From 2e47aafb02f3e46fc3e01799053e01835239151d Mon Sep 17 00:00:00 2001 From: Stefan Pintilie Date: Mon, 13 Feb 2023 09:59:23 -0500 Subject: [PATCH] [PowerPC] Fix float materialization patterns. Two of the float materialization patterns use the VSSRC regsiter class. This register class is not available before Power 8. The patterns will stay the same for Power 8 and up but must use the class F4RC for Power 7 and earlier. This patch fixes those patterns. Reviewed By: nemanjai, amyk, #powerpc Differential Revision: https://reviews.llvm.org/D142120 --- llvm/lib/Target/PowerPC/PPCInstrVSX.td | 27 ++- llvm/test/CodeGen/PowerPC/ppc-float-spill.ll | 270 +++++++++++++++++++++++++++ 2 files changed, 290 insertions(+), 7 deletions(-) create mode 100644 llvm/test/CodeGen/PowerPC/ppc-float-spill.ll diff --git a/llvm/lib/Target/PowerPC/PPCInstrVSX.td b/llvm/lib/Target/PowerPC/PPCInstrVSX.td index b62866e..5847504 100644 --- a/llvm/lib/Target/PowerPC/PPCInstrVSX.td +++ b/llvm/lib/Target/PowerPC/PPCInstrVSX.td @@ -126,6 +126,7 @@ def IsLittleEndian : Predicate<"Subtarget->isLittleEndian()">; def IsBigEndian : Predicate<"!Subtarget->isLittleEndian()">; def IsPPC64 : Predicate<"Subtarget->isPPC64()">; def HasOnlySwappingMemOps : Predicate<"!Subtarget->hasP9Vector()">; +def NoP8Vector : Predicate<"!Subtarget->hasP8Vector()">; def HasP8Vector : Predicate<"Subtarget->hasP8Vector()">; def HasDirectMove : Predicate<"Subtarget->hasDirectMove()">; def NoP9Vector : Predicate<"!Subtarget->hasP9Vector()">; @@ -2451,6 +2452,7 @@ def DblwdCmp { // [HasVSX, NoP9Vector, IsBigEndian] // [HasVSX, HasOnlySwappingMemOps] // [HasVSX, HasOnlySwappingMemOps, IsBigEndian] +// [HasVSX, NoP8Vector] // [HasVSX, HasP8Vector] // [HasVSX, HasP8Vector, IsBigEndian] // [HasVSX, HasP8Vector, IsBigEndian, IsPPC64] @@ -2504,17 +2506,10 @@ def : Pat<(v4i32 (or (and (vnot v4i32:$C), v4i32:$A), def : Pat<(f64 (fpimm0neg)), (f64 (XSNEGDP (XXLXORdpz)))>; -def : Pat<(f32 (fpimm0neg)), - (f32 (COPY_TO_REGCLASS (XSNEGDP (XXLXORdpz)), VSSRC))>; - def : Pat<(f64 (nzFPImmExactInti5:$A)), (COPY_TO_REGCLASS (XVCVSXWDP (COPY_TO_REGCLASS (VSPLTISW (getFPAs5BitExactInt fpimm:$A)), VSRC)), VSFRC)>; -def : Pat<(f32 (nzFPImmExactInti5:$A)), - (COPY_TO_REGCLASS (XVCVSXWDP (COPY_TO_REGCLASS - (VSPLTISW (getFPAs5BitExactInt fpimm:$A)), VSRC)), VSSRC)>; - // Additional fnmsub pattern for PPC specific ISD opcode def : Pat<(PPCfnmsub f64:$A, f64:$B, f64:$C), (XSNMSUBADP $C, $A, $B)>; @@ -3245,6 +3240,17 @@ let Predicates = [HasVSX, HasOnlySwappingMemOps, IsBigEndian] in { (SUBREG_TO_REG (i64 1), (XFLOADf64 ForceXForm:$src), sub_64)>; } // HasVSX, HasOnlySwappingMemOps, IsBigEndian +// Target before Power8 with VSX. +let Predicates = [HasVSX, NoP8Vector] in { +def : Pat<(f32 (fpimm0neg)), + (f32 (COPY_TO_REGCLASS (XSNEGDP (XXLXORdpz)), F4RC))>; + +def : Pat<(f32 (nzFPImmExactInti5:$A)), + (COPY_TO_REGCLASS (XVCVSXWDP (COPY_TO_REGCLASS + (VSPLTISW (getFPAs5BitExactInt fpimm:$A)), VSRC)), F4RC)>; + +} // HasVSX, NoP8Vector + // Any Power8 VSX subtarget. let Predicates = [HasVSX, HasP8Vector] in { def : Pat<(int_ppc_vsx_xxleqv v4i32:$A, v4i32:$B), @@ -3256,6 +3262,13 @@ def : Pat<(f32 (fpround (f64 (extloadf32 ForceXForm:$src)))), def : Pat<(f64 (any_fpextend f32:$src)), (COPY_TO_REGCLASS $src, VSFRC)>; +def : Pat<(f32 (fpimm0neg)), + (f32 (COPY_TO_REGCLASS (XSNEGDP (XXLXORdpz)), VSSRC))>; + +def : Pat<(f32 (nzFPImmExactInti5:$A)), + (COPY_TO_REGCLASS (XVCVSXWDP (COPY_TO_REGCLASS + (VSPLTISW (getFPAs5BitExactInt fpimm:$A)), VSRC)), VSSRC)>; + def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLT)), (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>; def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULT)), diff --git a/llvm/test/CodeGen/PowerPC/ppc-float-spill.ll b/llvm/test/CodeGen/PowerPC/ppc-float-spill.ll new file mode 100644 index 0000000..0de70b8 --- /dev/null +++ b/llvm/test/CodeGen/PowerPC/ppc-float-spill.ll @@ -0,0 +1,270 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \ +; RUN: -mcpu=pwr7 -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \ +; RUN: -O0 --data-sections < %s | FileCheck %s --check-prefix LE-LINUX-P7 +; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \ +; RUN: -mcpu=pwr8 -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \ +; RUN: -O0 --data-sections < %s | FileCheck %s --check-prefix LE-LINUX-P8 +; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu \ +; RUN: -mcpu=pwr7 -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \ +; RUN: -O0 --data-sections < %s | FileCheck %s --check-prefix BE-LINUX-P7 +; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu \ +; RUN: -mcpu=pwr8 -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \ +; RUN: -O0 --data-sections < %s | FileCheck %s --check-prefix BE-LINUX-P8 +; RUN: llc -verify-machineinstrs -mtriple=powerpc64-ibm-aix \ +; RUN: -mcpu=pwr7 -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \ +; RUN: -O0 --data-sections < %s | FileCheck %s --check-prefix AIX64-P7 +; RUN: llc -verify-machineinstrs -mtriple=powerpc64-ibm-aix \ +; RUN: -mcpu=pwr8 -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \ +; RUN: -O0 --data-sections < %s | FileCheck %s --check-prefix AIX64-P8 +; RUN: llc -verify-machineinstrs -mtriple=powerpc-ibm-aix \ +; RUN: -mcpu=pwr7 -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \ +; RUN: -O0 --data-sections < %s | FileCheck %s --check-prefix AIX32-P7 +; RUN: llc -verify-machineinstrs -mtriple=powerpc-ibm-aix \ +; RUN: -mcpu=pwr8 -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \ +; RUN: -O0 --data-sections < %s | FileCheck %s --check-prefix AIX32-P8 + +;; Note that the above run lines all use -O0 because this causes the spill that +;; exposed the issue in the first place. Higher opt levels will remove the spills +;; and as a result will not expose the issue. + +define dso_local void @caller() #0 { +; LE-LINUX-P7-LABEL: caller: +; LE-LINUX-P7: # %bb.0: # %entry +; LE-LINUX-P7-NEXT: mflr r0 +; LE-LINUX-P7-NEXT: stdu r1, -112(r1) +; LE-LINUX-P7-NEXT: std r0, 128(r1) +; LE-LINUX-P7-NEXT: addis r3, r2, .LCPI0_1@toc@ha +; LE-LINUX-P7-NEXT: lfs f1, .LCPI0_1@toc@l(r3) +; LE-LINUX-P7-NEXT: bl callee +; LE-LINUX-P7-NEXT: nop +; LE-LINUX-P7-NEXT: stfs f1, 100(r1) # 4-byte Folded Spill +; LE-LINUX-P7-NEXT: addis r3, r2, .LCPI0_1@toc@ha +; LE-LINUX-P7-NEXT: lfs f1, .LCPI0_1@toc@l(r3) +; LE-LINUX-P7-NEXT: bl callee +; LE-LINUX-P7-NEXT: nop +; LE-LINUX-P7-NEXT: fmr f2, f1 +; LE-LINUX-P7-NEXT: lfs f1, 100(r1) # 4-byte Folded Reload +; LE-LINUX-P7-NEXT: addis r3, r2, .LCPI0_0@toc@ha +; LE-LINUX-P7-NEXT: lfs f0, .LCPI0_0@toc@l(r3) +; LE-LINUX-P7-NEXT: fmuls f0, f0, f2 +; LE-LINUX-P7-NEXT: stfs f1, 104(r1) +; LE-LINUX-P7-NEXT: stfs f0, 108(r1) +; LE-LINUX-P7-NEXT: addi r1, r1, 112 +; LE-LINUX-P7-NEXT: ld r0, 16(r1) +; LE-LINUX-P7-NEXT: mtlr r0 +; LE-LINUX-P7-NEXT: blr +; +; LE-LINUX-P8-LABEL: caller: +; LE-LINUX-P8: # %bb.0: # %entry +; LE-LINUX-P8-NEXT: mflr r0 +; LE-LINUX-P8-NEXT: stdu r1, -64(r1) +; LE-LINUX-P8-NEXT: std r0, 80(r1) +; LE-LINUX-P8-NEXT: vspltisw v2, 1 +; LE-LINUX-P8-NEXT: xxlor vs0, v2, v2 +; LE-LINUX-P8-NEXT: xvcvsxwdp vs0, vs0 +; LE-LINUX-P8-NEXT: fmr f1, f0 +; LE-LINUX-P8-NEXT: li r3, 48 +; LE-LINUX-P8-NEXT: stxsspx f1, r1, r3 # 4-byte Folded Spill +; LE-LINUX-P8-NEXT: bl callee +; LE-LINUX-P8-NEXT: nop +; LE-LINUX-P8-NEXT: fmr f0, f1 +; LE-LINUX-P8-NEXT: li r3, 48 +; LE-LINUX-P8-NEXT: lxsspx f1, r1, r3 # 4-byte Folded Reload +; LE-LINUX-P8-NEXT: stfs f0, 52(r1) # 4-byte Folded Spill +; LE-LINUX-P8-NEXT: bl callee +; LE-LINUX-P8-NEXT: nop +; LE-LINUX-P8-NEXT: fmr f0, f1 +; LE-LINUX-P8-NEXT: lfs f1, 52(r1) # 4-byte Folded Reload +; LE-LINUX-P8-NEXT: xsaddsp f0, f0, f0 +; LE-LINUX-P8-NEXT: stfs f1, 56(r1) +; LE-LINUX-P8-NEXT: stfs f0, 60(r1) +; LE-LINUX-P8-NEXT: addi r1, r1, 64 +; LE-LINUX-P8-NEXT: ld r0, 16(r1) +; LE-LINUX-P8-NEXT: mtlr r0 +; LE-LINUX-P8-NEXT: blr +; +; BE-LINUX-P7-LABEL: caller: +; BE-LINUX-P7: # %bb.0: # %entry +; BE-LINUX-P7-NEXT: mflr r0 +; BE-LINUX-P7-NEXT: stdu r1, -128(r1) +; BE-LINUX-P7-NEXT: std r0, 144(r1) +; BE-LINUX-P7-NEXT: addis r3, r2, .LCPI0_1@toc@ha +; BE-LINUX-P7-NEXT: lfs f1, .LCPI0_1@toc@l(r3) +; BE-LINUX-P7-NEXT: bl callee +; BE-LINUX-P7-NEXT: nop +; BE-LINUX-P7-NEXT: stfs f1, 116(r1) # 4-byte Folded Spill +; BE-LINUX-P7-NEXT: addis r3, r2, .LCPI0_1@toc@ha +; BE-LINUX-P7-NEXT: lfs f1, .LCPI0_1@toc@l(r3) +; BE-LINUX-P7-NEXT: bl callee +; BE-LINUX-P7-NEXT: nop +; BE-LINUX-P7-NEXT: fmr f2, f1 +; BE-LINUX-P7-NEXT: lfs f1, 116(r1) # 4-byte Folded Reload +; BE-LINUX-P7-NEXT: addis r3, r2, .LCPI0_0@toc@ha +; BE-LINUX-P7-NEXT: lfs f0, .LCPI0_0@toc@l(r3) +; BE-LINUX-P7-NEXT: fmuls f0, f0, f2 +; BE-LINUX-P7-NEXT: stfs f1, 120(r1) +; BE-LINUX-P7-NEXT: stfs f0, 124(r1) +; BE-LINUX-P7-NEXT: addi r1, r1, 128 +; BE-LINUX-P7-NEXT: ld r0, 16(r1) +; BE-LINUX-P7-NEXT: mtlr r0 +; BE-LINUX-P7-NEXT: blr +; +; BE-LINUX-P8-LABEL: caller: +; BE-LINUX-P8: # %bb.0: # %entry +; BE-LINUX-P8-NEXT: mflr r0 +; BE-LINUX-P8-NEXT: stdu r1, -144(r1) +; BE-LINUX-P8-NEXT: std r0, 160(r1) +; BE-LINUX-P8-NEXT: vspltisw v2, 1 +; BE-LINUX-P8-NEXT: xxlor vs0, v2, v2 +; BE-LINUX-P8-NEXT: xvcvsxwdp vs0, vs0 +; BE-LINUX-P8-NEXT: fmr f1, f0 +; BE-LINUX-P8-NEXT: li r3, 128 +; BE-LINUX-P8-NEXT: stxsspx f1, r1, r3 # 4-byte Folded Spill +; BE-LINUX-P8-NEXT: bl callee +; BE-LINUX-P8-NEXT: nop +; BE-LINUX-P8-NEXT: fmr f0, f1 +; BE-LINUX-P8-NEXT: li r3, 128 +; BE-LINUX-P8-NEXT: lxsspx f1, r1, r3 # 4-byte Folded Reload +; BE-LINUX-P8-NEXT: stfs f0, 132(r1) # 4-byte Folded Spill +; BE-LINUX-P8-NEXT: bl callee +; BE-LINUX-P8-NEXT: nop +; BE-LINUX-P8-NEXT: fmr f0, f1 +; BE-LINUX-P8-NEXT: lfs f1, 132(r1) # 4-byte Folded Reload +; BE-LINUX-P8-NEXT: xsaddsp f0, f0, f0 +; BE-LINUX-P8-NEXT: stfs f1, 136(r1) +; BE-LINUX-P8-NEXT: stfs f0, 140(r1) +; BE-LINUX-P8-NEXT: addi r1, r1, 144 +; BE-LINUX-P8-NEXT: ld r0, 16(r1) +; BE-LINUX-P8-NEXT: mtlr r0 +; BE-LINUX-P8-NEXT: blr +; +; AIX64-P7-LABEL: caller: +; AIX64-P7: # %bb.0: # %entry +; AIX64-P7-NEXT: mflr r0 +; AIX64-P7-NEXT: stdu r1, -128(r1) +; AIX64-P7-NEXT: std r0, 144(r1) +; AIX64-P7-NEXT: vspltisw v2, 1 +; AIX64-P7-NEXT: xxlor vs0, v2, v2 +; AIX64-P7-NEXT: xvcvsxwdp vs0, vs0 +; AIX64-P7-NEXT: fmr f1, f0 +; AIX64-P7-NEXT: bl .callee[PR] +; AIX64-P7-NEXT: nop +; AIX64-P7-NEXT: stfs f1, 116(r1) # 4-byte Folded Spill +; AIX64-P7-NEXT: vspltisw v2, 1 +; AIX64-P7-NEXT: xxlor vs0, v2, v2 +; AIX64-P7-NEXT: xvcvsxwdp vs0, vs0 +; AIX64-P7-NEXT: fmr f1, f0 +; AIX64-P7-NEXT: bl .callee[PR] +; AIX64-P7-NEXT: nop +; AIX64-P7-NEXT: fmr f2, f1 +; AIX64-P7-NEXT: lfs f1, 116(r1) # 4-byte Folded Reload +; AIX64-P7-NEXT: ld r3, L..C0(r2) # %const.0 +; AIX64-P7-NEXT: lfs f0, 0(r3) +; AIX64-P7-NEXT: fmuls f0, f0, f2 +; AIX64-P7-NEXT: stfs f1, 120(r1) +; AIX64-P7-NEXT: stfs f0, 124(r1) +; AIX64-P7-NEXT: addi r1, r1, 128 +; AIX64-P7-NEXT: ld r0, 16(r1) +; AIX64-P7-NEXT: mtlr r0 +; AIX64-P7-NEXT: blr +; +; AIX64-P8-LABEL: caller: +; AIX64-P8: # %bb.0: # %entry +; AIX64-P8-NEXT: mflr r0 +; AIX64-P8-NEXT: stdu r1, -144(r1) +; AIX64-P8-NEXT: std r0, 160(r1) +; AIX64-P8-NEXT: vspltisw v2, 1 +; AIX64-P8-NEXT: xxlor vs0, v2, v2 +; AIX64-P8-NEXT: xvcvsxwdp vs0, vs0 +; AIX64-P8-NEXT: fmr f1, f0 +; AIX64-P8-NEXT: li r3, 128 +; AIX64-P8-NEXT: stxsspx f1, r1, r3 # 4-byte Folded Spill +; AIX64-P8-NEXT: bl .callee[PR] +; AIX64-P8-NEXT: nop +; AIX64-P8-NEXT: fmr f0, f1 +; AIX64-P8-NEXT: li r3, 128 +; AIX64-P8-NEXT: lxsspx f1, r1, r3 # 4-byte Folded Reload +; AIX64-P8-NEXT: stfs f0, 132(r1) # 4-byte Folded Spill +; AIX64-P8-NEXT: bl .callee[PR] +; AIX64-P8-NEXT: nop +; AIX64-P8-NEXT: fmr f0, f1 +; AIX64-P8-NEXT: lfs f1, 132(r1) # 4-byte Folded Reload +; AIX64-P8-NEXT: xsaddsp f0, f0, f0 +; AIX64-P8-NEXT: stfs f1, 136(r1) +; AIX64-P8-NEXT: stfs f0, 140(r1) +; AIX64-P8-NEXT: addi r1, r1, 144 +; AIX64-P8-NEXT: ld r0, 16(r1) +; AIX64-P8-NEXT: mtlr r0 +; AIX64-P8-NEXT: blr +; +; AIX32-P7-LABEL: caller: +; AIX32-P7: # %bb.0: # %entry +; AIX32-P7-NEXT: mflr r0 +; AIX32-P7-NEXT: stwu r1, -80(r1) +; AIX32-P7-NEXT: stw r0, 88(r1) +; AIX32-P7-NEXT: vspltisw v2, 1 +; AIX32-P7-NEXT: xxlor vs0, v2, v2 +; AIX32-P7-NEXT: xvcvsxwdp vs0, vs0 +; AIX32-P7-NEXT: fmr f1, f0 +; AIX32-P7-NEXT: stfs f1, 64(r1) # 4-byte Folded Spill +; AIX32-P7-NEXT: bl .callee[PR] +; AIX32-P7-NEXT: nop +; AIX32-P7-NEXT: fmr f0, f1 +; AIX32-P7-NEXT: lfs f1, 64(r1) # 4-byte Folded Reload +; AIX32-P7-NEXT: stfs f0, 68(r1) # 4-byte Folded Spill +; AIX32-P7-NEXT: bl .callee[PR] +; AIX32-P7-NEXT: nop +; AIX32-P7-NEXT: fmr f0, f1 +; AIX32-P7-NEXT: lfs f1, 68(r1) # 4-byte Folded Reload +; AIX32-P7-NEXT: fadds f0, f0, f0 +; AIX32-P7-NEXT: stfs f1, 72(r1) +; AIX32-P7-NEXT: stfs f0, 76(r1) +; AIX32-P7-NEXT: addi r1, r1, 80 +; AIX32-P7-NEXT: lwz r0, 8(r1) +; AIX32-P7-NEXT: mtlr r0 +; AIX32-P7-NEXT: blr +; +; AIX32-P8-LABEL: caller: +; AIX32-P8: # %bb.0: # %entry +; AIX32-P8-NEXT: mflr r0 +; AIX32-P8-NEXT: stwu r1, -80(r1) +; AIX32-P8-NEXT: stw r0, 88(r1) +; AIX32-P8-NEXT: vspltisw v2, 1 +; AIX32-P8-NEXT: xxlor vs0, v2, v2 +; AIX32-P8-NEXT: xvcvsxwdp vs0, vs0 +; AIX32-P8-NEXT: fmr f1, f0 +; AIX32-P8-NEXT: li r3, 64 +; AIX32-P8-NEXT: stxsspx f1, r1, r3 # 4-byte Folded Spill +; AIX32-P8-NEXT: bl .callee[PR] +; AIX32-P8-NEXT: nop +; AIX32-P8-NEXT: fmr f0, f1 +; AIX32-P8-NEXT: li r3, 64 +; AIX32-P8-NEXT: lxsspx f1, r1, r3 # 4-byte Folded Reload +; AIX32-P8-NEXT: stfs f0, 68(r1) # 4-byte Folded Spill +; AIX32-P8-NEXT: bl .callee[PR] +; AIX32-P8-NEXT: nop +; AIX32-P8-NEXT: fmr f0, f1 +; AIX32-P8-NEXT: lfs f1, 68(r1) # 4-byte Folded Reload +; AIX32-P8-NEXT: xsaddsp f0, f0, f0 +; AIX32-P8-NEXT: stfs f1, 72(r1) +; AIX32-P8-NEXT: stfs f0, 76(r1) +; AIX32-P8-NEXT: addi r1, r1, 80 +; AIX32-P8-NEXT: lwz r0, 8(r1) +; AIX32-P8-NEXT: mtlr r0 +; AIX32-P8-NEXT: blr +entry: + %com1 = alloca { float, float }, align 8 + %0 = call contract float @callee(float 1.000000e+00) + %1 = call contract float @callee(float 1.000000e+00) + %mult = fmul contract float 2.000000e+00, %1 + %addr0 = getelementptr inbounds { float, float }, ptr %com1, i32 0, i32 0 + store float %0, ptr %addr0, align 4 + %addr1 = getelementptr inbounds { float, float }, ptr %com1, i32 0, i32 1 + store float %mult, ptr %addr1, align 4 + ret void +} + +declare float @callee(float) #0 + +attributes #0 = { nounwind } -- 2.7.4