GroupAlone2], (instregex "RRXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Shift significand left/right
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat7LSU, LSU, DFU, GroupAlone], (instregex "S(L|R)DT$")>;
def : InstRW<[WLat11LSU, LSU, DFU4, GroupAlone], (instregex "S(L|R)XT$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Insert biased exponent
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat5, FXU, DFU, GroupAlone], (instregex "IEDTR$")>;
[SystemZ] Improve handling of instructions which expand to several groups Some instructions expand to more than one decoder group. This has been hitherto ignored, but is handled with this patch. Review: Ulrich Weigand https://reviews.llvm.org/D50187 llvm-svn: 338849
2018-08-03 18:43:05 +08:00
def : InstRW<[WLat7, FXU, DFU4, GroupAlone2], (instregex "IEXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
//===----------------------------------------------------------------------===//
// DFP: Comparisons
//===----------------------------------------------------------------------===//
// Compare
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, DFU, NormalGr], (instregex "(K|C)DTR$")>;
def : InstRW<[WLat10, DFU2, NormalGr], (instregex "(K|C)XTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Compare biased exponent
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat4, DFU, NormalGr], (instregex "CEDTR$")>;
def : InstRW<[WLat5, DFU2, NormalGr], (instregex "CEXTR$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
// Test Data Class/Group
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat9, LSU, DFU, NormalGr], (instregex "TD(C|G)DT$")>;
def : InstRW<[WLat10, LSU, DFU, NormalGr], (instregex "TD(C|G)ET$")>;
def : InstRW<[WLat10, LSU, DFU2, NormalGr], (instregex "TD(C|G)XT$")>;
[SystemZ] Add decimal floating-point instructions This adds assembler / disassembler support for the decimal floating-point instructions. Since LLVM does not yet have support for decimal float types, these cannot be used for codegen at this point. llvm-svn: 304203
2017-05-30 18:15:16 +08:00
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
// -------------------------------- System ---------------------------------- //
//===----------------------------------------------------------------------===//
// System: Program-Status Word Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, WLat30, MCD], (instregex "EPSW$")>;
def : InstRW<[WLat30, MCD], (instregex "LPSW(E)?$")>;
def : InstRW<[WLat3, FXU, GroupAlone], (instregex "IPK$")>;
def : InstRW<[WLat1, LSU, EndGroup], (instregex "SPKA$")>;
def : InstRW<[WLat1, LSU, EndGroup], (instregex "SSM$")>;
def : InstRW<[WLat1, FXU, LSU, GroupAlone], (instregex "ST(N|O)SM$")>;
def : InstRW<[WLat3, FXU, NormalGr], (instregex "IAC$")>;
def : InstRW<[WLat1, LSU, EndGroup], (instregex "SAC(F)?$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Control Register Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat10, WLat10, LSU2, GroupAlone], (instregex "LCTL(G)?$")>;
def : InstRW<[WLat1, FXU5, LSU5, GroupAlone], (instregex "STCT(L|G)$")>;
def : InstRW<[LSULatency, LSU, NormalGr], (instregex "E(P|S)A(I)?R$")>;
def : InstRW<[WLat30, MCD], (instregex "SSA(I)?R$")>;
def : InstRW<[WLat30, MCD], (instregex "ESEA$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Prefix-Register Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "S(T)?PX$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Storage-Key and Real Memory Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "ISKE$")>;
def : InstRW<[WLat30, MCD], (instregex "IVSK$")>;
def : InstRW<[WLat30, MCD], (instregex "SSKE(Opt)?$")>;
def : InstRW<[WLat30, MCD], (instregex "RRB(E|M)$")>;
def : InstRW<[WLat30, MCD], (instregex "PFMF$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "TB$")>;
def : InstRW<[WLat30, MCD], (instregex "PGIN$")>;
def : InstRW<[WLat30, MCD], (instregex "PGOUT$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Dynamic-Address-Translation Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "IPTE(Opt)?(Opt)?$")>;
def : InstRW<[WLat30, MCD], (instregex "IDTE(Opt)?$")>;
def : InstRW<[WLat30, MCD], (instregex "PTLB$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "CSP(G)?$")>;
def : InstRW<[WLat30, WLat30, WLat30, MCD], (instregex "LPTEA$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "LRA(Y|G)?$")>;
def : InstRW<[WLat30, MCD], (instregex "STRAG$")>;
def : InstRW<[WLat30, MCD], (instregex "LURA(G)?$")>;
def : InstRW<[WLat30, MCD], (instregex "STUR(A|G)$")>;
def : InstRW<[WLat30, MCD], (instregex "TPROT$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Memory-move Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "MVC(K|P|S)$")>;
def : InstRW<[WLat30, MCD], (instregex "MVC(S|D)K$")>;
def : InstRW<[WLat30, MCD], (instregex "MVCOS$")>;
def : InstRW<[WLat30, MCD], (instregex "MVPG$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Address-Space Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "LASP$")>;
def : InstRW<[WLat1, LSU, GroupAlone], (instregex "PALB$")>;
def : InstRW<[WLat30, MCD], (instregex "PC$")>;
def : InstRW<[WLat30, MCD], (instregex "PR$")>;
def : InstRW<[WLat30, MCD], (instregex "PT(I)?$")>;
def : InstRW<[WLat30, MCD], (instregex "RP$")>;
def : InstRW<[WLat30, MCD], (instregex "BS(G|A)$")>;
def : InstRW<[WLat30, MCD], (instregex "TAR$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Linkage-Stack Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "BAKR$")>;
def : InstRW<[WLat30, MCD], (instregex "EREG(G)?$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "(E|M)STA$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Time-Related Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "PTFF$")>;
def : InstRW<[WLat30, MCD], (instregex "SCK$")>;
def : InstRW<[WLat30, MCD], (instregex "SCKPF$")>;
def : InstRW<[WLat30, MCD], (instregex "SCKC$")>;
def : InstRW<[WLat30, MCD], (instregex "SPT$")>;
def : InstRW<[WLat30, MCD], (instregex "STCK(F)?$")>;
def : InstRW<[WLat30, MCD], (instregex "STCKE$")>;
def : InstRW<[WLat30, MCD], (instregex "STCKC$")>;
def : InstRW<[WLat30, MCD], (instregex "STPT$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: CPU-Related Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "STAP$")>;
def : InstRW<[WLat30, MCD], (instregex "STIDP$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "STSI$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "STFL(E)?$")>;
def : InstRW<[WLat30, MCD], (instregex "ECAG$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "ECTG$")>;
def : InstRW<[WLat30, MCD], (instregex "PTF$")>;
def : InstRW<[WLat30, MCD], (instregex "PCKMO$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: Miscellaneous Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "SVC$")>;
def : InstRW<[WLat1, FXU, GroupAlone], (instregex "MC$")>;
def : InstRW<[WLat30, MCD], (instregex "DIAG$")>;
def : InstRW<[WLat30, MCD], (instregex "TRAC(E|G)$")>;
def : InstRW<[WLat30, MCD], (instregex "TRAP(2|4)$")>;
def : InstRW<[WLat30, MCD], (instregex "SIG(P|A)$")>;
def : InstRW<[WLat30, MCD], (instregex "SIE$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: CPU-Measurement Facility Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat1, FXU, NormalGr], (instregex "LPP$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "ECPGA$")>;
def : InstRW<[WLat30, WLat30, MCD], (instregex "E(C|P)CTR$")>;
def : InstRW<[WLat30, MCD], (instregex "LCCTL$")>;
def : InstRW<[WLat30, MCD], (instregex "L(P|S)CTL$")>;
def : InstRW<[WLat30, MCD], (instregex "Q(S|CTR)I$")>;
def : InstRW<[WLat30, MCD], (instregex "S(C|P)CTR$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
//===----------------------------------------------------------------------===//
// System: I/O Instructions
//===----------------------------------------------------------------------===//
[SystemZ] Reimplent SchedModel IssueWidth and WriteRes/ReadAdvance mappings. As a consequence of recent discussions (http://lists.llvm.org/pipermail/llvm-dev/2018-May/123164.html), this patch changes the SystemZ SchedModels so that the IssueWidth is 6, which is the decoder capacity, and NumMicroOps become the number of decoder slots needed per instruction. In addition, the SchedWrite latencies now match the MachineInstructions def-operand indexes, and ReadAdvances have been added on instructions with one register operand and one memory operand. Review: Ulrich Weigand https://reviews.llvm.org/D47008 llvm-svn: 337538
2018-07-20 17:40:43 +08:00
def : InstRW<[WLat30, MCD], (instregex "(C|H|R|X)SCH$")>;
def : InstRW<[WLat30, MCD], (instregex "(M|S|ST|T)SCH$")>;
def : InstRW<[WLat30, MCD], (instregex "RCHP$")>;
def : InstRW<[WLat30, MCD], (instregex "SCHM$")>;
def : InstRW<[WLat30, MCD], (instregex "STC(PS|RW)$")>;
def : InstRW<[WLat30, MCD], (instregex "TPI$")>;
def : InstRW<[WLat30, MCD], (instregex "SAL$")>;
[SystemZ] Add all remaining instructions This adds all remaining instructions that were still missing, mostly privileged and semi-privileged system-level instructions. These are provided for use with the assembler and disassembler only. This brings the LLVM assembler / disassembler to parity with the GNU binutils tools. llvm-svn: 306876
2017-07-01 04:43:40 +08:00
[SystemZ] Post-RA scheduler implementation Post-RA sched strategy and scheduling instruction annotations for z196, zEC12 and z13. This scheduler optimizes decoder grouping and balances processor resources (including side steering the FPd unit instructions). The SystemZHazardRecognizer keeps track of the scheduling state, which can be dumped with -debug-only=misched. Reviers: Ulrich Weigand, Andrew Trick. https://reviews.llvm.org/D17260 llvm-svn: 284704
2016-10-20 16:27:16 +08:00
}