OSchip
/
llvm-project
mirror of https://github.com/THU-DSP-LAB/llvm-project.git
13
0
Fork 6
Code Issues Packages Projects Releases Wiki Activity

[X86] Make a bunch of merge masked binops commutable for loading folding. 

This primarily affects add/fadd/mul/fmul/and/or/xor/pmuludq/pmuldq/max/min/fmaxc/fminc/pmaddwd/pavg.

We already commuted the unmasked and zero masked versions.

I've added 512-bit stack folding tests for most of the instructions
affected. I've tested needing commuting and not commuting across
unmasked, merged masked, and zero masked. The 128/256 bit instructions
should behave similarly.

llvm-svn: 362746
This commit is contained in:
Craig Topper 2019-06-06 21:00:04 +00:00
parent 980d3645df
commit f320f26716
3 changed files with 2488 additions and 66 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
llvm/lib/Target/X86/X86InstrAVX512.td
View File

@ -277,10 +277,9 @@ multiclass AVX512_maskable<bits<8> O, Format F, X86VectorVTInfo _,
multiclass AVX512_maskable_scalar<bits<8> O, Format F, X86VectorVTInfo _,
dag Outs, dag Ins, string OpcodeStr,
string AttSrcAsm, string IntelSrcAsm,
dag RHS,
bit IsCommutable = 0> :
dag RHS> :
AVX512_maskable<O, F, _, Outs, Ins, OpcodeStr, AttSrcAsm, IntelSrcAsm,
RHS, IsCommutable, 0, IsCommutable, X86selects>;
RHS, 0, 0, 0, X86selects>;
// Similar to AVX512_maskable but in this case one of the source operands
// ($src1) is already tied to $dst so we just use that for the preserved
@ -4602,7 +4601,7 @@ multiclass avx512_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
(ins _.RC:$src1, _.RC:$src2), OpcodeStr,
"$src2, $src1", "$src1, $src2",
(_.VT (OpNode _.RC:$src1, _.RC:$src2)),
IsCommutable>, AVX512BIBase, EVEX_4V,
IsCommutable, IsCommutable>, AVX512BIBase, EVEX_4V,
Sched<[sched]>;
defm rm : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
@ -4838,7 +4837,7 @@ multiclass avx512_packs_rm<bits<8> opc, string OpcodeStr,
(_Dst.VT (OpNode
(_Src.VT _Src.RC:$src1),
(_Src.VT _Src.RC:$src2))),
IsCommutable>,
IsCommutable, IsCommutable>,
EVEX_CD8<_Src.EltSize, CD8VF>, EVEX_4V, Sched<[sched]>;
defm rm : AVX512_maskable<opc, MRMSrcMem, _Dst, (outs _Dst.RC:$dst),
(ins _Src.RC:$src1, _Src.MemOp:$src2), OpcodeStr,
@ -5531,13 +5530,13 @@ defm VMAXCSDZ : avx512_comutable_binop_s<0x5F, "vmaxsd", f64x_info, X86fmaxc,
multiclass avx512_fp_packed<bits<8> opc, string OpcodeStr, SDPatternOperator OpNode,
X86VectorVTInfo _, X86FoldableSchedWrite sched,
bit IsCommutable,
bit IsKZCommutable = IsCommutable> {
bit IsKCommutable = IsCommutable> {
let ExeDomain = _.ExeDomain, hasSideEffects = 0 in {
defm rr: AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
(ins _.RC:$src1, _.RC:$src2), OpcodeStr##_.Suffix,
"$src2, $src1", "$src1, $src2",
(_.VT (OpNode _.RC:$src1, _.RC:$src2)), IsCommutable, 0,
IsKZCommutable>,
(_.VT (OpNode _.RC:$src1, _.RC:$src2)), IsCommutable,
IsKCommutable, IsKCommutable>,
EVEX_4V, Sched<[sched]>;
let mayLoad = 1 in {
defm rm: AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),

212
llvm/test/CodeGen/X86/stack-folding-fp-avx512.ll
View File

@ -16,6 +16,28 @@ define <8 x double> @stack_fold_addpd_zmm(<8 x double> %a0, <8 x double> %a1) {
ret <8 x double> %2
}
define <8 x double> @stack_fold_addpd_zmm_k(<8 x double> %a0, <8 x double> %a1, i8 %mask, <8 x double>* %passthru) {
;CHECK-LABEL: stack_fold_addpd_zmm_k:
;CHECK: vaddpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fadd <8 x double> %a0, %a1
%3 = bitcast i8 %mask to <8 x i1>
%4 = load <8 x double>, <8 x double>* %passthru
%5 = select <8 x i1> %3, <8 x double> %2, <8 x double> %4
ret <8 x double> %5
}
define <8 x double> @stack_fold_addpd_zmm_k_commuted(<8 x double> %a0, <8 x double> %a1, i8 %mask, <8 x double>* %passthru) {
;CHECK-LABEL: stack_fold_addpd_zmm_k_commuted:
;CHECK: vaddpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fadd <8 x double> %a1, %a0
%3 = bitcast i8 %mask to <8 x i1>
%4 = load <8 x double>, <8 x double>* %passthru
%5 = select <8 x i1> %3, <8 x double> %2, <8 x double> %4
ret <8 x double> %5
}
define <8 x double> @stack_fold_addpd_zmm_kz(<8 x double> %a0, <8 x double> %a1, i8 %mask) {
;CHECK-LABEL: stack_fold_addpd_zmm_kz
;CHECK: vaddpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {z} {{.*#+}} 64-byte Folded Reload
@ -34,6 +56,28 @@ define <16 x float> @stack_fold_addps_zmm(<16 x float> %a0, <16 x float> %a1) {
ret <16 x float> %2
}
define <16 x float> @stack_fold_addps_zmm_k(<16 x float> %a0, <16 x float> %a1, i16 %mask, <16 x float>* %passthru) {
;CHECK-LABEL: stack_fold_addps_zmm_k:
;CHECK: vaddps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fadd <16 x float> %a0, %a1
%3 = bitcast i16 %mask to <16 x i1>
%4 = load <16 x float>, <16 x float>* %passthru
%5 = select <16 x i1> %3, <16 x float> %2, <16 x float> %4
ret <16 x float> %5
}
define <16 x float> @stack_fold_addps_zmm_k_commuted(<16 x float> %a0, <16 x float> %a1, i16 %mask, <16 x float>* %passthru) {
;CHECK-LABEL: stack_fold_addps_zmm_k_commuted:
;CHECK: vaddps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fadd <16 x float> %a1, %a0
%3 = bitcast i16 %mask to <16 x i1>
%4 = load <16 x float>, <16 x float>* %passthru
%5 = select <16 x i1> %3, <16 x float> %2, <16 x float> %4
ret <16 x float> %5
}
define <16 x float> @stack_fold_addps_zmm_kz(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
;CHECK-LABEL: stack_fold_addps_zmm_kz
;CHECK: vaddps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {z} {{.*#+}} 64-byte Folded Reload
@ -311,6 +355,28 @@ define <8 x double> @stack_fold_maxpd_zmm_commutable(<8 x double> %a0, <8 x doub
ret <8 x double> %2
}
define <8 x double> @stack_fold_maxpd_zmm_commutable_k(<8 x double> %a0, <8 x double> %a1, i8 %mask, <8 x double>* %passthru) #1 {
;CHECK-LABEL: stack_fold_maxpd_zmm_commutable_k:
;CHECK: vmaxpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = call <8 x double> @llvm.x86.avx512.max.pd.512(<8 x double> %a0, <8 x double> %a1, i32 4)
%3 = bitcast i8 %mask to <8 x i1>
%4 = load <8 x double>, <8 x double>* %passthru
%5 = select <8 x i1> %3, <8 x double> %2, <8 x double> %4
ret <8 x double> %5
}
define <8 x double> @stack_fold_maxpd_zmm_commutable_k_commuted(<8 x double> %a0, <8 x double> %a1, i8 %mask, <8 x double>* %passthru) #1 {
;CHECK-LABEL: stack_fold_maxpd_zmm_commutable_k_commuted:
;CHECK: vmaxpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = call <8 x double> @llvm.x86.avx512.max.pd.512(<8 x double> %a1, <8 x double> %a0, i32 4)
%3 = bitcast i8 %mask to <8 x i1>
%4 = load <8 x double>, <8 x double>* %passthru
%5 = select <8 x i1> %3, <8 x double> %2, <8 x double> %4
ret <8 x double> %5
}
define <8 x double> @stack_fold_maxpd_zmm_commutable_kz(<8 x double> %a0, <8 x double> %a1, i8 %mask) #1 {
;CHECK-LABEL: stack_fold_maxpd_zmm_commutable_kz
;CHECK: vmaxpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {z} {{.*#+}} 64-byte Folded Reload
@ -338,6 +404,28 @@ define <16 x float> @stack_fold_maxps_zmm_commutable(<16 x float> %a0, <16 x flo
ret <16 x float> %2
}
define <16 x float> @stack_fold_maxps_zmm_commutable_k(<16 x float> %a0, <16 x float> %a1, i16 %mask, <16 x float>* %passthru) #1 {
;CHECK-LABEL: stack_fold_maxps_zmm_commutable_k:
;CHECK: vmaxps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = call <16 x float> @llvm.x86.avx512.max.ps.512(<16 x float> %a0, <16 x float> %a1, i32 4)
%3 = bitcast i16 %mask to <16 x i1>
%4 = load <16 x float>, <16 x float>* %passthru
%5 = select <16 x i1> %3, <16 x float> %2, <16 x float> %4
ret <16 x float> %5
}
define <16 x float> @stack_fold_maxps_zmm_commutable_k_commuted(<16 x float> %a0, <16 x float> %a1, i16 %mask, <16 x float>* %passthru) #1 {
;CHECK-LABEL: stack_fold_maxps_zmm_commutable_k_commuted:
;CHECK: vmaxps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = call <16 x float> @llvm.x86.avx512.max.ps.512(<16 x float> %a1, <16 x float> %a0, i32 4)
%3 = bitcast i16 %mask to <16 x i1>
%4 = load <16 x float>, <16 x float>* %passthru
%5 = select <16 x i1> %3, <16 x float> %2, <16 x float> %4
ret <16 x float> %5
}
define <16 x float> @stack_fold_maxps_zmm_commutable_kz(<16 x float> %a0, <16 x float> %a1, i16 %mask) #1 {
;CHECK-LABEL: stack_fold_maxps_zmm_commutable_kz
;CHECK: vmaxps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {z} {{.*#+}} 64-byte Folded Reload
@ -365,6 +453,28 @@ define <8 x double> @stack_fold_minpd_zmm_commutable(<8 x double> %a0, <8 x doub
ret <8 x double> %2
}
define <8 x double> @stack_fold_minpd_zmm_commutable_k(<8 x double> %a0, <8 x double> %a1, i8 %mask, <8 x double>* %passthru) #1 {
;CHECK-LABEL: stack_fold_minpd_zmm_commutable_k:
;CHECK: vminpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = call <8 x double> @llvm.x86.avx512.min.pd.512(<8 x double> %a0, <8 x double> %a1, i32 4)
%3 = bitcast i8 %mask to <8 x i1>
%4 = load <8 x double>, <8 x double>* %passthru
%5 = select <8 x i1> %3, <8 x double> %2, <8 x double> %4
ret <8 x double> %5
}
define <8 x double> @stack_fold_minpd_zmm_commutable_k_commuted(<8 x double> %a0, <8 x double> %a1, i8 %mask, <8 x double>* %passthru) #1 {
;CHECK-LABEL: stack_fold_minpd_zmm_commutable_k_commuted:
;CHECK: vminpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = call <8 x double> @llvm.x86.avx512.min.pd.512(<8 x double> %a1, <8 x double> %a0, i32 4)
%3 = bitcast i8 %mask to <8 x i1>
%4 = load <8 x double>, <8 x double>* %passthru
%5 = select <8 x i1> %3, <8 x double> %2, <8 x double> %4
ret <8 x double> %5
}
define <8 x double> @stack_fold_minpd_zmm_commutable_kz(<8 x double> %a0, <8 x double> %a1, i8 %mask) #1 {
;CHECK-LABEL: stack_fold_minpd_zmm_commutable_kz
;CHECK: vminpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {z} {{.*#+}} 64-byte Folded Reload
@ -392,6 +502,28 @@ define <16 x float> @stack_fold_minps_zmm_commutable(<16 x float> %a0, <16 x flo
ret <16 x float> %2
}
define <16 x float> @stack_fold_minps_zmm_commutable_k(<16 x float> %a0, <16 x float> %a1, i16 %mask, <16 x float>* %passthru) #1 {
;CHECK-LABEL: stack_fold_minps_zmm_commutable_k:
;CHECK: vminps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = call <16 x float> @llvm.x86.avx512.min.ps.512(<16 x float> %a0, <16 x float> %a1, i32 4)
%3 = bitcast i16 %mask to <16 x i1>
%4 = load <16 x float>, <16 x float>* %passthru
%5 = select <16 x i1> %3, <16 x float> %2, <16 x float> %4
ret <16 x float> %5
}
define <16 x float> @stack_fold_minps_zmm_commutable_k_commuted(<16 x float> %a0, <16 x float> %a1, i16 %mask, <16 x float>* %passthru) #1 {
;CHECK-LABEL: stack_fold_minps_zmm_commutable_k_commuted:
;CHECK: vminps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = call <16 x float> @llvm.x86.avx512.min.ps.512(<16 x float> %a1, <16 x float> %a0, i32 4)
%3 = bitcast i16 %mask to <16 x i1>
%4 = load <16 x float>, <16 x float>* %passthru
%5 = select <16 x i1> %3, <16 x float> %2, <16 x float> %4
ret <16 x float> %5
}
define <16 x float> @stack_fold_minps_zmm_commutable_kz(<16 x float> %a0, <16 x float> %a1, i16 %mask) #1 {
;CHECK-LABEL: stack_fold_minps_zmm_commutable_kz
;CHECK: vminps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {z} {{.*#+}} 64-byte Folded Reload
@ -402,6 +534,86 @@ define <16 x float> @stack_fold_minps_zmm_commutable_kz(<16 x float> %a0, <16 x
ret <16 x float> %4
}
define <8 x double> @stack_fold_mulpd_zmm(<8 x double> %a0, <8 x double> %a1) {
;CHECK-LABEL: stack_fold_mulpd_zmm
;CHECK: vmulpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fmul <8 x double> %a0, %a1
ret <8 x double> %2
}
define <8 x double> @stack_fold_mulpd_zmm_k(<8 x double> %a0, <8 x double> %a1, i8 %mask, <8 x double>* %passthru) {
;CHECK-LABEL: stack_fold_mulpd_zmm_k:
;CHECK: vmulpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fmul <8 x double> %a0, %a1
%3 = bitcast i8 %mask to <8 x i1>
%4 = load <8 x double>, <8 x double>* %passthru
%5 = select <8 x i1> %3, <8 x double> %2, <8 x double> %4
ret <8 x double> %5
}
define <8 x double> @stack_fold_mulpd_zmm_k_commuted(<8 x double> %a0, <8 x double> %a1, i8 %mask, <8 x double>* %passthru) {
;CHECK-LABEL: stack_fold_mulpd_zmm_k_commuted:
;CHECK: vmulpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fmul <8 x double> %a1, %a0
%3 = bitcast i8 %mask to <8 x i1>
%4 = load <8 x double>, <8 x double>* %passthru
%5 = select <8 x i1> %3, <8 x double> %2, <8 x double> %4
ret <8 x double> %5
}
define <8 x double> @stack_fold_mulpd_zmm_kz(<8 x double> %a0, <8 x double> %a1, i8 %mask) {
;CHECK-LABEL: stack_fold_mulpd_zmm_kz
;CHECK: vmulpd {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {z} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fmul <8 x double> %a1, %a0
%3 = bitcast i8 %mask to <8 x i1>
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> zeroinitializer
ret <8 x double> %4
}
define <16 x float> @stack_fold_mulps_zmm(<16 x float> %a0, <16 x float> %a1) {
;CHECK-LABEL: stack_fold_mulps_zmm
;CHECK: vmulps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fmul <16 x float> %a0, %a1
ret <16 x float> %2
}
define <16 x float> @stack_fold_mulps_zmm_k(<16 x float> %a0, <16 x float> %a1, i16 %mask, <16 x float>* %passthru) {
;CHECK-LABEL: stack_fold_mulps_zmm_k:
;CHECK: vmulps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fmul <16 x float> %a0, %a1
%3 = bitcast i16 %mask to <16 x i1>
%4 = load <16 x float>, <16 x float>* %passthru
%5 = select <16 x i1> %3, <16 x float> %2, <16 x float> %4
ret <16 x float> %5
}
define <16 x float> @stack_fold_mulps_zmm_k_commuted(<16 x float> %a0, <16 x float> %a1, i16 %mask, <16 x float>* %passthru) {
;CHECK-LABEL: stack_fold_mulps_zmm_k_commuted:
;CHECK: vmulps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fmul <16 x float> %a1, %a0
%3 = bitcast i16 %mask to <16 x i1>
%4 = load <16 x float>, <16 x float>* %passthru
%5 = select <16 x i1> %3, <16 x float> %2, <16 x float> %4
ret <16 x float> %5
}
define <16 x float> @stack_fold_mulps_zmm_kz(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
;CHECK-LABEL: stack_fold_mulps_zmm_kz
;CHECK: vmulps {{-?[0-9]*}}(%rsp), {{%zmm[0-9][0-9]*}}, {{%zmm[0-9][0-9]*}} {{{%k[1-7]}}} {z} {{.*#+}} 64-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
%2 = fmul <16 x float> %a1, %a0
%3 = bitcast i16 %mask to <16 x i1>
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> zeroinitializer
ret <16 x float> %4
}
define double @stack_fold_mulsd(double %a0, double %a1) {
;CHECK-LABEL: stack_fold_mulsd
;CHECK: vmulsd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}}, {{%xmm[0-9][0-9]*}} {{.*#+}} 8-byte Folded Reload

2327
llvm/test/CodeGen/X86/stack-folding-int-avx512.ll
View File

File diff suppressed because it is too large Load Diff