maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/PowerPC/fast-isel-conversion.ll

424 lines
9.1 KiB
LLVM
Raw Normal View History

[PowerPC] Temporarily disable VSX for PowerPC fast-isel tests Patch by Bill Seurer; some comment formatting changes by me. There are a few PowerPC test cases for FastISel support that currently fail with VSX support enabled. The temporary workaround under discussion in http://reviews.llvm.org/D5362 helps, but the tests still fail because they specify -fast-isel-abort, and the VSX workaround punts back to SelectionDAG. We have plans to fix FastISel permanently for VSX, but until that's in place these tests are preventing us from enabling VSX by default. Therefore we are adding -mattr=-vsx to these tests until the full support is ready. llvm-svn: 220172
2014-10-20 04:48:47 +08:00
; FIXME: FastISel currently returns false if it hits code that uses VSX
Change the fast-isel-abort option from bool to int to enable "levels" Summary: Currently fast-isel-abort will only abort for regular instructions, and just warn for function calls, terminators, function arguments. There is already fast-isel-abort-args but nothing for calls and terminators. This change turns the fast-isel-abort options into an integer option, so that multiple levels of strictness can be defined. This will help no being surprised when the "abort" option indeed does not abort, and enables the possibility to write test that verifies that no intrinsics are forgotten by fast-isel. Reviewers: resistor, echristo Subscribers: jfb, llvm-commits Differential Revision: http://reviews.llvm.org/D7941 From: Mehdi Amini <mehdi.amini@apple.com> llvm-svn: 230775
2015-02-28 02:32:11 +08:00
; registers and with -fast-isel-abort=1 turned on the test case will then fail.
[PowerPC] Temporarily disable VSX for PowerPC fast-isel tests Patch by Bill Seurer; some comment formatting changes by me. There are a few PowerPC test cases for FastISel support that currently fail with VSX support enabled. The temporary workaround under discussion in http://reviews.llvm.org/D5362 helps, but the tests still fail because they specify -fast-isel-abort, and the VSX workaround punts back to SelectionDAG. We have plans to fix FastISel permanently for VSX, but until that's in place these tests are preventing us from enabling VSX by default. Therefore we are adding -mattr=-vsx to these tests until the full support is ready. llvm-svn: 220172
2014-10-20 04:48:47 +08:00
; When fastisel better supports VSX fix up this test case.
;
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; RUN: llc < %s -O0 -verify-machineinstrs -fast-isel-abort=1 -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 -mattr=-vsx | FileCheck %s
; RUN: llc < %s -O0 -verify-machineinstrs -fast-isel-abort=1 -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr8 -mattr=-vsx | FileCheck %s
[PowerPC] Temporarily disable VSX for PowerPC fast-isel tests Patch by Bill Seurer; some comment formatting changes by me. There are a few PowerPC test cases for FastISel support that currently fail with VSX support enabled. The temporary workaround under discussion in http://reviews.llvm.org/D5362 helps, but the tests still fail because they specify -fast-isel-abort, and the VSX workaround punts back to SelectionDAG. We have plans to fix FastISel permanently for VSX, but until that's in place these tests are preventing us from enabling VSX by default. Therefore we are adding -mattr=-vsx to these tests until the full support is ready. llvm-svn: 220172
2014-10-20 04:48:47 +08:00
; RUN: llc < %s -O0 -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu -mcpu=970 -mattr=-vsx | FileCheck %s --check-prefix=PPC970
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
Change the fast-isel-abort option from bool to int to enable "levels" Summary: Currently fast-isel-abort will only abort for regular instructions, and just warn for function calls, terminators, function arguments. There is already fast-isel-abort-args but nothing for calls and terminators. This change turns the fast-isel-abort options into an integer option, so that multiple levels of strictness can be defined. This will help no being surprised when the "abort" option indeed does not abort, and enables the possibility to write test that verifies that no intrinsics are forgotten by fast-isel. Reviewers: resistor, echristo Subscribers: jfb, llvm-commits Differential Revision: http://reviews.llvm.org/D7941 From: Mehdi Amini <mehdi.amini@apple.com> llvm-svn: 230775
2015-02-28 02:32:11 +08:00
;; Tests for 970 don't use -fast-isel-abort=1 because we intentionally punt
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
;; to SelectionDAG in some cases.
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
; Test sitofp
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @sitofp_single_i64(i64 %a, float %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: sitofp_single_i64
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: sitofp_single_i64
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca float, align 4
%conv = sitofp i64 %a to float
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: std
; CHECK: lfd
; CHECK: fcfids
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
; PPC970: frsp
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store float %conv, float* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @sitofp_single_i32(i32 %a, float %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: sitofp_single_i32
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: sitofp_single_i32
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca float, align 4
%conv = sitofp i32 %a to float
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: std
; CHECK-NEXT: li
; CHECK-NEXT: lfiwax
; CHECK-NEXT: fcfids
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
; PPC970: frsp
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store float %conv, float* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @sitofp_single_i16(i16 %a, float %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: sitofp_single_i16
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: sitofp_single_i16
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca float, align 4
%conv = sitofp i16 %a to float
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: extsh
; CHECK: std
; CHECK: lfd
; CHECK: fcfids
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: extsh
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
; PPC970: frsp
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store float %conv, float* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @sitofp_single_i8(i8 %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: sitofp_single_i8
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: sitofp_single_i8
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca float, align 4
%conv = sitofp i8 %a to float
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: extsb
; CHECK: std
; CHECK: lfd
; CHECK: fcfids
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: extsb
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
; PPC970: frsp
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store float %conv, float* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @sitofp_double_i32(i32 %a, double %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: sitofp_double_i32
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: sitofp_double_i32
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca double, align 8
%conv = sitofp i32 %a to double
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: std
; CHECK-NOT: ori
; CHECK: li
; CHECK-NOT: ori
; CHECK: lfiwax
; CHECK: fcfid
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store double %conv, double* %b.addr, align 8
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @sitofp_double_i64(i64 %a, double %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: sitofp_double_i64
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: sitofp_double_i64
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca double, align 8
%conv = sitofp i64 %a to double
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: std
; CHECK: lfd
; CHECK: fcfid
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store double %conv, double* %b.addr, align 8
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @sitofp_double_i16(i16 %a, double %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: sitofp_double_i16
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: sitofp_double_i16
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca double, align 8
%conv = sitofp i16 %a to double
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: extsh
; CHECK: std
; CHECK: lfd
; CHECK: fcfid
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: extsh
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store double %conv, double* %b.addr, align 8
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @sitofp_double_i8(i8 %a, double %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: sitofp_double_i8
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: sitofp_double_i8
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca double, align 8
%conv = sitofp i8 %a to double
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: extsb
; CHECK: std
; CHECK: lfd
; CHECK: fcfid
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: extsb
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store double %conv, double* %b.addr, align 8
ret void
}
; Test uitofp
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @uitofp_single_i64(i64 %a, float %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: uitofp_single_i64
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: uitofp_single_i64
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca float, align 4
%conv = uitofp i64 %a to float
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: std
; CHECK: lfd
; CHECK: fcfidus
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970-NOT: fcfidus
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store float %conv, float* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @uitofp_single_i32(i32 %a, float %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: uitofp_single_i32
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: uitofp_single_i32
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca float, align 4
%conv = uitofp i32 %a to float
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: std
; CHECK-NOT: ori
; CHECK: li
; CHECK-NOT: ori
; CHECK: lfiwzx
; CHECK: fcfidus
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970-NOT: lfiwzx
; PPC970-NOT: fcfidus
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store float %conv, float* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @uitofp_single_i16(i16 %a, float %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: uitofp_single_i16
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: uitofp_single_i16
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca float, align 4
%conv = uitofp i16 %a to float
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: clrldi {{[0-9]+}}, {{[0-9]+}}, 48
; CHECK: std
; CHECK: lfd
; CHECK: fcfidus
[PowerPC] Enable printing instructions using aliases TableGen had been nicely generating code to print a number of instructions using shorter aliases (and PowerPC has plenty of short mnemonics), but we were not calling it. For some of the aliases we support in the parser, TableGen can't infer the "inverse" alias relationship, so there is still more to do. Thus, after some hours of updating test cases... llvm-svn: 235616
2015-04-24 02:30:38 +08:00
; PPC970: clrlwi {{[0-9]+}}, {{[0-9]+}}, 16
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
; PPC970: frsp
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store float %conv, float* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @uitofp_single_i8(i8 %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: uitofp_single_i8
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: uitofp_single_i8
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca float, align 4
%conv = uitofp i8 %a to float
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: clrldi {{[0-9]+}}, {{[0-9]+}}, 56
; CHECK: std
; CHECK: lfd
; CHECK: fcfidus
[PowerPC] Enable printing instructions using aliases TableGen had been nicely generating code to print a number of instructions using shorter aliases (and PowerPC has plenty of short mnemonics), but we were not calling it. For some of the aliases we support in the parser, TableGen can't infer the "inverse" alias relationship, so there is still more to do. Thus, after some hours of updating test cases... llvm-svn: 235616
2015-04-24 02:30:38 +08:00
; PPC970: clrlwi {{[0-9]+}}, {{[0-9]+}}, 24
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
; PPC970: frsp
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store float %conv, float* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @uitofp_double_i64(i64 %a, double %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: uitofp_double_i64
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: uitofp_double_i64
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca double, align 8
%conv = uitofp i64 %a to double
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: std
; CHECK: lfd
; CHECK: fcfidu
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970-NOT: fcfidu
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store double %conv, double* %b.addr, align 8
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @uitofp_double_i32(i32 %a, double %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: uitofp_double_i32
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: uitofp_double_i32
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca double, align 8
%conv = uitofp i32 %a to double
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: std
; CHECK-NEXT: li
; CHECK-NEXT: lfiwzx
; CHECK-NEXT: fcfidu
; CHECKLE: fcfidu
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970-NOT: lfiwzx
; PPC970-NOT: fcfidu
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store double %conv, double* %b.addr, align 8
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @uitofp_double_i16(i16 %a, double %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: uitofp_double_i16
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: uitofp_double_i16
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca double, align 8
%conv = uitofp i16 %a to double
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: clrldi {{[0-9]+}}, {{[0-9]+}}, 48
; CHECK: std
; CHECK: lfd
; CHECK: fcfidu
[PowerPC] Enable printing instructions using aliases TableGen had been nicely generating code to print a number of instructions using shorter aliases (and PowerPC has plenty of short mnemonics), but we were not calling it. For some of the aliases we support in the parser, TableGen can't infer the "inverse" alias relationship, so there is still more to do. Thus, after some hours of updating test cases... llvm-svn: 235616
2015-04-24 02:30:38 +08:00
; PPC970: clrlwi {{[0-9]+}}, {{[0-9]+}}, 16
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store double %conv, double* %b.addr, align 8
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @uitofp_double_i8(i8 %a, double %b) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: uitofp_double_i8
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: uitofp_double_i8
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca double, align 8
%conv = uitofp i8 %a to double
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: clrldi {{[0-9]+}}, {{[0-9]+}}, 56
; CHECK: std
; CHECK: lfd
; CHECK: fcfidu
[PowerPC] Enable printing instructions using aliases TableGen had been nicely generating code to print a number of instructions using shorter aliases (and PowerPC has plenty of short mnemonics), but we were not calling it. For some of the aliases we support in the parser, TableGen can't infer the "inverse" alias relationship, so there is still more to do. Thus, after some hours of updating test cases... llvm-svn: 235616
2015-04-24 02:30:38 +08:00
; PPC970: clrlwi {{[0-9]+}}, {{[0-9]+}}, 24
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: std
; PPC970: lfd
; PPC970: fcfid
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store double %conv, double* %b.addr, align 8
ret void
}
; Test fptosi
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @fptosi_float_i32(float %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fptosi_float_i32
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fptosi_float_i32
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca i32, align 4
%conv = fptosi float %a to i32
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fctiwz
; CHECK: stfd
; CHECK: lwa
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fctiwz
; PPC970: stfd
; PPC970: lwa
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store i32 %conv, i32* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @fptosi_float_i64(float %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fptosi_float_i64
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fptosi_float_i64
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca i64, align 4
%conv = fptosi float %a to i64
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fctidz
; CHECK: stfd
; CHECK: ld
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fctidz
; PPC970: stfd
; PPC970: ld
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store i64 %conv, i64* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @fptosi_double_i32(double %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fptosi_double_i32
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fptosi_double_i32
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca i32, align 8
%conv = fptosi double %a to i32
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fctiwz
; CHECK: stfd
; CHECK: lwa
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fctiwz
; PPC970: stfd
; PPC970: lwa
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store i32 %conv, i32* %b.addr, align 8
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @fptosi_double_i64(double %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fptosi_double_i64
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fptosi_double_i64
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca i64, align 8
%conv = fptosi double %a to i64
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fctidz
; CHECK: stfd
; CHECK: ld
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fctidz
; PPC970: stfd
; PPC970: ld
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store i64 %conv, i64* %b.addr, align 8
ret void
}
; Test fptoui
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @fptoui_float_i32(float %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fptoui_float_i32
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fptoui_float_i32
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca i32, align 4
%conv = fptoui float %a to i32
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fctiwuz
; CHECK: stfd
; CHECK: lwz
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fctidz
; PPC970: stfd
; PPC970: lwz
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store i32 %conv, i32* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @fptoui_float_i64(float %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fptoui_float_i64
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fptoui_float_i64
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca i64, align 4
%conv = fptoui float %a to i64
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fctiduz
; CHECK: stfd
; CHECK: ld
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970-NOT: fctiduz
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store i64 %conv, i64* %b.addr, align 4
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @fptoui_double_i32(double %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fptoui_double_i32
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fptoui_double_i32
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca i32, align 8
%conv = fptoui double %a to i32
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fctiwuz
; CHECK: stfd
; CHECK: lwz
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fctidz
; PPC970: stfd
; PPC970: lwz
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store i32 %conv, i32* %b.addr, align 8
ret void
}
Clean up function attributes on PPC fast-isel tests. llvm-svn: 243079
2015-07-24 09:07:50 +08:00
define void @fptoui_double_i64(double %a) nounwind {
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
entry:
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fptoui_double_i64
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970: fptoui_double_i64
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
%b.addr = alloca i64, align 8
%conv = fptoui double %a to i64
[PowerPC] Utilize DQ-Form instructions for spill/restore and fix FrameIndex elimination to only use `lis/addi` if necessary. Currently we produce a bunch of unnecessary code when emitting the prologue/epilogue for spills/restores. Namely, if the load from stack slot/store to stack slot instruction is an X-Form instruction, we will always produce an LIS/ORI sequence for the stack offset. Furthermore, we have not exploited the P9 vector D-Form loads/stores for this purpose. This patch address both issues. Specifying the D-Form load as the instruction to use for stack spills/reloads should be safe because: 1. The stack should be aligned according to the ABI 2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will check for the offset being a multiple of 16 and will convert it to an X-Form instruction if it isn't. Differential Revision : https://reviews.llvm.org/D38758 llvm-svn: 315500
2017-10-12 04:20:58 +08:00
; CHECK: fctiduz
; CHECK: stfd
; CHECK: ld
[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction) PR20071 identifies a problem in PowerPC's fast-isel implementation for floating-point conversion to integer. The fctiduz instruction was added in Power ISA 2.06 (i.e., Power7 and later). However, this instruction is being generated regardless of which 64-bit PowerPC target is selected. The intent is for fast-isel to punt to DAG selection when this instruction is not available. This patch implements that change. For testing purposes, the existing fast-isel-conversion.ll test adds a RUN line for -mcpu=970 and tests for the expected code generation. Additionally, the existing test fast-isel-conversion-p5.ll was found to be incorrectly expecting the unavailable instruction to be generated. I've removed these test variants since we have adequate coverage in fast-isel-conversion.ll. llvm-svn: 211627
2014-06-25 04:05:18 +08:00
; PPC970-NOT: fctiduz
[PowerPC] Add handling for conversions to fast-isel. Yet another chunk of fast-isel code. This one handles various conversions involving floating-point. (It also includes some miscellaneous handling throughout the back end for LWA_32 and LWAX_32 that should have been part of the load-store patch.) llvm-svn: 189677
2013-08-30 23:18:11 +08:00
store i64 %conv, i64* %b.addr, align 8
ret void
}