386 changes: 193 additions & 193 deletions llvm/test/CodeGen/X86/avx-schedule.ll
View file

Large diffs are not rendered by default.

66 changes: 33 additions & 33 deletions llvm/test/CodeGen/X86/avx2-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,9 +15,9 @@ define <32 x i8> @test_pabsb(<32 x i8> %a0, <32 x i8> *%a1) {
; HASWELL-LABEL: test_pabsb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsb %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpabsb (%rdi), %ymm1 # sched: [5:0.50]
; HASWELL-NEXT: vpabsb (%rdi), %ymm1 # sched: [1:0.50]
; HASWELL-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pabsb:
; ZNVER1: # BB#0:
Expand All @@ -44,9 +44,9 @@ define <8 x i32> @test_pabsd(<8 x i32> %a0, <8 x i32> *%a1) {
; HASWELL-LABEL: test_pabsd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsd %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpabsd (%rdi), %ymm1 # sched: [5:0.50]
; HASWELL-NEXT: vpabsd (%rdi), %ymm1 # sched: [1:0.50]
; HASWELL-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pabsd:
; ZNVER1: # BB#0:
Expand All @@ -73,9 +73,9 @@ define <16 x i16> @test_pabsw(<16 x i16> %a0, <16 x i16> *%a1) {
; HASWELL-LABEL: test_pabsw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsw %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpabsw (%rdi), %ymm1 # sched: [5:0.50]
; HASWELL-NEXT: vpabsw (%rdi), %ymm1 # sched: [1:0.50]
; HASWELL-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pabsw:
; ZNVER1: # BB#0:
Expand All @@ -101,8 +101,8 @@ define <32 x i8> @test_paddb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> *%a2) {
; HASWELL-LABEL: test_paddb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpaddb %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpaddb (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpaddb (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_paddb:
; ZNVER1: # BB#0:
Expand All @@ -125,8 +125,8 @@ define <8 x i32> @test_paddd(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
; HASWELL-LABEL: test_paddd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpaddd %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpaddd (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpaddd (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_paddd:
; ZNVER1: # BB#0:
Expand All @@ -149,8 +149,8 @@ define <4 x i64> @test_paddq(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
; HASWELL-LABEL: test_paddq:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpaddq (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpaddq (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_paddq:
; ZNVER1: # BB#0:
Expand All @@ -173,8 +173,8 @@ define <16 x i16> @test_paddw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2) {
; HASWELL-LABEL: test_paddw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpaddw %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpaddw (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpaddw (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_paddw:
; ZNVER1: # BB#0:
Expand All @@ -198,9 +198,9 @@ define <4 x i64> @test_pand(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
; HASWELL-LABEL: test_pand:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpand %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; HASWELL-NEXT: vpand (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: vpand (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pand:
; ZNVER1: # BB#0:
Expand All @@ -226,9 +226,9 @@ define <4 x i64> @test_pandn(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
; HASWELL-LABEL: test_pandn:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpandn %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; HASWELL-NEXT: vpandn (%rdi), %ymm0, %ymm1 # sched: [5:0.50]
; HASWELL-NEXT: vpandn (%rdi), %ymm0, %ymm1 # sched: [1:0.50]
; HASWELL-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pandn:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -256,7 +256,7 @@ define <8 x i32> @test_pmulld(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
; HASWELL: # BB#0:
; HASWELL-NEXT: vpmulld %ymm1, %ymm0, %ymm0 # sched: [10:2.00]
; HASWELL-NEXT: vpmulld (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pmulld:
; ZNVER1: # BB#0:
Expand All @@ -279,8 +279,8 @@ define <16 x i16> @test_pmullw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2)
; HASWELL-LABEL: test_pmullw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpmullw %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
; HASWELL-NEXT: vpmullw (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpmullw (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pmullw:
; ZNVER1: # BB#0:
Expand All @@ -304,9 +304,9 @@ define <4 x i64> @test_por(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
; HASWELL-LABEL: test_por:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpor %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; HASWELL-NEXT: vpor (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: vpor (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_por:
; ZNVER1: # BB#0:
Expand All @@ -331,8 +331,8 @@ define <32 x i8> @test_psubb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> *%a2) {
; HASWELL-LABEL: test_psubb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsubb %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpsubb (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpsubb (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_psubb:
; ZNVER1: # BB#0:
Expand All @@ -355,8 +355,8 @@ define <8 x i32> @test_psubd(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> *%a2) {
; HASWELL-LABEL: test_psubd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsubd %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpsubd (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpsubd (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_psubd:
; ZNVER1: # BB#0:
Expand All @@ -379,8 +379,8 @@ define <4 x i64> @test_psubq(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
; HASWELL-LABEL: test_psubq:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsubq %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpsubq (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpsubq (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_psubq:
; ZNVER1: # BB#0:
Expand All @@ -403,8 +403,8 @@ define <16 x i16> @test_psubw(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> *%a2) {
; HASWELL-LABEL: test_psubw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsubw %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpsubw (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpsubw (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_psubw:
; ZNVER1: # BB#0:
Expand All @@ -428,9 +428,9 @@ define <4 x i64> @test_pxor(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> *%a2) {
; HASWELL-LABEL: test_pxor:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpxor %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; HASWELL-NEXT: vpxor (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; HASWELL-NEXT: vpxor (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: vpaddq %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pxor:
; ZNVER1: # BB#0:
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/X86/avx512-cmp.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -126,11 +126,11 @@ entry:
define i32 @test8(i32 %a1, i32 %a2, i32 %a3) {
; ALL-LABEL: test8:
; ALL: ## BB#0:
; ALL-NEXT: notl %edi
; ALL-NEXT: xorl $-2147483648, %esi ## imm = 0x80000000
; ALL-NEXT: testl %edx, %edx
; ALL-NEXT: movl $1, %eax
; ALL-NEXT: cmovel %eax, %edx
; ALL-NEXT: notl %edi
; ALL-NEXT: orl %edi, %esi
; ALL-NEXT: cmovnel %edx, %eax
; ALL-NEXT: retq
Expand Down
54 changes: 41 additions & 13 deletions llvm/test/CodeGen/X86/avx512-cvt.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1530,19 +1530,19 @@ define <4 x double> @uitofp_4i1_double(<4 x i32> %a) {
}

define <2 x float> @uitofp_2i1_float(<2 x i32> %a) {
; NOVL-LABEL: uitofp_2i1_float:
; NOVL: # BB#0:
; NOVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; NOVL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
; NOVL-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
; NOVL-NEXT: vpextrb $8, %xmm0, %eax
; NOVL-NEXT: andl $1, %eax
; NOVL-NEXT: vcvtsi2ssl %eax, %xmm2, %xmm1
; NOVL-NEXT: vpextrb $0, %xmm0, %eax
; NOVL-NEXT: andl $1, %eax
; NOVL-NEXT: vcvtsi2ssl %eax, %xmm2, %xmm0
; NOVL-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[2,3]
; NOVL-NEXT: retq
; KNL-LABEL: uitofp_2i1_float:
; KNL: # BB#0:
; KNL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; KNL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
; KNL-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
; KNL-NEXT: vpextrb $8, %xmm0, %eax
; KNL-NEXT: andl $1, %eax
; KNL-NEXT: vpextrb $0, %xmm0, %ecx
; KNL-NEXT: vcvtsi2ssl %eax, %xmm2, %xmm0
; KNL-NEXT: andl $1, %ecx
; KNL-NEXT: vcvtsi2ssl %ecx, %xmm2, %xmm1
; KNL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
; KNL-NEXT: retq
;
; VL-LABEL: uitofp_2i1_float:
; VL: # BB#0:
Expand All @@ -1552,6 +1552,34 @@ define <2 x float> @uitofp_2i1_float(<2 x i32> %a) {
; VL-NEXT: vpbroadcastd {{.*}}(%rip), %xmm0 {%k1} {z}
; VL-NEXT: vcvtudq2ps %xmm0, %xmm0
; VL-NEXT: retq
;
; AVX512DQ-LABEL: uitofp_2i1_float:
; AVX512DQ: # BB#0:
; AVX512DQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512DQ-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
; AVX512DQ-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
; AVX512DQ-NEXT: vpextrb $8, %xmm0, %eax
; AVX512DQ-NEXT: andl $1, %eax
; AVX512DQ-NEXT: vcvtsi2ssl %eax, %xmm2, %xmm1
; AVX512DQ-NEXT: vpextrb $0, %xmm0, %eax
; AVX512DQ-NEXT: andl $1, %eax
; AVX512DQ-NEXT: vcvtsi2ssl %eax, %xmm2, %xmm0
; AVX512DQ-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[2,3]
; AVX512DQ-NEXT: retq
;
; AVX512BW-LABEL: uitofp_2i1_float:
; AVX512BW: # BB#0:
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
; AVX512BW-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
; AVX512BW-NEXT: vpextrb $8, %xmm0, %eax
; AVX512BW-NEXT: andl $1, %eax
; AVX512BW-NEXT: vcvtsi2ssl %eax, %xmm2, %xmm1
; AVX512BW-NEXT: vpextrb $0, %xmm0, %eax
; AVX512BW-NEXT: andl $1, %eax
; AVX512BW-NEXT: vcvtsi2ssl %eax, %xmm2, %xmm0
; AVX512BW-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[2,3]
; AVX512BW-NEXT: retq
%mask = icmp ult <2 x i32> %a, zeroinitializer
%1 = uitofp <2 x i1> %mask to <2 x float>
ret <2 x float> %1
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/X86/avx512-ext.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -48,8 +48,8 @@ define <8 x i16> @sext_8x8mem_to_8x16(<8 x i8> *%i , <8 x i1> %mask) nounwind re
define <16 x i16> @zext_16x8mem_to_16x16(<16 x i8> *%i , <16 x i1> %mask) nounwind readnone {
; KNL-LABEL: zext_16x8mem_to_16x16:
; KNL: # BB#0:
; KNL-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; KNL-NEXT: vpmovzxbw {{.*#+}} ymm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
; KNL-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; KNL-NEXT: vpsllw $15, %ymm0, %ymm0
; KNL-NEXT: vpsraw $15, %ymm0, %ymm0
; KNL-NEXT: vpand %ymm1, %ymm0, %ymm0
Expand All @@ -70,8 +70,8 @@ define <16 x i16> @zext_16x8mem_to_16x16(<16 x i8> *%i , <16 x i1> %mask) nounwi
define <16 x i16> @sext_16x8mem_to_16x16(<16 x i8> *%i , <16 x i1> %mask) nounwind readnone {
; KNL-LABEL: sext_16x8mem_to_16x16:
; KNL: # BB#0:
; KNL-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; KNL-NEXT: vpmovsxbw (%rdi), %ymm1
; KNL-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; KNL-NEXT: vpsllw $15, %ymm0, %ymm0
; KNL-NEXT: vpsraw $15, %ymm0, %ymm0
; KNL-NEXT: vpand %ymm1, %ymm0, %ymm0
Expand Down
47 changes: 28 additions & 19 deletions llvm/test/CodeGen/X86/avx512-insert-extract.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -936,7 +936,6 @@ define i32 @test_insertelement_v32i1(i32 %a, i32 %b, <32 x i32> %x , <32 x i32>
; KNL-NEXT: subq $32, %rsp
; KNL-NEXT: xorl %eax, %eax
; KNL-NEXT: cmpl %esi, %edi
; KNL-NEXT: setb %al
; KNL-NEXT: vpcmpltud %zmm3, %zmm1, %k0
; KNL-NEXT: kshiftlw $14, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
Expand Down Expand Up @@ -1062,6 +1061,7 @@ define i32 @test_insertelement_v32i1(i32 %a, i32 %b, <32 x i32> %x , <32 x i32>
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %ecx
; KNL-NEXT: vpinsrb $14, %ecx, %xmm0, %xmm0
; KNL-NEXT: setb %al
; KNL-NEXT: kshiftrw $15, %k0, %k0
; KNL-NEXT: kmovw %k0, %ecx
; KNL-NEXT: vpinsrb $15, %ecx, %xmm0, %xmm0
Expand Down Expand Up @@ -1112,23 +1112,23 @@ define i8 @test_iinsertelement_v4i1(i32 %a, i32 %b, <4 x i32> %x , <4 x i32> %y)
; KNL-LABEL: test_iinsertelement_v4i1:
; KNL: ## BB#0:
; KNL-NEXT: cmpl %esi, %edi
; KNL-NEXT: setb %al
; KNL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
; KNL-NEXT: vpxor %xmm2, %xmm0, %xmm0
; KNL-NEXT: vpxor %xmm2, %xmm1, %xmm1
; KNL-NEXT: vpcmpgtd %xmm0, %xmm1, %xmm0
; KNL-NEXT: vpextrb $4, %xmm0, %ecx
; KNL-NEXT: kmovw %ecx, %k1
; KNL-NEXT: vpextrb $4, %xmm0, %eax
; KNL-NEXT: setb %cl
; KNL-NEXT: kmovw %eax, %k1
; KNL-NEXT: vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
; KNL-NEXT: vpextrb $0, %xmm0, %ecx
; KNL-NEXT: kmovw %ecx, %k1
; KNL-NEXT: vpextrb $0, %xmm0, %eax
; KNL-NEXT: kmovw %eax, %k1
; KNL-NEXT: vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
; KNL-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,8,2,3,4,5,6,7]
; KNL-NEXT: vpermi2q %zmm1, %zmm2, %zmm3
; KNL-NEXT: vpsllq $63, %zmm3, %zmm1
; KNL-NEXT: vptestmq %zmm1, %zmm1, %k1
; KNL-NEXT: vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
; KNL-NEXT: kmovw %eax, %k1
; KNL-NEXT: kmovw %ecx, %k1
; KNL-NEXT: vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
; KNL-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,8,3,4,5,6,7]
; KNL-NEXT: vpermi2q %zmm2, %zmm1, %zmm3
Expand Down Expand Up @@ -1902,14 +1902,23 @@ define i16 @test_extractelement_variable_v32i16(<32 x i16> %t1, i32 %index) {
}

define i8 @test_extractelement_variable_v16i8(<16 x i8> %t1, i32 %index) {
; CHECK-LABEL: test_extractelement_variable_v16i8:
; CHECK: ## BB#0:
; CHECK-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; CHECK-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
; CHECK-NEXT: andl $15, %edi
; CHECK-NEXT: leaq -{{[0-9]+}}(%rsp), %rax
; CHECK-NEXT: movb (%rdi,%rax), %al
; CHECK-NEXT: retq
; KNL-LABEL: test_extractelement_variable_v16i8:
; KNL: ## BB#0:
; KNL-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: andl $15, %edi
; KNL-NEXT: leaq -{{[0-9]+}}(%rsp), %rax
; KNL-NEXT: movb (%rdi,%rax), %al
; KNL-NEXT: retq
;
; SKX-LABEL: test_extractelement_variable_v16i8:
; SKX: ## BB#0:
; SKX-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; SKX-NEXT: vmovaps %xmm0, -{{[0-9]+}}(%rsp)
; SKX-NEXT: andl $15, %edi
; SKX-NEXT: leaq -{{[0-9]+}}(%rsp), %rax
; SKX-NEXT: movb (%rdi,%rax), %al
; SKX-NEXT: retq
%t2 = extractelement <16 x i8> %t1, i32 %index
ret i8 %t2
}
Expand All @@ -1927,8 +1936,8 @@ define i8 @test_extractelement_variable_v32i8(<32 x i8> %t1, i32 %index) {
; KNL-NEXT: .cfi_def_cfa_register %rbp
; KNL-NEXT: andq $-32, %rsp
; KNL-NEXT: subq $64, %rsp
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: vmovaps %ymm0, (%rsp)
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: andl $31, %edi
; KNL-NEXT: movq %rsp, %rax
; KNL-NEXT: movb (%rdi,%rax), %al
Expand Down Expand Up @@ -1975,9 +1984,9 @@ define i8 @test_extractelement_variable_v64i8(<64 x i8> %t1, i32 %index) {
; KNL-NEXT: .cfi_def_cfa_register %rbp
; KNL-NEXT: andq $-64, %rsp
; KNL-NEXT: subq $128, %rsp
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: vmovaps %ymm1, {{[0-9]+}}(%rsp)
; KNL-NEXT: vmovaps %ymm0, (%rsp)
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: andl $63, %edi
; KNL-NEXT: movq %rsp, %rax
; KNL-NEXT: movb (%rdi,%rax), %al
Expand Down Expand Up @@ -2066,12 +2075,12 @@ define i8 @test_extractelement_variable_v64i8_indexi8(<64 x i8> %t1, i8 %index)
define zeroext i8 @test_extractelement_varible_v2i1(<2 x i64> %a, <2 x i64> %b, i32 %index) {
; KNL-LABEL: test_extractelement_varible_v2i1:
; KNL: ## BB#0:
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; KNL-NEXT: vpxor %xmm2, %xmm1, %xmm1
; KNL-NEXT: vpxor %xmm2, %xmm0, %xmm0
; KNL-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0
; KNL-NEXT: vmovdqa %xmm0, -{{[0-9]+}}(%rsp)
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: andl $1, %edi
; KNL-NEXT: movl -24(%rsp,%rdi,8), %eax
; KNL-NEXT: andl $1, %eax
Expand All @@ -2096,12 +2105,12 @@ define zeroext i8 @test_extractelement_varible_v2i1(<2 x i64> %a, <2 x i64> %b,
define zeroext i8 @test_extractelement_varible_v4i1(<4 x i32> %a, <4 x i32> %b, i32 %index) {
; KNL-LABEL: test_extractelement_varible_v4i1:
; KNL: ## BB#0:
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
; KNL-NEXT: vpxor %xmm2, %xmm1, %xmm1
; KNL-NEXT: vpxor %xmm2, %xmm0, %xmm0
; KNL-NEXT: vpcmpgtd %xmm1, %xmm0, %xmm0
; KNL-NEXT: vmovdqa %xmm0, -{{[0-9]+}}(%rsp)
; KNL-NEXT: ## kill: %EDI<def> %EDI<kill> %RDI<def>
; KNL-NEXT: andl $3, %edi
; KNL-NEXT: movl -24(%rsp,%rdi,4), %eax
; KNL-NEXT: andl $1, %eax
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/X86/avx512-intrinsics-upgrade.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2880,7 +2880,6 @@ declare <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32>, <16 x i32>, <8
define <4 x float> @test_mask_vextractf32x4(<4 x float> %b, <16 x float> %a, i8 %mask) {
; CHECK-LABEL: test_mask_vextractf32x4:
; CHECK: ## BB#0:
; CHECK-NEXT: vextractf32x4 $2, %zmm1, %xmm1
; CHECK-NEXT: kmovw %edi, %k0
; CHECK-NEXT: kshiftlw $12, %k0, %k1
; CHECK-NEXT: kshiftrw $15, %k1, %k1
Expand All @@ -2898,6 +2897,7 @@ define <4 x float> @test_mask_vextractf32x4(<4 x float> %b, <16 x float> %a, i8
; CHECK-NEXT: vpinsrb $8, %eax, %xmm2, %xmm2
; CHECK-NEXT: kmovw %k1, %eax
; CHECK-NEXT: vpinsrb $12, %eax, %xmm2, %xmm2
; CHECK-NEXT: vextractf32x4 $2, %zmm1, %xmm1
; CHECK-NEXT: vpslld $31, %xmm2, %xmm2
; CHECK-NEXT: vblendvps %xmm2, %xmm1, %xmm0, %xmm0
; CHECK-NEXT: retq
Expand Down Expand Up @@ -2941,7 +2941,6 @@ declare <4 x i64> @llvm.x86.avx512.mask.vextracti64x4.512(<8 x i64>, i32, <4 x i
define <4 x i32> @test_maskz_vextracti32x4(<16 x i32> %a, i8 %mask) {
; CHECK-LABEL: test_maskz_vextracti32x4:
; CHECK: ## BB#0:
; CHECK-NEXT: vextracti32x4 $2, %zmm0, %xmm0
; CHECK-NEXT: kmovw %edi, %k0
; CHECK-NEXT: kshiftlw $12, %k0, %k1
; CHECK-NEXT: kshiftrw $15, %k1, %k1
Expand All @@ -2959,6 +2958,7 @@ define <4 x i32> @test_maskz_vextracti32x4(<16 x i32> %a, i8 %mask) {
; CHECK-NEXT: vpinsrb $8, %eax, %xmm1, %xmm1
; CHECK-NEXT: kmovw %k1, %eax
; CHECK-NEXT: vpinsrb $12, %eax, %xmm1, %xmm1
; CHECK-NEXT: vextracti32x4 $2, %zmm0, %xmm0
; CHECK-NEXT: vpslld $31, %xmm1, %xmm1
; CHECK-NEXT: vpsrad $31, %xmm1, %xmm1
; CHECK-NEXT: vpand %xmm0, %xmm1, %xmm0
Expand Down
554 changes: 259 additions & 295 deletions llvm/test/CodeGen/X86/avx512-mask-op.ll
View file

Large diffs are not rendered by default.

54 changes: 27 additions & 27 deletions llvm/test/CodeGen/X86/avx512-vec-cmp.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -269,8 +269,6 @@ define i32 @test12_v32i32(<32 x i32> %a, <32 x i32> %b) nounwind {
; KNL-NEXT: vpinsrb $15, %eax, %xmm1, %xmm1
; KNL-NEXT: vpmovsxbd %xmm1, %zmm1
; KNL-NEXT: vpslld $31, %zmm1, %zmm1
; KNL-NEXT: vptestmd %zmm1, %zmm1, %k0
; KNL-NEXT: kmovw %k0, {{[0-9]+}}(%rsp)
; KNL-NEXT: vpcmpeqd %zmm2, %zmm0, %k0
; KNL-NEXT: kshiftlw $14, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
Expand Down Expand Up @@ -327,11 +325,13 @@ define i32 @test12_v32i32(<32 x i32> %a, <32 x i32> %b) nounwind {
; KNL-NEXT: kshiftlw $2, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $13, %eax, %xmm0, %xmm0
; KNL-NEXT: kshiftlw $1, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0
; KNL-NEXT: kmovw %k1, %ecx
; KNL-NEXT: vptestmd %zmm1, %zmm1, %k1
; KNL-NEXT: vpinsrb $13, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $14, %ecx, %xmm0, %xmm0
; KNL-NEXT: kmovw %k1, {{[0-9]+}}(%rsp)
; KNL-NEXT: kshiftrw $15, %k0, %k0
; KNL-NEXT: kmovw %k0, %eax
; KNL-NEXT: vpinsrb $15, %eax, %xmm0, %xmm0
Expand Down Expand Up @@ -577,75 +577,75 @@ define i64 @test12_v64i16(<64 x i16> %a, <64 x i16> %b) nounwind {
; KNL-NEXT: vpinsrb $15, %eax, %xmm0, %xmm0
; KNL-NEXT: vpmovsxbd %xmm0, %zmm0
; KNL-NEXT: vpslld $31, %zmm0, %zmm0
; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0
; KNL-NEXT: kmovw %k0, {{[0-9]+}}(%rsp)
; KNL-NEXT: vpcmpeqw %ymm6, %ymm2, %ymm0
; KNL-NEXT: vpmovsxwd %ymm0, %zmm0
; KNL-NEXT: vpslld $31, %zmm0, %zmm0
; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0
; KNL-NEXT: vpcmpeqw %ymm6, %ymm2, %ymm1
; KNL-NEXT: vpmovsxwd %ymm1, %zmm1
; KNL-NEXT: vpslld $31, %zmm1, %zmm1
; KNL-NEXT: vptestmd %zmm1, %zmm1, %k0
; KNL-NEXT: kshiftlw $14, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: kshiftlw $15, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %ecx
; KNL-NEXT: vmovd %ecx, %xmm0
; KNL-NEXT: vpinsrb $1, %eax, %xmm0, %xmm0
; KNL-NEXT: vmovd %ecx, %xmm1
; KNL-NEXT: vpinsrb $1, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $13, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $2, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $12, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $3, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $3, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $11, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $4, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $10, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $5, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $5, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $9, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $6, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $6, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $8, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $7, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $7, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $7, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $8, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $6, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $9, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $9, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $5, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $10, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $4, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $11, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $11, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $3, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $12, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $2, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $13, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $13, %eax, %xmm1, %xmm1
; KNL-NEXT: kshiftlw $1, %k0, %k1
; KNL-NEXT: kshiftrw $15, %k1, %k1
; KNL-NEXT: kmovw %k1, %eax
; KNL-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $14, %eax, %xmm1, %xmm1
; KNL-NEXT: vptestmd %zmm0, %zmm0, %k1
; KNL-NEXT: kshiftrw $15, %k0, %k0
; KNL-NEXT: kmovw %k0, %eax
; KNL-NEXT: vpinsrb $15, %eax, %xmm0, %xmm0
; KNL-NEXT: vpinsrb $15, %eax, %xmm1, %xmm0
; KNL-NEXT: kmovw %k1, {{[0-9]+}}(%rsp)
; KNL-NEXT: vpmovsxbd %xmm0, %zmm0
; KNL-NEXT: vpslld $31, %zmm0, %zmm0
; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0
Expand Down
1,512 changes: 754 additions & 758 deletions llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
View file

Large diffs are not rendered by default.

44 changes: 22 additions & 22 deletions llvm/test/CodeGen/X86/avx512bwvl-intrinsics-upgrade.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2750,23 +2750,23 @@ define <8 x i32> @test_mask_cmp_b_256(<32 x i8> %a0, <32 x i8> %a1, i32 %mask) {
; CHECK-NEXT: vpcmpgtb %ymm0, %ymm1, %k0 {%k1} ## encoding: [0x62,0xf1,0x75,0x29,0x64,0xc0]
; CHECK-NEXT: kmovd %k0, %r9d ## encoding: [0xc5,0x7b,0x93,0xc8]
; CHECK-NEXT: vpcmpleb %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf3,0x7d,0x29,0x3f,0xc1,0x02]
; CHECK-NEXT: kmovd %k0, %r10d ## encoding: [0xc5,0x7b,0x93,0xd0]
; CHECK-NEXT: kxord %k0, %k0, %k0 ## encoding: [0xc4,0xe1,0xfd,0x47,0xc0]
; CHECK-NEXT: kmovd %k0, %esi ## encoding: [0xc5,0xfb,0x93,0xf0]
; CHECK-NEXT: kmovd %k0, %edx ## encoding: [0xc5,0xfb,0x93,0xd0]
; CHECK-NEXT: vpcmpneqb %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf3,0x7d,0x29,0x3f,0xc1,0x04]
; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT: kmovd %k0, %esi ## encoding: [0xc5,0xfb,0x93,0xf0]
; CHECK-NEXT: vpcmpleb %ymm0, %ymm1, %k0 {%k1} ## encoding: [0x62,0xf3,0x75,0x29,0x3f,0xc0,0x02]
; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT: vpcmpgtb %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0x64,0xc1]
; CHECK-NEXT: kmovd %k0, %edx ## encoding: [0xc5,0xfb,0x93,0xd0]
; CHECK-NEXT: vmovd %eax, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc0]
; CHECK-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc1,0x01]
; CHECK-NEXT: vpinsrd $2, %edx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc2,0x02]
; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
; CHECK-NEXT: vmovd %esi, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc6]
; CHECK-NEXT: vpinsrd $1, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc0,0x01]
; CHECK-NEXT: vpinsrd $2, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc1,0x02]
; CHECK-NEXT: vpinsrd $3, %edi, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc7,0x03]
; CHECK-NEXT: vmovd %r8d, %xmm1 ## EVEX TO VEX Compression encoding: [0xc4,0xc1,0x79,0x6e,0xc8]
; CHECK-NEXT: vpinsrd $1, %r9d, %xmm1, %xmm1 ## encoding: [0xc4,0xc3,0x71,0x22,0xc9,0x01]
; CHECK-NEXT: vpinsrd $2, %r10d, %xmm1, %xmm1 ## encoding: [0xc4,0xc3,0x71,0x22,0xca,0x02]
; CHECK-NEXT: vpinsrd $3, %esi, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x22,0xce,0x03]
; CHECK-NEXT: vpinsrd $2, %edx, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x22,0xca,0x02]
; CHECK-NEXT: kxord %k0, %k0, %k0 ## encoding: [0xc4,0xe1,0xfd,0x47,0xc0]
; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT: vpinsrd $3, %eax, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x22,0xc8,0x03]
; CHECK-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe3,0x75,0x38,0xc0,0x01]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res0 = call i32 @llvm.x86.avx512.mask.cmp.b.256(<32 x i8> %a0, <32 x i8> %a1, i32 0, i32 %mask)
Expand Down Expand Up @@ -2848,23 +2848,23 @@ define <8 x i32> @test_mask_ucmp_b_256(<32 x i8> %a0, <32 x i8> %a1, i32 %mask)
; CHECK-NEXT: vpcmpltub %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf3,0x7d,0x29,0x3e,0xc1,0x01]
; CHECK-NEXT: kmovd %k0, %r9d ## encoding: [0xc5,0x7b,0x93,0xc8]
; CHECK-NEXT: vpcmpleub %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf3,0x7d,0x29,0x3e,0xc1,0x02]
; CHECK-NEXT: kmovd %k0, %r10d ## encoding: [0xc5,0x7b,0x93,0xd0]
; CHECK-NEXT: kxord %k0, %k0, %k0 ## encoding: [0xc4,0xe1,0xfd,0x47,0xc0]
; CHECK-NEXT: kmovd %k0, %esi ## encoding: [0xc5,0xfb,0x93,0xf0]
; CHECK-NEXT: kmovd %k0, %edx ## encoding: [0xc5,0xfb,0x93,0xd0]
; CHECK-NEXT: vpcmpneqb %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf3,0x7d,0x29,0x3f,0xc1,0x04]
; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT: kmovd %k0, %esi ## encoding: [0xc5,0xfb,0x93,0xf0]
; CHECK-NEXT: vpcmpnltub %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf3,0x7d,0x29,0x3e,0xc1,0x05]
; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT: vpcmpnleub %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf3,0x7d,0x29,0x3e,0xc1,0x06]
; CHECK-NEXT: kmovd %k0, %edx ## encoding: [0xc5,0xfb,0x93,0xd0]
; CHECK-NEXT: vmovd %eax, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc0]
; CHECK-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc1,0x01]
; CHECK-NEXT: vpinsrd $2, %edx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc2,0x02]
; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
; CHECK-NEXT: vmovd %esi, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc6]
; CHECK-NEXT: vpinsrd $1, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc0,0x01]
; CHECK-NEXT: vpinsrd $2, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc1,0x02]
; CHECK-NEXT: vpinsrd $3, %edi, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x22,0xc7,0x03]
; CHECK-NEXT: vmovd %r8d, %xmm1 ## EVEX TO VEX Compression encoding: [0xc4,0xc1,0x79,0x6e,0xc8]
; CHECK-NEXT: vpinsrd $1, %r9d, %xmm1, %xmm1 ## encoding: [0xc4,0xc3,0x71,0x22,0xc9,0x01]
; CHECK-NEXT: vpinsrd $2, %r10d, %xmm1, %xmm1 ## encoding: [0xc4,0xc3,0x71,0x22,0xca,0x02]
; CHECK-NEXT: vpinsrd $3, %esi, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x22,0xce,0x03]
; CHECK-NEXT: vpinsrd $2, %edx, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x22,0xca,0x02]
; CHECK-NEXT: kxord %k0, %k0, %k0 ## encoding: [0xc4,0xe1,0xfd,0x47,0xc0]
; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT: vpinsrd $3, %eax, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x22,0xc8,0x03]
; CHECK-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe3,0x75,0x38,0xc0,0x01]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res0 = call i32 @llvm.x86.avx512.mask.ucmp.b.256(<32 x i8> %a0, <32 x i8> %a1, i32 0, i32 %mask)
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/X86/avx512dq-intrinsics-upgrade.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,6 @@ declare <2 x double> @llvm.x86.avx512.mask.vextractf64x2.512(<8 x double>, i32,
define <2 x double>@test_int_x86_avx512_mask_vextractf64x2_512(<8 x double> %x0, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vextractf64x2_512:
; CHECK: ## BB#0:
; CHECK-NEXT: vextractf32x4 $1, %zmm0, %xmm0
; CHECK-NEXT: kmovw %edi, %k0
; CHECK-NEXT: kshiftlb $7, %k0, %k1
; CHECK-NEXT: kshiftrb $7, %k1, %k1
Expand All @@ -16,6 +15,7 @@ define <2 x double>@test_int_x86_avx512_mask_vextractf64x2_512(<8 x double> %x0,
; CHECK-NEXT: kmovw %k1, %ecx
; CHECK-NEXT: vmovd %ecx, %xmm2
; CHECK-NEXT: vpinsrb $8, %eax, %xmm2, %xmm2
; CHECK-NEXT: vextractf32x4 $1, %zmm0, %xmm0
; CHECK-NEXT: vpsllq $63, %xmm2, %xmm2
; CHECK-NEXT: vpsraq $63, %zmm2, %zmm2
; CHECK-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/X86/avx512vl-vec-cmp.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -314,8 +314,8 @@ define <4 x i64> @test256_11(<4 x i64> %x, <4 x i64>* %y.ptr, <4 x i64> %x1, <4
;
; NoVLX-LABEL: test256_11:
; NoVLX: # BB#0:
; NoVLX-NEXT: vpcmpgtq %ymm2, %ymm1, %ymm2
; NoVLX-NEXT: vpcmpgtq (%rdi), %ymm0, %ymm3
; NoVLX-NEXT: vpcmpgtq %ymm2, %ymm1, %ymm2
; NoVLX-NEXT: vpand %ymm2, %ymm3, %ymm2
; NoVLX-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
; NoVLX-NEXT: retq
Expand Down Expand Up @@ -824,8 +824,8 @@ define <2 x i64> @test128_11(<2 x i64> %x, <2 x i64>* %y.ptr, <2 x i64> %x1, <2
;
; NoVLX-LABEL: test128_11:
; NoVLX: # BB#0:
; NoVLX-NEXT: vpcmpgtq %xmm2, %xmm1, %xmm2
; NoVLX-NEXT: vpcmpgtq (%rdi), %xmm0, %xmm3
; NoVLX-NEXT: vpcmpgtq %xmm2, %xmm1, %xmm2
; NoVLX-NEXT: vpand %xmm2, %xmm3, %xmm2
; NoVLX-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
; NoVLX-NEXT: retq
Expand Down
16,702 changes: 7,731 additions & 8,971 deletions llvm/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll
View file

Large diffs are not rendered by default.

56 changes: 28 additions & 28 deletions llvm/test/CodeGen/X86/bmi-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,10 +20,10 @@ define i16 @test_andn_i16(i16 zeroext %a0, i16 zeroext %a1, i16 *%a2) {
; HASWELL: # BB#0:
; HASWELL-NEXT: andnl %esi, %edi, %eax # sched: [1:0.50]
; HASWELL-NEXT: notl %edi # sched: [1:0.25]
; HASWELL-NEXT: andw (%rdx), %di # sched: [5:0.50]
; HASWELL-NEXT: andw (%rdx), %di # sched: [1:0.50]
; HASWELL-NEXT: addl %edi, %eax # sched: [1:0.25]
; HASWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_andn_i16:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -61,9 +61,9 @@ define i32 @test_andn_i32(i32 %a0, i32 %a1, i32 *%a2) {
; HASWELL-LABEL: test_andn_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: andnl %esi, %edi, %ecx # sched: [1:0.50]
; HASWELL-NEXT: andnl (%rdx), %edi, %eax # sched: [4:0.50]
; HASWELL-NEXT: andnl (%rdx), %edi, %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_andn_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -97,9 +97,9 @@ define i64 @test_andn_i64(i64 %a0, i64 %a1, i64 *%a2) {
; HASWELL-LABEL: test_andn_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: andnq %rsi, %rdi, %rcx # sched: [1:0.50]
; HASWELL-NEXT: andnq (%rdx), %rdi, %rax # sched: [4:0.50]
; HASWELL-NEXT: andnq (%rdx), %rdi, %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_andn_i64:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -132,10 +132,10 @@ define i32 @test_bextr_i32(i32 %a0, i32 %a1, i32 *%a2) {
;
; HASWELL-LABEL: test_bextr_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: bextrl %edi, (%rdx), %ecx # sched: [6:0.50]
; HASWELL-NEXT: bextrl %edi, (%rdx), %ecx # sched: [2:0.50]
; HASWELL-NEXT: bextrl %edi, %esi, %eax # sched: [2:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_bextr_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -168,10 +168,10 @@ define i64 @test_bextr_i64(i64 %a0, i64 %a1, i64 *%a2) {
;
; HASWELL-LABEL: test_bextr_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: bextrq %rdi, (%rdx), %rcx # sched: [6:0.50]
; HASWELL-NEXT: bextrq %rdi, (%rdx), %rcx # sched: [2:0.50]
; HASWELL-NEXT: bextrq %rdi, %rsi, %rax # sched: [2:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_bextr_i64:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -204,10 +204,10 @@ define i32 @test_blsi_i32(i32 %a0, i32 *%a1) {
;
; HASWELL-LABEL: test_blsi_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: blsil (%rsi), %ecx # sched: [4:0.50]
; HASWELL-NEXT: blsil (%rsi), %ecx # sched: [1:0.50]
; HASWELL-NEXT: blsil %edi, %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_blsi_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -241,10 +241,10 @@ define i64 @test_blsi_i64(i64 %a0, i64 *%a1) {
;
; HASWELL-LABEL: test_blsi_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: blsiq (%rsi), %rcx # sched: [4:0.50]
; HASWELL-NEXT: blsiq (%rsi), %rcx # sched: [1:0.50]
; HASWELL-NEXT: blsiq %rdi, %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_blsi_i64:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -278,10 +278,10 @@ define i32 @test_blsmsk_i32(i32 %a0, i32 *%a1) {
;
; HASWELL-LABEL: test_blsmsk_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: blsmskl (%rsi), %ecx # sched: [4:0.50]
; HASWELL-NEXT: blsmskl (%rsi), %ecx # sched: [1:0.50]
; HASWELL-NEXT: blsmskl %edi, %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_blsmsk_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -315,10 +315,10 @@ define i64 @test_blsmsk_i64(i64 %a0, i64 *%a1) {
;
; HASWELL-LABEL: test_blsmsk_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: blsmskq (%rsi), %rcx # sched: [4:0.50]
; HASWELL-NEXT: blsmskq (%rsi), %rcx # sched: [1:0.50]
; HASWELL-NEXT: blsmskq %rdi, %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_blsmsk_i64:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -352,10 +352,10 @@ define i32 @test_blsr_i32(i32 %a0, i32 *%a1) {
;
; HASWELL-LABEL: test_blsr_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: blsrl (%rsi), %ecx # sched: [4:0.50]
; HASWELL-NEXT: blsrl (%rsi), %ecx # sched: [1:0.50]
; HASWELL-NEXT: blsrl %edi, %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_blsr_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -389,10 +389,10 @@ define i64 @test_blsr_i64(i64 %a0, i64 *%a1) {
;
; HASWELL-LABEL: test_blsr_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: blsrq (%rsi), %rcx # sched: [4:0.50]
; HASWELL-NEXT: blsrq (%rsi), %rcx # sched: [1:0.50]
; HASWELL-NEXT: blsrq %rdi, %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_blsr_i64:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -427,11 +427,11 @@ define i16 @test_cttz_i16(i16 zeroext %a0, i16 *%a1) {
;
; HASWELL-LABEL: test_cttz_i16:
; HASWELL: # BB#0:
; HASWELL-NEXT: tzcntw (%rsi), %cx # sched: [7:1.00]
; HASWELL-NEXT: tzcntw (%rsi), %cx # sched: [3:1.00]
; HASWELL-NEXT: tzcntw %di, %ax # sched: [3:1.00]
; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_cttz_i16:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -466,10 +466,10 @@ define i32 @test_cttz_i32(i32 %a0, i32 *%a1) {
;
; HASWELL-LABEL: test_cttz_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: tzcntl (%rsi), %ecx # sched: [7:1.00]
; HASWELL-NEXT: tzcntl (%rsi), %ecx # sched: [3:1.00]
; HASWELL-NEXT: tzcntl %edi, %eax # sched: [3:1.00]
; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_cttz_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -502,10 +502,10 @@ define i64 @test_cttz_i64(i64 %a0, i64 *%a1) {
;
; HASWELL-LABEL: test_cttz_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: tzcntq (%rsi), %rcx # sched: [7:1.00]
; HASWELL-NEXT: tzcntq (%rsi), %rcx # sched: [3:1.00]
; HASWELL-NEXT: tzcntq %rdi, %rax # sched: [3:1.00]
; HASWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_cttz_i64:
; BTVER2: # BB#0:
Expand Down
60 changes: 30 additions & 30 deletions llvm/test/CodeGen/X86/bmi2-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,10 +15,10 @@ define i32 @test_bzhi_i32(i32 %a0, i32 %a1, i32 *%a2) {
;
; HASWELL-LABEL: test_bzhi_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: bzhil %edi, (%rdx), %ecx # sched: [4:0.50]
; HASWELL-NEXT: bzhil %edi, (%rdx), %ecx # sched: [1:0.50]
; HASWELL-NEXT: bzhil %edi, %esi, %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_bzhi_i32:
; ZNVER1: # BB#0:
Expand All @@ -44,10 +44,10 @@ define i64 @test_bzhi_i64(i64 %a0, i64 %a1, i64 *%a2) {
;
; HASWELL-LABEL: test_bzhi_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: bzhiq %rdi, (%rdx), %rcx # sched: [4:0.50]
; HASWELL-NEXT: bzhiq %rdi, (%rdx), %rcx # sched: [1:0.50]
; HASWELL-NEXT: bzhiq %rdi, %rsi, %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_bzhi_i64:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -80,9 +80,9 @@ define i64 @test_mulx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; HASWELL-NEXT: movq %rdx, %rax # sched: [1:0.25]
; HASWELL-NEXT: movq %rdi, %rdx # sched: [1:0.25]
; HASWELL-NEXT: mulxq %rsi, %rsi, %rcx # sched: [4:1.00]
; HASWELL-NEXT: mulxq (%rax), %rdx, %rax # sched: [8:1.00]
; HASWELL-NEXT: mulxq (%rax), %rdx, %rax # sched: [4:1.00]
; HASWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_mulx_i64:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -116,10 +116,10 @@ define i32 @test_pdep_i32(i32 %a0, i32 %a1, i32 *%a2) {
;
; HASWELL-LABEL: test_pdep_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: pdepl (%rdx), %edi, %ecx # sched: [7:1.00]
; HASWELL-NEXT: pdepl (%rdx), %edi, %ecx # sched: [3:1.00]
; HASWELL-NEXT: pdepl %esi, %edi, %eax # sched: [3:1.00]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pdep_i32:
; ZNVER1: # BB#0:
Expand All @@ -145,10 +145,10 @@ define i64 @test_pdep_i64(i64 %a0, i64 %a1, i64 *%a2) {
;
; HASWELL-LABEL: test_pdep_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: pdepq (%rdx), %rdi, %rcx # sched: [7:1.00]
; HASWELL-NEXT: pdepq (%rdx), %rdi, %rcx # sched: [3:1.00]
; HASWELL-NEXT: pdepq %rsi, %rdi, %rax # sched: [3:1.00]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pdep_i64:
; ZNVER1: # BB#0:
Expand All @@ -174,10 +174,10 @@ define i32 @test_pext_i32(i32 %a0, i32 %a1, i32 *%a2) {
;
; HASWELL-LABEL: test_pext_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: pextl (%rdx), %edi, %ecx # sched: [7:1.00]
; HASWELL-NEXT: pextl (%rdx), %edi, %ecx # sched: [3:1.00]
; HASWELL-NEXT: pextl %esi, %edi, %eax # sched: [3:1.00]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pext_i32:
; ZNVER1: # BB#0:
Expand All @@ -203,10 +203,10 @@ define i64 @test_pext_i64(i64 %a0, i64 %a1, i64 *%a2) {
;
; HASWELL-LABEL: test_pext_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: pextq (%rdx), %rdi, %rcx # sched: [7:1.00]
; HASWELL-NEXT: pextq (%rdx), %rdi, %rcx # sched: [3:1.00]
; HASWELL-NEXT: pextq %rsi, %rdi, %rax # sched: [3:1.00]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_pext_i64:
; ZNVER1: # BB#0:
Expand All @@ -233,9 +233,9 @@ define i32 @test_rorx_i32(i32 %a0, i32 %a1, i32 *%a2) {
; HASWELL-LABEL: test_rorx_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: rorxl $5, %edi, %ecx # sched: [1:0.50]
; HASWELL-NEXT: rorxl $5, (%rdx), %eax # sched: [5:0.50]
; HASWELL-NEXT: rorxl $5, (%rdx), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_rorx_i32:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -265,9 +265,9 @@ define i64 @test_rorx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; HASWELL-LABEL: test_rorx_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: rorxq $5, %rdi, %rcx # sched: [1:0.50]
; HASWELL-NEXT: rorxq $5, (%rdx), %rax # sched: [5:0.50]
; HASWELL-NEXT: rorxq $5, (%rdx), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_rorx_i64:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -297,9 +297,9 @@ define i32 @test_sarx_i32(i32 %a0, i32 %a1, i32 *%a2) {
; HASWELL-LABEL: test_sarx_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: sarxl %esi, %edi, %ecx # sched: [1:0.50]
; HASWELL-NEXT: sarxl %esi, (%rdx), %eax # sched: [5:0.50]
; HASWELL-NEXT: sarxl %esi, (%rdx), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sarx_i32:
; ZNVER1: # BB#0:
Expand All @@ -325,9 +325,9 @@ define i64 @test_sarx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; HASWELL-LABEL: test_sarx_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: sarxq %rsi, %rdi, %rcx # sched: [1:0.50]
; HASWELL-NEXT: sarxq %rsi, (%rdx), %rax # sched: [5:0.50]
; HASWELL-NEXT: sarxq %rsi, (%rdx), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sarx_i64:
; ZNVER1: # BB#0:
Expand All @@ -353,9 +353,9 @@ define i32 @test_shlx_i32(i32 %a0, i32 %a1, i32 *%a2) {
; HASWELL-LABEL: test_shlx_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: shlxl %esi, %edi, %ecx # sched: [1:0.50]
; HASWELL-NEXT: shlxl %esi, (%rdx), %eax # sched: [5:0.50]
; HASWELL-NEXT: shlxl %esi, (%rdx), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_shlx_i32:
; ZNVER1: # BB#0:
Expand All @@ -381,9 +381,9 @@ define i64 @test_shlx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; HASWELL-LABEL: test_shlx_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: shlxq %rsi, %rdi, %rcx # sched: [1:0.50]
; HASWELL-NEXT: shlxq %rsi, (%rdx), %rax # sched: [5:0.50]
; HASWELL-NEXT: shlxq %rsi, (%rdx), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_shlx_i64:
; ZNVER1: # BB#0:
Expand All @@ -409,9 +409,9 @@ define i32 @test_shrx_i32(i32 %a0, i32 %a1, i32 *%a2) {
; HASWELL-LABEL: test_shrx_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: shrxl %esi, %edi, %ecx # sched: [1:0.50]
; HASWELL-NEXT: shrxl %esi, (%rdx), %eax # sched: [5:0.50]
; HASWELL-NEXT: shrxl %esi, (%rdx), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_shrx_i32:
; ZNVER1: # BB#0:
Expand All @@ -437,9 +437,9 @@ define i64 @test_shrx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; HASWELL-LABEL: test_shrx_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: shrxq %rsi, %rdi, %rcx # sched: [1:0.50]
; HASWELL-NEXT: shrxq %rsi, (%rdx), %rax # sched: [5:0.50]
; HASWELL-NEXT: shrxq %rsi, (%rdx), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_shrx_i64:
; ZNVER1: # BB#0:
Expand Down
24 changes: 12 additions & 12 deletions llvm/test/CodeGen/X86/f16c-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -23,10 +23,10 @@ define <4 x float> @test_vcvtph2ps_128(<8 x i16> %a0, <8 x i16> *%a1) {
;
; HASWELL-LABEL: test_vcvtph2ps_128:
; HASWELL: # BB#0:
; HASWELL-NEXT: vcvtph2ps (%rdi), %xmm1 # sched: [7:1.00]
; HASWELL-NEXT: vcvtph2ps %xmm0, %xmm0 # sched: [4:1.00]
; HASWELL-NEXT: vcvtph2ps (%rdi), %xmm1 # sched: [1:1.00]
; HASWELL-NEXT: vcvtph2ps %xmm0, %xmm0 # sched: [2:1.00]
; HASWELL-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_vcvtph2ps_128:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -66,10 +66,10 @@ define <8 x float> @test_vcvtph2ps_256(<8 x i16> %a0, <8 x i16> *%a1) {
;
; HASWELL-LABEL: test_vcvtph2ps_256:
; HASWELL: # BB#0:
; HASWELL-NEXT: vcvtph2ps (%rdi), %ymm1 # sched: [7:1.00]
; HASWELL-NEXT: vcvtph2ps %xmm0, %ymm0 # sched: [4:1.00]
; HASWELL-NEXT: vcvtph2ps (%rdi), %ymm1 # sched: [1:1.00]
; HASWELL-NEXT: vcvtph2ps %xmm0, %ymm0 # sched: [2:1.00]
; HASWELL-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_vcvtph2ps_256:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -108,8 +108,8 @@ define <8 x i16> @test_vcvtps2ph_128(<4 x float> %a0, <4 x float> %a1, <4 x i16>
; HASWELL-LABEL: test_vcvtps2ph_128:
; HASWELL: # BB#0:
; HASWELL-NEXT: vcvtps2ph $0, %xmm0, %xmm0 # sched: [4:1.00]
; HASWELL-NEXT: vcvtps2ph $0, %xmm1, (%rdi) # sched: [8:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vcvtps2ph $0, %xmm1, (%rdi) # sched: [4:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_vcvtps2ph_128:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -147,10 +147,10 @@ define <8 x i16> @test_vcvtps2ph_256(<8 x float> %a0, <8 x float> %a1, <8 x i16>
;
; HASWELL-LABEL: test_vcvtps2ph_256:
; HASWELL: # BB#0:
; HASWELL-NEXT: vcvtps2ph $0, %ymm0, %xmm0 # sched: [4:1.00]
; HASWELL-NEXT: vcvtps2ph $0, %ymm1, (%rdi) # sched: [8:1.00]
; HASWELL-NEXT: vzeroupper # sched: [1:?]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vcvtps2ph $0, %ymm0, %xmm0 # sched: [6:1.00]
; HASWELL-NEXT: vcvtps2ph $0, %ymm1, (%rdi) # sched: [6:1.00]
; HASWELL-NEXT: vzeroupper # sched: [4:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_vcvtps2ph_256:
; BTVER2: # BB#0:
Expand Down
22 changes: 11 additions & 11 deletions llvm/test/CodeGen/X86/lea32-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,7 @@ define i32 @test_lea_offset(i32) {
; HASWELL: # BB#0:
; HASWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; HASWELL-NEXT: leal -24(%rdi), %eax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_offset:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -97,7 +97,7 @@ define i32 @test_lea_offset_big(i32) {
; HASWELL: # BB#0:
; HASWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; HASWELL-NEXT: leal 1024(%rdi), %eax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_offset_big:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -155,7 +155,7 @@ define i32 @test_lea_add(i32, i32) {
; HASWELL-NEXT: # kill: %ESI<def> %ESI<kill> %RSI<def>
; HASWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; HASWELL-NEXT: leal (%rdi,%rsi), %eax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -217,7 +217,7 @@ define i32 @test_lea_add_offset(i32, i32) {
; HASWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; HASWELL-NEXT: leal (%rdi,%rsi), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl $16, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_offset:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -283,7 +283,7 @@ define i32 @test_lea_add_offset_big(i32, i32) {
; HASWELL-NEXT: leal (%rdi,%rsi), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl $-4096, %eax # imm = 0xF000
; HASWELL-NEXT: # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_offset_big:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -338,7 +338,7 @@ define i32 @test_lea_mul(i32) {
; HASWELL: # BB#0:
; HASWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; HASWELL-NEXT: leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_mul:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -393,7 +393,7 @@ define i32 @test_lea_mul_offset(i32) {
; HASWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; HASWELL-NEXT: leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl $-32, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_mul_offset:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -452,7 +452,7 @@ define i32 @test_lea_mul_offset_big(i32) {
; HASWELL-NEXT: leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl $10000, %eax # imm = 0x2710
; HASWELL-NEXT: # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_mul_offset_big:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -510,7 +510,7 @@ define i32 @test_lea_add_scale(i32, i32) {
; HASWELL-NEXT: # kill: %ESI<def> %ESI<kill> %RSI<def>
; HASWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; HASWELL-NEXT: leal (%rdi,%rsi,2), %eax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_scale:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -573,7 +573,7 @@ define i32 @test_lea_add_scale_offset(i32, i32) {
; HASWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; HASWELL-NEXT: leal (%rdi,%rsi,4), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl $96, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_scale_offset:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -640,7 +640,7 @@ define i32 @test_lea_add_scale_offset_big(i32, i32) {
; HASWELL-NEXT: leal (%rdi,%rsi,8), %eax # sched: [1:0.50]
; HASWELL-NEXT: addl $-1200, %eax # imm = 0xFB50
; HASWELL-NEXT: # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_scale_offset_big:
; BTVER2: # BB#0:
Expand Down
22 changes: 11 additions & 11 deletions llvm/test/CodeGen/X86/lea64-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@ define i64 @test_lea_offset(i64) {
; HASWELL-LABEL: test_lea_offset:
; HASWELL: # BB#0:
; HASWELL-NEXT: leaq -24(%rdi), %rax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_offset:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -85,7 +85,7 @@ define i64 @test_lea_offset_big(i64) {
; HASWELL-LABEL: test_lea_offset_big:
; HASWELL: # BB#0:
; HASWELL-NEXT: leaq 1024(%rdi), %rax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_offset_big:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -131,7 +131,7 @@ define i64 @test_lea_add(i64, i64) {
; HASWELL-LABEL: test_lea_add:
; HASWELL: # BB#0:
; HASWELL-NEXT: leaq (%rdi,%rsi), %rax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -179,7 +179,7 @@ define i64 @test_lea_add_offset(i64, i64) {
; HASWELL: # BB#0:
; HASWELL-NEXT: leaq (%rdi,%rsi), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq $16, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_offset:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -231,7 +231,7 @@ define i64 @test_lea_add_offset_big(i64, i64) {
; HASWELL-NEXT: leaq (%rdi,%rsi), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq $-4096, %rax # imm = 0xF000
; HASWELL-NEXT: # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_offset_big:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -277,7 +277,7 @@ define i64 @test_lea_mul(i64) {
; HASWELL-LABEL: test_lea_mul:
; HASWELL: # BB#0:
; HASWELL-NEXT: leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_mul:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -325,7 +325,7 @@ define i64 @test_lea_mul_offset(i64) {
; HASWELL: # BB#0:
; HASWELL-NEXT: leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq $-32, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_mul_offset:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -377,7 +377,7 @@ define i64 @test_lea_mul_offset_big(i64) {
; HASWELL-NEXT: leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq $10000, %rax # imm = 0x2710
; HASWELL-NEXT: # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_mul_offset_big:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -423,7 +423,7 @@ define i64 @test_lea_add_scale(i64, i64) {
; HASWELL-LABEL: test_lea_add_scale:
; HASWELL: # BB#0:
; HASWELL-NEXT: leaq (%rdi,%rsi,2), %rax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_scale:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -472,7 +472,7 @@ define i64 @test_lea_add_scale_offset(i64, i64) {
; HASWELL: # BB#0:
; HASWELL-NEXT: leaq (%rdi,%rsi,4), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq $96, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_scale_offset:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -525,7 +525,7 @@ define i64 @test_lea_add_scale_offset_big(i64, i64) {
; HASWELL-NEXT: leaq (%rdi,%rsi,8), %rax # sched: [1:0.50]
; HASWELL-NEXT: addq $-1200, %rax # imm = 0xFB50
; HASWELL-NEXT: # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lea_add_scale_offset_big:
; BTVER2: # BB#0:
Expand Down
18 changes: 9 additions & 9 deletions llvm/test/CodeGen/X86/lzcnt-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -17,11 +17,11 @@ define i16 @test_ctlz_i16(i16 zeroext %a0, i16 *%a1) {
;
; HASWELL-LABEL: test_ctlz_i16:
; HASWELL: # BB#0:
; HASWELL-NEXT: lzcntw (%rsi), %cx
; HASWELL-NEXT: lzcntw %di, %ax
; HASWELL-NEXT: lzcntw (%rsi), %cx # sched: [3:1.00]
; HASWELL-NEXT: lzcntw %di, %ax # sched: [3:1.00]
; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctlz_i16:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -56,10 +56,10 @@ define i32 @test_ctlz_i32(i32 %a0, i32 *%a1) {
;
; HASWELL-LABEL: test_ctlz_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: lzcntl (%rsi), %ecx
; HASWELL-NEXT: lzcntl %edi, %eax
; HASWELL-NEXT: lzcntl (%rsi), %ecx # sched: [3:1.00]
; HASWELL-NEXT: lzcntl %edi, %eax # sched: [3:1.00]
; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctlz_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -92,10 +92,10 @@ define i64 @test_ctlz_i64(i64 %a0, i64 *%a1) {
;
; HASWELL-LABEL: test_ctlz_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: lzcntq (%rsi), %rcx
; HASWELL-NEXT: lzcntq %rdi, %rax
; HASWELL-NEXT: lzcntq (%rsi), %rcx # sched: [3:1.00]
; HASWELL-NEXT: lzcntq %rdi, %rax # sched: [3:1.00]
; HASWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctlz_i64:
; BTVER2: # BB#0:
Expand Down
10 changes: 5 additions & 5 deletions llvm/test/CodeGen/X86/movbe-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -33,9 +33,9 @@ define i16 @test_ctlz_i16(i16 *%a0, i16 %a1, i16 *%a2) {
;
; HASWELL-LABEL: test_ctlz_i16:
; HASWELL: # BB#0:
; HASWELL-NEXT: movbew (%rdi), %ax # sched: [6:0.50]
; HASWELL-NEXT: movbew (%rdi), %ax # sched: [1:0.50]
; HASWELL-NEXT: movbew %si, (%rdx) # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctlz_i16:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -83,7 +83,7 @@ define i32 @test_ctlz_i32(i32 *%a0, i32 %a1, i32 *%a2) {
; HASWELL: # BB#0:
; HASWELL-NEXT: movbel (%rdi), %eax # sched: [1:0.50]
; HASWELL-NEXT: movbel %esi, (%rdx) # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctlz_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -129,9 +129,9 @@ define i64 @test_ctlz_i64(i64 *%a0, i64 %a1, i64 *%a2) {
;
; HASWELL-LABEL: test_ctlz_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: movbeq (%rdi), %rax # sched: [6:0.50]
; HASWELL-NEXT: movbeq (%rdi), %rax # sched: [1:0.50]
; HASWELL-NEXT: movbeq %rsi, (%rdx) # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctlz_i64:
; BTVER2: # BB#0:
Expand Down
178 changes: 89 additions & 89 deletions llvm/test/CodeGen/X86/mul-constant-i32.ll
View file

Large diffs are not rendered by default.

136 changes: 68 additions & 68 deletions llvm/test/CodeGen/X86/mul-constant-i64.ll
View file

Large diffs are not rendered by default.

12 changes: 6 additions & 6 deletions llvm/test/CodeGen/X86/popcnt-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -37,11 +37,11 @@ define i16 @test_ctpop_i16(i16 zeroext %a0, i16 *%a1) {
;
; HASWELL-LABEL: test_ctpop_i16:
; HASWELL: # BB#0:
; HASWELL-NEXT: popcntw (%rsi), %cx # sched: [7:1.00]
; HASWELL-NEXT: popcntw (%rsi), %cx # sched: [3:1.00]
; HASWELL-NEXT: popcntw %di, %ax # sched: [3:1.00]
; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctpop_i16:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -90,10 +90,10 @@ define i32 @test_ctpop_i32(i32 %a0, i32 *%a1) {
;
; HASWELL-LABEL: test_ctpop_i32:
; HASWELL: # BB#0:
; HASWELL-NEXT: popcntl (%rsi), %ecx # sched: [7:1.00]
; HASWELL-NEXT: popcntl (%rsi), %ecx # sched: [3:1.00]
; HASWELL-NEXT: popcntl %edi, %eax # sched: [3:1.00]
; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctpop_i32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -140,10 +140,10 @@ define i64 @test_ctpop_i64(i64 %a0, i64 *%a1) {
;
; HASWELL-LABEL: test_ctpop_i64:
; HASWELL: # BB#0:
; HASWELL-NEXT: popcntq (%rsi), %rcx # sched: [7:1.00]
; HASWELL-NEXT: popcntq (%rsi), %rcx # sched: [3:1.00]
; HASWELL-NEXT: popcntq %rdi, %rax # sched: [3:1.00]
; HASWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_ctpop_i64:
; BTVER2: # BB#0:
Expand Down
68 changes: 34 additions & 34 deletions llvm/test/CodeGen/X86/pr32329.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -36,33 +36,33 @@ define void @foo() local_unnamed_addr {
; X86-NEXT: .cfi_offset %ebx, -12
; X86-NEXT: .Lcfi7:
; X86-NEXT: .cfi_offset %ebp, -8
; X86-NEXT: movl obj, %edx
; X86-NEXT: movsbl var_27, %eax
; X86-NEXT: movzwl var_2, %esi
; X86-NEXT: movl var_310, %ecx
; X86-NEXT: imull %eax, %ecx
; X86-NEXT: movl obj, %esi
; X86-NEXT: addl var_24, %ecx
; X86-NEXT: andl $4194303, %edx # imm = 0x3FFFFF
; X86-NEXT: leal (%edx,%edx), %ebx
; X86-NEXT: subl %eax, %ebx
; X86-NEXT: movl %ebx, %edi
; X86-NEXT: subl %esi, %edi
; X86-NEXT: imull %edi, %ecx
; X86-NEXT: movzwl var_2, %edi
; X86-NEXT: andl $4194303, %esi # imm = 0x3FFFFF
; X86-NEXT: leal (%esi,%esi), %edx
; X86-NEXT: subl %eax, %edx
; X86-NEXT: movl %edx, %ebx
; X86-NEXT: subl %edi, %ebx
; X86-NEXT: imull %ebx, %ecx
; X86-NEXT: addl $-1437483407, %ecx # imm = 0xAA51BE71
; X86-NEXT: movl $9, %esi
; X86-NEXT: movl $9, %edi
; X86-NEXT: xorl %ebp, %ebp
; X86-NEXT: shldl %cl, %esi, %ebp
; X86-NEXT: shlxl %ecx, %esi, %esi
; X86-NEXT: shldl %cl, %edi, %ebp
; X86-NEXT: shlxl %ecx, %edi, %edi
; X86-NEXT: testb $32, %cl
; X86-NEXT: cmovnel %esi, %ebp
; X86-NEXT: cmovnel %edi, %ebp
; X86-NEXT: movl $0, %ecx
; X86-NEXT: cmovnel %ecx, %esi
; X86-NEXT: cmpl %edx, %edi
; X86-NEXT: cmovnel %ecx, %edi
; X86-NEXT: movl %ebp, var_50+4
; X86-NEXT: movl %esi, var_50
; X86-NEXT: cmpl %esi, %ebx
; X86-NEXT: setge var_205
; X86-NEXT: imull %eax, %ebx
; X86-NEXT: movb %bl, var_218
; X86-NEXT: imull %eax, %edx
; X86-NEXT: movl %edi, var_50
; X86-NEXT: movb %dl, var_218
; X86-NEXT: popl %esi
; X86-NEXT: popl %edi
; X86-NEXT: popl %ebx
Expand All @@ -71,25 +71,25 @@ define void @foo() local_unnamed_addr {
;
; X64-LABEL: foo:
; X64: # BB#0: # %entry
; X64-NEXT: movl {{.*}}(%rip), %eax
; X64-NEXT: movsbl {{.*}}(%rip), %r9d
; X64-NEXT: movsbl {{.*}}(%rip), %eax
; X64-NEXT: movl {{.*}}(%rip), %ecx
; X64-NEXT: imull %eax, %ecx
; X64-NEXT: movl {{.*}}(%rip), %edx
; X64-NEXT: addl {{.*}}(%rip), %ecx
; X64-NEXT: movzwl {{.*}}(%rip), %r8d
; X64-NEXT: movl {{.*}}(%rip), %esi
; X64-NEXT: imull %r9d, %esi
; X64-NEXT: addl {{.*}}(%rip), %esi
; X64-NEXT: andl $4194303, %eax # imm = 0x3FFFFF
; X64-NEXT: leal (%rax,%rax), %edi
; X64-NEXT: subl %r9d, %edi
; X64-NEXT: movl %edi, %edx
; X64-NEXT: subl %r8d, %edx
; X64-NEXT: imull %edx, %esi
; X64-NEXT: addl $-1437483407, %esi # imm = 0xAA51BE71
; X64-NEXT: movl $9, %ecx
; X64-NEXT: shlxq %rsi, %rcx, %rcx
; X64-NEXT: movq %rcx, {{.*}}(%rip)
; X64-NEXT: cmpl %eax, %edx
; X64-NEXT: andl $4194303, %edx # imm = 0x3FFFFF
; X64-NEXT: leal (%rdx,%rdx), %edi
; X64-NEXT: subl %eax, %edi
; X64-NEXT: movl %edi, %esi
; X64-NEXT: subl %r8d, %esi
; X64-NEXT: imull %esi, %ecx
; X64-NEXT: addl $-1437483407, %ecx # imm = 0xAA51BE71
; X64-NEXT: movl $9, %r8d
; X64-NEXT: cmpl %edx, %esi
; X64-NEXT: setge {{.*}}(%rip)
; X64-NEXT: imull %r9d, %edi
; X64-NEXT: shlxq %rcx, %r8, %rcx
; X64-NEXT: imull %eax, %edi
; X64-NEXT: movq %rcx, {{.*}}(%rip)
; X64-NEXT: movb %dil, {{.*}}(%rip)
; X64-NEXT: retq
entry:
Expand Down
192 changes: 96 additions & 96 deletions llvm/test/CodeGen/X86/recip-fastmath.ll
View file

Large diffs are not rendered by default.

392 changes: 196 additions & 196 deletions llvm/test/CodeGen/X86/recip-fastmath2.ll
View file

Large diffs are not rendered by default.

14 changes: 7 additions & 7 deletions llvm/test/CodeGen/X86/sha-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ define <4 x i32> @test_sha1msg1(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; CANNONLAKE: # BB#0:
; CANNONLAKE-NEXT: sha1msg1 %xmm1, %xmm0
; CANNONLAKE-NEXT: sha1msg1 (%rdi), %xmm0
; CANNONLAKE-NEXT: retq # sched: [1:1.00]
; CANNONLAKE-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sha1msg1:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -56,7 +56,7 @@ define <4 x i32> @test_sha1msg2(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; CANNONLAKE: # BB#0:
; CANNONLAKE-NEXT: sha1msg2 %xmm1, %xmm0
; CANNONLAKE-NEXT: sha1msg2 (%rdi), %xmm0
; CANNONLAKE-NEXT: retq # sched: [1:1.00]
; CANNONLAKE-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sha1msg2:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -87,7 +87,7 @@ define <4 x i32> @test_sha1nexte(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; CANNONLAKE: # BB#0:
; CANNONLAKE-NEXT: sha1nexte %xmm1, %xmm0
; CANNONLAKE-NEXT: sha1nexte (%rdi), %xmm0
; CANNONLAKE-NEXT: retq # sched: [1:1.00]
; CANNONLAKE-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sha1nexte:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -118,7 +118,7 @@ define <4 x i32> @test_sha1rnds4(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; CANNONLAKE: # BB#0:
; CANNONLAKE-NEXT: sha1rnds4 $3, %xmm1, %xmm0
; CANNONLAKE-NEXT: sha1rnds4 $3, (%rdi), %xmm0
; CANNONLAKE-NEXT: retq # sched: [1:1.00]
; CANNONLAKE-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sha1rnds4:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -153,7 +153,7 @@ define <4 x i32> @test_sha256msg1(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2)
; CANNONLAKE: # BB#0:
; CANNONLAKE-NEXT: sha256msg1 %xmm1, %xmm0
; CANNONLAKE-NEXT: sha256msg1 (%rdi), %xmm0
; CANNONLAKE-NEXT: retq # sched: [1:1.00]
; CANNONLAKE-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sha256msg1:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -184,7 +184,7 @@ define <4 x i32> @test_sha256msg2(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2)
; CANNONLAKE: # BB#0:
; CANNONLAKE-NEXT: sha256msg2 %xmm1, %xmm0
; CANNONLAKE-NEXT: sha256msg2 (%rdi), %xmm0
; CANNONLAKE-NEXT: retq # sched: [1:1.00]
; CANNONLAKE-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sha256msg2:
; ZNVER1: # BB#0:
Expand Down Expand Up @@ -224,7 +224,7 @@ define <4 x i32> @test_sha256rnds2(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2,
; CANNONLAKE-NEXT: sha256rnds2 %xmm0, %xmm1, %xmm3
; CANNONLAKE-NEXT: sha256rnds2 %xmm0, (%rdi), %xmm3
; CANNONLAKE-NEXT: vmovaps %xmm3, %xmm0 # sched: [1:1.00]
; CANNONLAKE-NEXT: retq # sched: [1:1.00]
; CANNONLAKE-NEXT: retq # sched: [2:1.00]
;
; ZNVER1-LABEL: test_sha256rnds2:
; ZNVER1: # BB#0:
Expand Down
196 changes: 98 additions & 98 deletions llvm/test/CodeGen/X86/sse-schedule.ll
View file

Large diffs are not rendered by default.

484 changes: 242 additions & 242 deletions llvm/test/CodeGen/X86/sse2-schedule.ll
View file

Large diffs are not rendered by default.

46 changes: 23 additions & 23 deletions llvm/test/CodeGen/X86/sse3-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -37,8 +37,8 @@ define <2 x double> @test_addsubpd(<2 x double> %a0, <2 x double> %a1, <2 x doub
; HASWELL-LABEL: test_addsubpd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_addsubpd:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -86,8 +86,8 @@ define <4 x float> @test_addsubps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; HASWELL-LABEL: test_addsubps:
; HASWELL: # BB#0:
; HASWELL-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_addsubps:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -135,8 +135,8 @@ define <2 x double> @test_haddpd(<2 x double> %a0, <2 x double> %a1, <2 x double
; HASWELL-LABEL: test_haddpd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [9:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_haddpd:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -184,8 +184,8 @@ define <4 x float> @test_haddps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
; HASWELL-LABEL: test_haddps:
; HASWELL: # BB#0:
; HASWELL-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [9:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_haddps:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -233,8 +233,8 @@ define <2 x double> @test_hsubpd(<2 x double> %a0, <2 x double> %a1, <2 x double
; HASWELL-LABEL: test_hsubpd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [9:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_hsubpd:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -282,8 +282,8 @@ define <4 x float> @test_hsubps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
; HASWELL-LABEL: test_hsubps:
; HASWELL: # BB#0:
; HASWELL-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [9:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_hsubps:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -328,8 +328,8 @@ define <16 x i8> @test_lddqu(i8* %a0) {
;
; HASWELL-LABEL: test_lddqu:
; HASWELL: # BB#0:
; HASWELL-NEXT: vlddqu (%rdi), %xmm0 # sched: [4:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vlddqu (%rdi), %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_lddqu:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -379,7 +379,7 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; HASWELL-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
; HASWELL-NEXT: movl %esi, %ecx # sched: [1:0.25]
; HASWELL-NEXT: monitor # sched: [100:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_monitor:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -432,9 +432,9 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; HASWELL-LABEL: test_movddup:
; HASWELL: # BB#0:
; HASWELL-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
; HASWELL-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [4:0.50]
; HASWELL-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [1:0.50]
; HASWELL-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_movddup:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -489,9 +489,9 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; HASWELL-LABEL: test_movshdup:
; HASWELL: # BB#0:
; HASWELL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
; HASWELL-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [4:0.50]
; HASWELL-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [1:0.50]
; HASWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_movshdup:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -546,9 +546,9 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; HASWELL-LABEL: test_movsldup:
; HASWELL: # BB#0:
; HASWELL-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
; HASWELL-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [4:0.50]
; HASWELL-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [1:0.50]
; HASWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_movsldup:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -603,8 +603,8 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; HASWELL: # BB#0:
; HASWELL-NEXT: movl %edi, %ecx # sched: [1:0.25]
; HASWELL-NEXT: movl %esi, %eax # sched: [1:0.25]
; HASWELL-NEXT: mwait # sched: [100:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: mwait # sched: [20:2.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_mwait:
; BTVER2: # BB#0:
Expand Down
210 changes: 105 additions & 105 deletions llvm/test/CodeGen/X86/sse41-schedule.ll
View file

Large diffs are not rendered by default.

38 changes: 19 additions & 19 deletions llvm/test/CodeGen/X86/sse42-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@ define i32 @crc32_32_8(i32 %a0, i8 %a1, i8 *%a2) {
; HASWELL-NEXT: crc32b %sil, %edi # sched: [3:1.00]
; HASWELL-NEXT: crc32b (%rdx), %edi # sched: [7:1.00]
; HASWELL-NEXT: movl %edi, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: crc32_32_8:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -84,7 +84,7 @@ define i32 @crc32_32_16(i32 %a0, i16 %a1, i16 *%a2) {
; HASWELL-NEXT: crc32w %si, %edi # sched: [3:1.00]
; HASWELL-NEXT: crc32w (%rdx), %edi # sched: [7:1.00]
; HASWELL-NEXT: movl %edi, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: crc32_32_16:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -133,7 +133,7 @@ define i32 @crc32_32_32(i32 %a0, i32 %a1, i32 *%a2) {
; HASWELL-NEXT: crc32l %esi, %edi # sched: [3:1.00]
; HASWELL-NEXT: crc32l (%rdx), %edi # sched: [7:1.00]
; HASWELL-NEXT: movl %edi, %eax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: crc32_32_32:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -182,7 +182,7 @@ define i64 @crc32_64_8(i64 %a0, i8 %a1, i8 *%a2) nounwind {
; HASWELL-NEXT: crc32b %sil, %edi # sched: [3:1.00]
; HASWELL-NEXT: crc32b (%rdx), %edi # sched: [7:1.00]
; HASWELL-NEXT: movq %rdi, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: crc32_64_8:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -231,7 +231,7 @@ define i64 @crc32_64_64(i64 %a0, i64 %a1, i64 *%a2) {
; HASWELL-NEXT: crc32q %rsi, %rdi # sched: [3:1.00]
; HASWELL-NEXT: crc32q (%rdx), %rdi # sched: [7:1.00]
; HASWELL-NEXT: movq %rdi, %rax # sched: [1:0.25]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: crc32_64_64:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -297,14 +297,14 @@ define i32 @test_pcmpestri(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; HASWELL: # BB#0:
; HASWELL-NEXT: movl $7, %eax # sched: [1:0.25]
; HASWELL-NEXT: movl $7, %edx # sched: [1:0.25]
; HASWELL-NEXT: vpcmpestri $7, %xmm1, %xmm0 # sched: [11:3.00]
; HASWELL-NEXT: vpcmpestri $7, %xmm1, %xmm0 # sched: [18:4.00]
; HASWELL-NEXT: movl %ecx, %esi # sched: [1:0.25]
; HASWELL-NEXT: movl $7, %eax # sched: [1:0.25]
; HASWELL-NEXT: movl $7, %edx # sched: [1:0.25]
; HASWELL-NEXT: vpcmpestri $7, (%rdi), %xmm0 # sched: [11:3.00]
; HASWELL-NEXT: vpcmpestri $7, (%rdi), %xmm0 # sched: [18:4.00]
; HASWELL-NEXT: # kill: %ECX<def> %ECX<kill> %RCX<def>
; HASWELL-NEXT: leal (%rcx,%rsi), %eax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pcmpestri:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -374,11 +374,11 @@ define <16 x i8> @test_pcmpestrm(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; HASWELL: # BB#0:
; HASWELL-NEXT: movl $7, %eax # sched: [1:0.25]
; HASWELL-NEXT: movl $7, %edx # sched: [1:0.25]
; HASWELL-NEXT: vpcmpestrm $7, %xmm1, %xmm0 # sched: [10:4.00]
; HASWELL-NEXT: vpcmpestrm $7, %xmm1, %xmm0 # sched: [19:4.00]
; HASWELL-NEXT: movl $7, %eax # sched: [1:0.25]
; HASWELL-NEXT: movl $7, %edx # sched: [1:0.25]
; HASWELL-NEXT: vpcmpestrm $7, (%rdi), %xmm0 # sched: [10:3.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpcmpestrm $7, (%rdi), %xmm0 # sched: [19:4.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pcmpestrm:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -441,7 +441,7 @@ define i32 @test_pcmpistri(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; HASWELL-NEXT: vpcmpistri $7, (%rdi), %xmm0 # sched: [11:3.00]
; HASWELL-NEXT: # kill: %ECX<def> %ECX<kill> %RCX<def>
; HASWELL-NEXT: leal (%rcx,%rax), %eax # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pcmpistri:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -489,9 +489,9 @@ define <16 x i8> @test_pcmpistrm(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
;
; HASWELL-LABEL: test_pcmpistrm:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpcmpistrm $7, %xmm1, %xmm0 # sched: [10:3.00]
; HASWELL-NEXT: vpcmpistrm $7, (%rdi), %xmm0 # sched: [10:3.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpcmpistrm $7, %xmm1, %xmm0 # sched: [11:3.00]
; HASWELL-NEXT: vpcmpistrm $7, (%rdi), %xmm0 # sched: [11:3.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pcmpistrm:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -534,7 +534,7 @@ define <2 x i64> @test_pcmpgtq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
; HASWELL: # BB#0:
; HASWELL-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; HASWELL-NEXT: vpcmpgtq (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pcmpgtq:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -576,9 +576,9 @@ define <2 x i64> @test_pclmulqdq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
;
; HASWELL-LABEL: test_pclmulqdq:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0 # sched: [7:2.00]
; HASWELL-NEXT: vpclmulqdq $0, (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0 # sched: [11:2.00]
; HASWELL-NEXT: vpclmulqdq $0, (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pclmulqdq:
; BTVER2: # BB#0:
Expand Down
64 changes: 32 additions & 32 deletions llvm/test/CodeGen/X86/ssse3-schedule.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -42,9 +42,9 @@ define <16 x i8> @test_pabsb(<16 x i8> %a0, <16 x i8> *%a1) {
; HASWELL-LABEL: test_pabsb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsb %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: vpabsb (%rdi), %xmm1 # sched: [5:0.50]
; HASWELL-NEXT: vpabsb (%rdi), %xmm1 # sched: [1:0.50]
; HASWELL-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pabsb:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -100,9 +100,9 @@ define <4 x i32> @test_pabsd(<4 x i32> %a0, <4 x i32> *%a1) {
; HASWELL-LABEL: test_pabsd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsd %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: vpabsd (%rdi), %xmm1 # sched: [5:0.50]
; HASWELL-NEXT: vpabsd (%rdi), %xmm1 # sched: [1:0.50]
; HASWELL-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pabsd:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -158,9 +158,9 @@ define <8 x i16> @test_pabsw(<8 x i16> %a0, <8 x i16> *%a1) {
; HASWELL-LABEL: test_pabsw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpabsw %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: vpabsw (%rdi), %xmm1 # sched: [5:0.50]
; HASWELL-NEXT: vpabsw (%rdi), %xmm1 # sched: [1:0.50]
; HASWELL-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pabsw:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -216,8 +216,8 @@ define <8 x i16> @test_palignr(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; HASWELL-LABEL: test_palignr:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5] sched: [1:1.00]
; HASWELL-NEXT: vpalignr {{.*#+}} xmm0 = mem[14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13] sched: [5:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpalignr {{.*#+}} xmm0 = mem[14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13] sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_palignr:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -264,8 +264,8 @@ define <4 x i32> @test_phaddd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; HASWELL-LABEL: test_phaddd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vphaddd %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: vphaddd (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vphaddd (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_phaddd:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -313,8 +313,8 @@ define <8 x i16> @test_phaddsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; HASWELL-LABEL: test_phaddsw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vphaddsw %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: vphaddsw (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vphaddsw (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_phaddsw:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -362,8 +362,8 @@ define <8 x i16> @test_phaddw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; HASWELL-LABEL: test_phaddw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vphaddw %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: vphaddw (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vphaddw (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_phaddw:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -411,8 +411,8 @@ define <4 x i32> @test_phsubd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; HASWELL-LABEL: test_phsubd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vphsubd %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: vphsubd (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vphsubd (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_phsubd:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -460,8 +460,8 @@ define <8 x i16> @test_phsubsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; HASWELL-LABEL: test_phsubsw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vphsubsw %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: vphsubsw (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vphsubsw (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_phsubsw:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -509,8 +509,8 @@ define <8 x i16> @test_phsubw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; HASWELL-LABEL: test_phsubw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vphsubw %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: vphsubw (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vphsubw (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_phsubw:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -558,8 +558,8 @@ define <8 x i16> @test_pmaddubsw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; HASWELL-LABEL: test_pmaddubsw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; HASWELL-NEXT: vpmaddubsw (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpmaddubsw (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pmaddubsw:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -608,8 +608,8 @@ define <8 x i16> @test_pmulhrsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; HASWELL-LABEL: test_pmulhrsw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpmulhrsw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; HASWELL-NEXT: vpmulhrsw (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpmulhrsw (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pmulhrsw:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -657,8 +657,8 @@ define <16 x i8> @test_pshufb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; HASWELL-LABEL: test_pshufb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpshufb %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; HASWELL-NEXT: vpshufb (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpshufb (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_pshufb:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -710,8 +710,8 @@ define <16 x i8> @test_psignb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; HASWELL-LABEL: test_psignb:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsignb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: vpsignb (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpsignb (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_psignb:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -763,8 +763,8 @@ define <4 x i32> @test_psignd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; HASWELL-LABEL: test_psignd:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsignd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: vpsignd (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpsignd (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_psignd:
; BTVER2: # BB#0:
Expand Down Expand Up @@ -816,8 +816,8 @@ define <8 x i16> @test_psignw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; HASWELL-LABEL: test_psignw:
; HASWELL: # BB#0:
; HASWELL-NEXT: vpsignw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: vpsignw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; HASWELL-NEXT: retq # sched: [1:1.00]
; HASWELL-NEXT: vpsignw (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; HASWELL-NEXT: retq # sched: [2:1.00]
;
; BTVER2-LABEL: test_psignw:
; BTVER2: # BB#0:
Expand Down
12 changes: 6 additions & 6 deletions llvm/test/CodeGen/X86/vector-shift-ashr-512.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -201,14 +201,14 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
; AVX512DQ-NEXT: vpsraw $2, %ymm0, %ymm5
; AVX512DQ-NEXT: vpaddw %ymm2, %ymm2, %ymm8
; AVX512DQ-NEXT: vpblendvb %ymm8, %ymm5, %ymm0, %ymm0
; AVX512DQ-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
; AVX512DQ-NEXT: vpsraw $4, %ymm5, %ymm9
; AVX512DQ-NEXT: vpblendvb %ymm3, %ymm9, %ymm5, %ymm3
; AVX512DQ-NEXT: vpsraw $1, %ymm0, %ymm5
; AVX512DQ-NEXT: vpaddw %ymm8, %ymm8, %ymm9
; AVX512DQ-NEXT: vpblendvb %ymm9, %ymm5, %ymm0, %ymm0
; AVX512DQ-NEXT: vpsrlw $8, %ymm0, %ymm0
; AVX512DQ-NEXT: vpackuswb %ymm4, %ymm0, %ymm0
; AVX512DQ-NEXT: vpunpckhbw {{.*#+}} ymm4 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
; AVX512DQ-NEXT: vpsraw $4, %ymm4, %ymm5
; AVX512DQ-NEXT: vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
; AVX512DQ-NEXT: vpsraw $2, %ymm3, %ymm4
; AVX512DQ-NEXT: vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
; AVX512DQ-NEXT: vpsraw $1, %ymm3, %ymm4
Expand Down Expand Up @@ -328,14 +328,14 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) nounwind {
; AVX512DQ-NEXT: vpsraw $2, %ymm0, %ymm5
; AVX512DQ-NEXT: vpaddw %ymm2, %ymm2, %ymm8
; AVX512DQ-NEXT: vpblendvb %ymm8, %ymm5, %ymm0, %ymm0
; AVX512DQ-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
; AVX512DQ-NEXT: vpsraw $4, %ymm5, %ymm9
; AVX512DQ-NEXT: vpblendvb %ymm3, %ymm9, %ymm5, %ymm3
; AVX512DQ-NEXT: vpsraw $1, %ymm0, %ymm5
; AVX512DQ-NEXT: vpaddw %ymm8, %ymm8, %ymm9
; AVX512DQ-NEXT: vpblendvb %ymm9, %ymm5, %ymm0, %ymm0
; AVX512DQ-NEXT: vpsrlw $8, %ymm0, %ymm0
; AVX512DQ-NEXT: vpackuswb %ymm4, %ymm0, %ymm0
; AVX512DQ-NEXT: vpunpckhbw {{.*#+}} ymm4 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
; AVX512DQ-NEXT: vpsraw $4, %ymm4, %ymm5
; AVX512DQ-NEXT: vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
; AVX512DQ-NEXT: vpsraw $2, %ymm3, %ymm4
; AVX512DQ-NEXT: vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
; AVX512DQ-NEXT: vpsraw $1, %ymm3, %ymm4
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/X86/vector-shift-lshr-256.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -777,9 +777,9 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
;
; AVX512DQ-LABEL: splatvar_shift_v32i8:
; AVX512DQ: # BB#0:
; AVX512DQ-NEXT: vpbroadcastb %xmm1, %ymm1
; AVX512DQ-NEXT: vpsrlw $4, %ymm0, %ymm2
; AVX512DQ-NEXT: vpand {{.*}}(%rip), %ymm2, %ymm2
; AVX512DQ-NEXT: vpbroadcastb %xmm1, %ymm1
; AVX512DQ-NEXT: vpsllw $5, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
; AVX512DQ-NEXT: vpsrlw $2, %ymm0, %ymm2
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/X86/vector-shift-shl-256.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -713,9 +713,9 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
;
; AVX512DQ-LABEL: splatvar_shift_v32i8:
; AVX512DQ: # BB#0:
; AVX512DQ-NEXT: vpbroadcastb %xmm1, %ymm1
; AVX512DQ-NEXT: vpsllw $4, %ymm0, %ymm2
; AVX512DQ-NEXT: vpand {{.*}}(%rip), %ymm2, %ymm2
; AVX512DQ-NEXT: vpbroadcastb %xmm1, %ymm1
; AVX512DQ-NEXT: vpsllw $5, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
; AVX512DQ-NEXT: vpsllw $2, %ymm0, %ymm2
Expand Down
12 changes: 6 additions & 6 deletions llvm/test/CodeGen/X86/vector-shuffle-512-v32.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -68,13 +68,13 @@ define <32 x i16> @shuffle_v32i16_0f_1f_0e_16_0d_1d_04_1e_0b_1b_0a_1a_09_19_08_1
; KNL-NEXT: vpshufb {{.*#+}} xmm4 = xmm1[8,9,12,13,12,13,10,11,0,1,4,5,4,5,0,1]
; KNL-NEXT: vpshufd {{.*#+}} xmm1 = xmm2[0,1,0,3]
; KNL-NEXT: vpshuflw {{.*#+}} xmm2 = xmm1[0,3,2,2,4,5,6,7]
; KNL-NEXT: vinserti128 $1, %xmm2, %ymm4, %ymm1
; KNL-NEXT: vextracti128 $1, %ymm0, %xmm5
; KNL-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
; KNL-NEXT: vextracti128 $1, %ymm0, %xmm1
; KNL-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; KNL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[12,13,10,11,8,9,14,15,4,5,2,3,2,3,6,7]
; KNL-NEXT: vpshufb {{.*#+}} xmm5 = xmm5[6,7,2,3,4,5,6,7,2,3,2,3,0,1,14,15]
; KNL-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm0
; KNL-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7],ymm0[8],ymm1[9],ymm0[10],ymm1[11],ymm0[12],ymm1[13],ymm0[14],ymm1[15]
; KNL-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[6,7,2,3,4,5,6,7,2,3,2,3,0,1,14,15]
; KNL-NEXT: vinserti128 $1, %xmm2, %ymm4, %ymm5
; KNL-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; KNL-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm5[1],ymm0[2],ymm5[3],ymm0[4],ymm5[5],ymm0[6],ymm5[7],ymm0[8],ymm5[9],ymm0[10],ymm5[11],ymm0[12],ymm5[13],ymm0[14],ymm5[15]
; KNL-NEXT: vextracti128 $1, %ymm3, %xmm3
; KNL-NEXT: vpbroadcastw %xmm3, %ymm3
; KNL-NEXT: vmovdqa {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
Expand Down