88 changes: 44 additions & 44 deletions llvm/test/CodeGen/X86/avx512fp16-mov.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -178,7 +178,7 @@ define void @test5(half %x, ptr %y) {
;
; X86-LABEL: test5:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: vmovsh %xmm0, (%eax)
; X86-NEXT: retl
Expand All @@ -189,13 +189,13 @@ define void @test5(half %x, ptr %y) {
define half @test7(ptr %x) {
; X64-LABEL: test7:
; X64: # %bb.0:
; X64-NEXT: vmovsh (%rdi), %xmm0
; X64-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X64-NEXT: retq
;
; X86-LABEL: test7:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: vmovsh (%eax), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: retl
%y = load i16, ptr %x
%res = bitcast i16 %y to half
Expand Down Expand Up @@ -253,13 +253,13 @@ define <32 x i16> @test10c(ptr %x) {
define <8 x half> @test11(ptr %x) {
; X64-LABEL: test11:
; X64: # %bb.0:
; X64-NEXT: vmovsh (%rdi), %xmm0
; X64-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X64-NEXT: retq
;
; X86-LABEL: test11:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: vmovsh (%eax), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: retl
%y = load half, ptr %x, align 2
%res = insertelement <8 x half>zeroinitializer, half %y, i32 0
Expand All @@ -269,13 +269,13 @@ define <8 x half> @test11(ptr %x) {
define <16 x half> @test11b(ptr %x) {
; X64-LABEL: test11b:
; X64: # %bb.0:
; X64-NEXT: vmovsh (%rdi), %xmm0
; X64-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X64-NEXT: retq
;
; X86-LABEL: test11b:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: vmovsh (%eax), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: retl
%y = load half, ptr %x, align 2
%res = insertelement <16 x half>zeroinitializer, half %y, i32 0
Expand All @@ -285,13 +285,13 @@ define <16 x half> @test11b(ptr %x) {
define <32 x half> @test11c(ptr %x) {
; X64-LABEL: test11c:
; X64: # %bb.0:
; X64-NEXT: vmovsh (%rdi), %xmm0
; X64-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X64-NEXT: retq
;
; X86-LABEL: test11c:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: vmovsh (%eax), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: retl
%y = load half, ptr %x, align 2
%res = insertelement <32 x half>zeroinitializer, half %y, i32 0
Expand All @@ -307,7 +307,7 @@ define <8 x half> @test14(half %x) {
;
; X86-LABEL: test14:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: retl
%res = insertelement <8 x half>zeroinitializer, half %x, i32 0
ret <8 x half>%res
Expand All @@ -322,7 +322,7 @@ define <16 x half> @test14b(half %x) {
;
; X86-LABEL: test14b:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: retl
%res = insertelement <16 x half>zeroinitializer, half %x, i32 0
ret <16 x half>%res
Expand All @@ -337,7 +337,7 @@ define <32 x half> @test14c(half %x) {
;
; X86-LABEL: test14c:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: retl
%res = insertelement <32 x half>zeroinitializer, half %x, i32 0
ret <32 x half>%res
Expand Down Expand Up @@ -1253,7 +1253,7 @@ define half @test_movw2(i16 %x) {
;
; X86-LABEL: test_movw2:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: retl
%res = bitcast i16 %x to half
ret half %res
Expand Down Expand Up @@ -1358,7 +1358,7 @@ define half @extract_f16_8(<32 x half> %x, i64 %idx) nounwind {
; X64-NEXT: subq $128, %rsp
; X64-NEXT: andl $31, %edi
; X64-NEXT: vmovaps %zmm0, (%rsp)
; X64-NEXT: vmovsh (%rsp,%rdi,2), %xmm0
; X64-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X64-NEXT: movq %rbp, %rsp
; X64-NEXT: popq %rbp
; X64-NEXT: vzeroupper
Expand All @@ -1373,7 +1373,7 @@ define half @extract_f16_8(<32 x half> %x, i64 %idx) nounwind {
; X86-NEXT: movl 8(%ebp), %eax
; X86-NEXT: andl $31, %eax
; X86-NEXT: vmovaps %zmm0, (%esp)
; X86-NEXT: vmovsh (%esp,%eax,2), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: movl %ebp, %esp
; X86-NEXT: popl %ebp
; X86-NEXT: vzeroupper
Expand All @@ -1392,7 +1392,7 @@ define half @extract_f16_9(<64 x half> %x, i64 %idx) nounwind {
; X64-NEXT: andl $63, %edi
; X64-NEXT: vmovaps %zmm1, {{[0-9]+}}(%rsp)
; X64-NEXT: vmovaps %zmm0, (%rsp)
; X64-NEXT: vmovsh (%rsp,%rdi,2), %xmm0
; X64-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X64-NEXT: movq %rbp, %rsp
; X64-NEXT: popq %rbp
; X64-NEXT: vzeroupper
Expand All @@ -1408,7 +1408,7 @@ define half @extract_f16_9(<64 x half> %x, i64 %idx) nounwind {
; X86-NEXT: andl $63, %eax
; X86-NEXT: vmovaps %zmm1, {{[0-9]+}}(%esp)
; X86-NEXT: vmovaps %zmm0, (%esp)
; X86-NEXT: vmovsh (%esp,%eax,2), %xmm0
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: movl %ebp, %esp
; X86-NEXT: popl %ebp
; X86-NEXT: vzeroupper
Expand Down Expand Up @@ -1797,11 +1797,11 @@ define <8 x half> @build_vector_xxxxuuuu(half %a0, half %a1, half %a2, half %a3)
;
; X86-LABEL: build_vector_xxxxuuuu:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm2
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm2 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
; X86-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
; X86-NEXT: retl
Expand All @@ -1823,11 +1823,11 @@ define <8 x half> @build_vector_uuuuxxxx(half %a0, half %a1, half %a2, half %a3)
;
; X86-LABEL: build_vector_uuuuxxxx:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm2
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm2 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
; X86-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; X86-NEXT: vpbroadcastq %xmm0, %xmm0
Expand All @@ -1853,18 +1853,18 @@ define <8 x half> @build_vector_xxxxxxxx(half %a0, half %a1, half %a2, half %a3,
;
; X86-LABEL: build_vector_xxxxxxxx:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm2
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm2 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
; X86-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm2
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm2 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm2
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm3
; X86-NEXT: vmovsh {{.*#+}} xmm2 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm3 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
; X86-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; X86-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
Expand Down Expand Up @@ -1895,18 +1895,18 @@ define <16 x half> @build_vector_xxxxuuuuuuuuxxxx(half %a0, half %a1, half %a2,
;
; X86-LABEL: build_vector_xxxxuuuuuuuuxxxx:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm2
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm2 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
; X86-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm2
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm2 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm2
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm3
; X86-NEXT: vmovsh {{.*#+}} xmm2 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm3 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
; X86-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0],xmm2[0],zero,zero
; X86-NEXT: vpbroadcastq %xmm0, %xmm0
Expand Down Expand Up @@ -2006,10 +2006,10 @@ define <8 x half> @test21(half %a, half %b, half %c) nounwind {
;
; X86-LABEL: test21:
; X86: # %bb.0:
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; X86-NEXT: vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT: vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; X86-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X86-NEXT: vpbroadcastw %xmm1, %xmm1
Expand Down
8 changes: 4 additions & 4 deletions llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-zext.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -221,15 +221,15 @@ define <16 x i8> @ext_i16_16i8(i16 %a0) {
; AVX512F-LABEL: ext_i16_16i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: kmovw %edi, %k1
; AVX512F-NEXT: vpbroadcastd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0 {%k1} {z}
; AVX512F-NEXT: vpbroadcastd {{.*#+}} zmm0 {%k1} {z} = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
; AVX512VLBW-LABEL: ext_i16_16i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: kmovd %edi, %k1
; AVX512VLBW-NEXT: vmovdqu8 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1} {z}
; AVX512VLBW-NEXT: vmovdqu8 {{.*#+}} xmm0 {%k1} {z} = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512VLBW-NEXT: retq
%1 = bitcast i16 %a0 to <16 x i1>
%2 = zext <16 x i1> %1 to <16 x i8>
Expand Down Expand Up @@ -497,7 +497,7 @@ define <32 x i8> @ext_i32_32i8(i32 %a0) {
; AVX512VLBW-LABEL: ext_i32_32i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: kmovd %edi, %k1
; AVX512VLBW-NEXT: vmovdqu8 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0 {%k1} {z}
; AVX512VLBW-NEXT: vmovdqu8 {{.*#+}} ymm0 {%k1} {z} = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512VLBW-NEXT: retq
%1 = bitcast i32 %a0 to <32 x i1>
%2 = zext <32 x i1> %1 to <32 x i8>
Expand Down Expand Up @@ -885,7 +885,7 @@ define <64 x i8> @ext_i64_64i8(i64 %a0) {
; AVX512VLBW-LABEL: ext_i64_64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: kmovq %rdi, %k1
; AVX512VLBW-NEXT: vmovdqu8 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0 {%k1} {z}
; AVX512VLBW-NEXT: vmovdqu8 {{.*#+}} zmm0 {%k1} {z} = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512VLBW-NEXT: retq
%1 = bitcast i64 %a0 to <64 x i1>
%2 = zext <64 x i1> %1 to <64 x i8>
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/X86/combine-sdiv.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -95,7 +95,7 @@ define <4 x i32> @combine_vec_sdiv_by_minsigned(<4 x i32> %x) {
; AVX512BW-LABEL: combine_vec_sdiv_by_minsigned:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %k1
; AVX512BW-NEXT: vpbroadcastd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1} {z}
; AVX512BW-NEXT: vpbroadcastd {{.*#+}} xmm0 {%k1} {z} = [1,1,1,1]
; AVX512BW-NEXT: retq
;
; XOP-LABEL: combine_vec_sdiv_by_minsigned:
Expand Down
16 changes: 8 additions & 8 deletions llvm/test/CodeGen/X86/divrem-by-select.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -161,7 +161,7 @@ define <2 x i64> @udiv_indentity_zero(<2 x i1> %c, <2 x i64> %x) {
; CHECK-X64-V4-NEXT: vpsllq $63, %xmm0, %xmm0
; CHECK-X64-V4-NEXT: vpmovq2m %xmm0, %k0
; CHECK-X64-V4-NEXT: knotw %k0, %k1
; CHECK-X64-V4-NEXT: vpbroadcastq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1} {z}
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 {%k1} {z} = [1,1]
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm1, %rax
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm0, %rcx
; CHECK-X64-V4-NEXT: xorl %edx, %edx
Expand Down Expand Up @@ -206,7 +206,7 @@ define <2 x i64> @udiv_indentity_partial_zero(<2 x i1> %c, <2 x i64> %x) {
; CHECK-X64-V4-NEXT: vpsllq $63, %xmm0, %xmm0
; CHECK-X64-V4-NEXT: vpmovq2m %xmm0, %k1
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 = [1,1]
; CHECK-X64-V4-NEXT: vmovdqa64 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; CHECK-X64-V4-NEXT: vmovdqa64 {{.*#+}} xmm0 {%k1} = [0,0,0,0,0,0,0,0,5,0,0,0,0,0,0,0]
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm0, %rcx
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm1, %rax
; CHECK-X64-V4-NEXT: xorl %edx, %edx
Expand Down Expand Up @@ -251,7 +251,7 @@ define <2 x i64> @urem_identity_const(<2 x i1> %c, <2 x i64> %x) {
; CHECK-X64-V4-NEXT: vpsllq $63, %xmm0, %xmm0
; CHECK-X64-V4-NEXT: vpmovq2m %xmm0, %k1
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 = [11,11]
; CHECK-X64-V4-NEXT: vpbroadcastq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 {%k1} = [1,1]
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm0, %rcx
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm1, %rax
; CHECK-X64-V4-NEXT: xorl %edx, %edx
Expand Down Expand Up @@ -296,7 +296,7 @@ define <2 x i64> @sdiv_identity_const(<2 x i1> %c, <2 x i64> %x) {
; CHECK-X64-V4-NEXT: vpsllq $63, %xmm0, %xmm0
; CHECK-X64-V4-NEXT: vpmovq2m %xmm0, %k1
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 = [1,1]
; CHECK-X64-V4-NEXT: vmovdqa64 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; CHECK-X64-V4-NEXT: vmovdqa64 {{.*#+}} xmm0 {%k1} = [11,13]
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm0, %rcx
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm1, %rax
; CHECK-X64-V4-NEXT: cqto
Expand Down Expand Up @@ -341,7 +341,7 @@ define <2 x i64> @sdiv_identity_const_todo_better_nonzero(<2 x i1> %c, <2 x i64>
; CHECK-X64-V4-NEXT: vpsllq $63, %xmm0, %xmm0
; CHECK-X64-V4-NEXT: vpmovq2m %xmm0, %k1
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 = [1,1]
; CHECK-X64-V4-NEXT: vmovdqa64 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; CHECK-X64-V4-NEXT: vmovdqa64 {{.*#+}} xmm0 {%k1} = [11,17]
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm0, %rcx
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm1, %rax
; CHECK-X64-V4-NEXT: cqto
Expand Down Expand Up @@ -386,7 +386,7 @@ define <2 x i64> @srem_identity_const(<2 x i1> %c, <2 x i64> %x) {
; CHECK-X64-V4-NEXT: vpsllq $63, %xmm0, %xmm0
; CHECK-X64-V4-NEXT: vpmovq2m %xmm0, %k1
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 = [11,11]
; CHECK-X64-V4-NEXT: vpbroadcastq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 {%k1} = [1,1]
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm0, %rcx
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm1, %rax
; CHECK-X64-V4-NEXT: cqto
Expand Down Expand Up @@ -436,7 +436,7 @@ define <2 x i64> @udivrem_identity_const(<2 x i1> %c, <2 x i64> %x) {
; CHECK-X64-V4-NEXT: vpsllq $63, %xmm0, %xmm0
; CHECK-X64-V4-NEXT: vpmovq2m %xmm0, %k1
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 = [1,1]
; CHECK-X64-V4-NEXT: vpbroadcastq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 {%k1} = [11,11]
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm0, %rcx
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm1, %rax
; CHECK-X64-V4-NEXT: xorl %edx, %edx
Expand Down Expand Up @@ -493,7 +493,7 @@ define <2 x i64> @sdivrem_identity_const(<2 x i1> %c, <2 x i64> %x) {
; CHECK-X64-V4-NEXT: vpsllq $63, %xmm0, %xmm0
; CHECK-X64-V4-NEXT: vpmovq2m %xmm0, %k1
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 = [1,1]
; CHECK-X64-V4-NEXT: vpbroadcastq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; CHECK-X64-V4-NEXT: vpbroadcastq {{.*#+}} xmm0 {%k1} = [11,11]
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm0, %rcx
; CHECK-X64-V4-NEXT: vpextrq $1, %xmm1, %rax
; CHECK-X64-V4-NEXT: cqto
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/X86/fp128-cast.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1005,7 +1005,7 @@ define dso_local i32 @TestConst128(fp128 %v) nounwind {
; X64-SSE-LABEL: TestConst128:
; X64-SSE: # %bb.0: # %entry
; X64-SSE-NEXT: pushq %rax
; X64-SSE-NEXT: movaps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; X64-SSE-NEXT: movaps {{.*#+}} xmm1 = [1.0E+0]
; X64-SSE-NEXT: callq __gttf2@PLT
; X64-SSE-NEXT: xorl %ecx, %ecx
; X64-SSE-NEXT: testl %eax, %eax
Expand Down Expand Up @@ -1037,7 +1037,7 @@ define dso_local i32 @TestConst128(fp128 %v) nounwind {
; X64-AVX-LABEL: TestConst128:
; X64-AVX: # %bb.0: # %entry
; X64-AVX-NEXT: pushq %rax
; X64-AVX-NEXT: vmovaps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; X64-AVX-NEXT: vmovaps {{.*#+}} xmm1 = [1.0E+0]
; X64-AVX-NEXT: callq __gttf2@PLT
; X64-AVX-NEXT: xorl %ecx, %ecx
; X64-AVX-NEXT: testl %eax, %eax
Expand Down
12 changes: 6 additions & 6 deletions llvm/test/CodeGen/X86/fp128-i128.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -132,27 +132,27 @@ define fp128 @TestI128_1(fp128 %x) #0 {
; SSE: # %bb.0: # %entry
; SSE-NEXT: pushq %rax
; SSE-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SSE-NEXT: movaps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; SSE-NEXT: movaps {{.*#+}} xmm1 = [1.00000000000000000000000000000000005E-1]
; SSE-NEXT: callq __lttf2@PLT
; SSE-NEXT: xorl %ecx, %ecx
; SSE-NEXT: testl %eax, %eax
; SSE-NEXT: sets %cl
; SSE-NEXT: shll $4, %ecx
; SSE-NEXT: movaps {{\.?LCPI[0-9]+_[0-9]+}}(%rcx), %xmm0
; SSE-NEXT: movaps {{.*#+}} xmm0 = [?]
; SSE-NEXT: popq %rax
; SSE-NEXT: retq
;
; AVX-LABEL: TestI128_1:
; AVX: # %bb.0: # %entry
; AVX-NEXT: pushq %rax
; AVX-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; AVX-NEXT: vmovaps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; AVX-NEXT: vmovaps {{.*#+}} xmm1 = [1.00000000000000000000000000000000005E-1]
; AVX-NEXT: callq __lttf2@PLT
; AVX-NEXT: xorl %ecx, %ecx
; AVX-NEXT: testl %eax, %eax
; AVX-NEXT: sets %cl
; AVX-NEXT: shll $4, %ecx
; AVX-NEXT: vmovaps {{\.?LCPI[0-9]+_[0-9]+}}(%rcx), %xmm0
; AVX-NEXT: vmovaps {{.*#+}} xmm0 = [?]
; AVX-NEXT: popq %rax
; AVX-NEXT: retq
entry:
Expand Down Expand Up @@ -224,7 +224,7 @@ define fp128 @TestI128_3(fp128 %x, ptr nocapture readnone %ex) #0 {
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rcx
; SSE-NEXT: jmp .LBB4_3
; SSE-NEXT: .LBB4_2: # %if.then
; SSE-NEXT: movaps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; SSE-NEXT: movaps {{.*#+}} xmm1 = [5.36312317197703883982960999928233845E+154]
; SSE-NEXT: callq __multf3@PLT
; SSE-NEXT: movaps %xmm0, {{[0-9]+}}(%rsp)
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rcx
Expand All @@ -251,7 +251,7 @@ define fp128 @TestI128_3(fp128 %x, ptr nocapture readnone %ex) #0 {
; AVX-NEXT: movq {{[0-9]+}}(%rsp), %rcx
; AVX-NEXT: jmp .LBB4_3
; AVX-NEXT: .LBB4_2: # %if.then
; AVX-NEXT: vmovaps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; AVX-NEXT: vmovaps {{.*#+}} xmm1 = [5.36312317197703883982960999928233845E+154]
; AVX-NEXT: callq __multf3@PLT
; AVX-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp)
; AVX-NEXT: movq {{[0-9]+}}(%rsp), %rcx
Expand Down
2 changes: 2 additions & 0 deletions llvm/test/CodeGen/X86/stack-folding-fp-avx512fp16.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -727,6 +727,7 @@ define half @stack_fold_maxsh_commuted(half %a0, half %a1) #0 {
; CHECK-NEXT: nop
; CHECK-NEXT: #NO_APP
; CHECK-NEXT: vmovsh {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 4-byte Reload
; CHECK-NEXT: # xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; CHECK-NEXT: vmaxsh %xmm0, %xmm1, %xmm0
; CHECK-NEXT: retq
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
Expand Down Expand Up @@ -1044,6 +1045,7 @@ define half @stack_fold_minsh_commuted(half %a0, half %a1) #0 {
; CHECK-NEXT: nop
; CHECK-NEXT: #NO_APP
; CHECK-NEXT: vmovsh {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 4-byte Reload
; CHECK-NEXT: # xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; CHECK-NEXT: vminsh %xmm0, %xmm1, %xmm0
; CHECK-NEXT: retq
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/X86/var-permute-256.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1239,7 +1239,7 @@ define <4 x i64> @PR50356(<4 x i64> %0, <4 x i32> %1, <4 x i64> %2) unnamed_addr
; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512VL-NEXT: vpcmpgtq %ymm0, %ymm2, %k1
; AVX512VL-NEXT: vpmovzxbq {{.*#+}} ymm0 = [34,68,102,136]
; AVX512VL-NEXT: vmovdqa64 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0 {%k1}
; AVX512VL-NEXT: vmovdqa64 {{.*#+}} ymm0 {%k1} = [17,51,85,119]
; AVX512VL-NEXT: movq %rbp, %rsp
; AVX512VL-NEXT: popq %rbp
; AVX512VL-NEXT: retq
Expand Down
6 changes: 3 additions & 3 deletions llvm/test/CodeGen/X86/vselect-zero.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -160,7 +160,7 @@ define double @fsel_nonzero_true_val(double %x, double %y, double %z) {
; AVX512-LABEL: fsel_nonzero_true_val:
; AVX512: # %bb.0:
; AVX512-NEXT: vcmpeqsd %xmm1, %xmm0, %k1
; AVX512-NEXT: vmovsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2 {%k1}
; AVX512-NEXT: vmovsd {{.*#+}} xmm2 {%k1} = [4.2E+1,0.0E+0]
; AVX512-NEXT: vmovapd %xmm2, %xmm0
; AVX512-NEXT: retq
%cond = fcmp oeq double %x, %y
Expand Down Expand Up @@ -189,7 +189,7 @@ define double @fsel_nonzero_constants(double %x, double %y) {
; AVX512: # %bb.0:
; AVX512-NEXT: vcmpeqsd %xmm1, %xmm0, %k1
; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = [4.2E+1,0.0E+0]
; AVX512-NEXT: vmovsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; AVX512-NEXT: vmovsd {{.*#+}} xmm0 {%k1} = [1.2E+1,0.0E+0]
; AVX512-NEXT: retq
%cond = fcmp oeq double %x, %y
%r = select i1 %cond, double 12.0, double 42.0
Expand Down Expand Up @@ -227,7 +227,7 @@ define <2 x double> @vsel_nonzero_constants(<2 x double> %x, <2 x double> %y) {
; AVX512: # %bb.0:
; AVX512-NEXT: vcmplepd %xmm0, %xmm1, %k1
; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = [4.2E+1,0.0E+0]
; AVX512-NEXT: vmovapd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
; AVX512-NEXT: vmovapd {{.*#+}} xmm0 {%k1} = [1.2E+1,-1.0E+0]
; AVX512-NEXT: retq
%cond = fcmp oge <2 x double> %x, %y
%r = select <2 x i1> %cond, <2 x double> <double 12.0, double -1.0>, <2 x double> <double 42.0, double 0.0>
Expand Down