From: Craig Topper Date: Thu, 3 Oct 2019 03:16:27 +0000 (+0000) Subject: [X86] Add broadcast load folding patterns to NoVLX VPMULLQ/VPMAXSQ/VPMAXUQ/VPMINSQ... X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=1653d09ad34f76596c46ca5d5caa069498f831d1;p=llvm [X86] Add broadcast load folding patterns to NoVLX VPMULLQ/VPMAXSQ/VPMAXUQ/VPMINSQ/VPMINUQ patterns. More fixes for PR36191. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@373560 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td index 7030e3bf76e..65a9cb62174 100644 --- a/lib/Target/X86/X86InstrAVX512.td +++ b/lib/Target/X86/X86InstrAVX512.td @@ -5060,6 +5060,12 @@ let Predicates = [HasDQI, NoVLX] in { (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR256X:$src1, sub_ymm), (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR256X:$src2, sub_ymm)), sub_ymm)>; + def : Pat<(v4i64 (mul (v4i64 VR256X:$src1), (v4i64 (X86VBroadcastld64 addr:$src2)))), + (EXTRACT_SUBREG + (VPMULLQZrmb + (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR256X:$src1, sub_ymm), + addr:$src2), + sub_ymm)>; def : Pat<(v2i64 (mul (v2i64 VR128X:$src1), (v2i64 VR128X:$src2))), (EXTRACT_SUBREG @@ -5067,29 +5073,47 @@ let Predicates = [HasDQI, NoVLX] in { (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src1, sub_xmm), (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src2, sub_xmm)), sub_xmm)>; + def : Pat<(v2i64 (mul (v2i64 VR128X:$src1), (v2i64 (X86VBroadcastld64 addr:$src2)))), + (EXTRACT_SUBREG + (VPMULLQZrmb + (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src1, sub_xmm), + addr:$src2), + sub_xmm)>; } -multiclass avx512_min_max_lowering { +multiclass avx512_min_max_lowering { def : Pat<(v4i64 (OpNode VR256X:$src1, VR256X:$src2)), (EXTRACT_SUBREG - (Instr + (!cast(Instr#"rr") (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR256X:$src1, sub_ymm), (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR256X:$src2, sub_ymm)), sub_ymm)>; + def : Pat<(v4i64 (OpNode (v4i64 VR256X:$src1), (v4i64 (X86VBroadcastld64 addr:$src2)))), + (EXTRACT_SUBREG + (!cast(Instr#"rmb") + (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR256X:$src1, sub_ymm), + addr:$src2), + sub_ymm)>; def : Pat<(v2i64 (OpNode VR128X:$src1, VR128X:$src2)), (EXTRACT_SUBREG - (Instr + (!cast(Instr#"rr") (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src1, sub_xmm), (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src2, sub_xmm)), sub_xmm)>; + def : Pat<(v2i64 (OpNode (v2i64 VR128X:$src1), (v2i64 (X86VBroadcastld64 addr:$src2)))), + (EXTRACT_SUBREG + (!cast(Instr#"rmb") + (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR128X:$src1, sub_xmm), + addr:$src2), + sub_xmm)>; } let Predicates = [HasAVX512, NoVLX] in { - defm : avx512_min_max_lowering; - defm : avx512_min_max_lowering; - defm : avx512_min_max_lowering; - defm : avx512_min_max_lowering; + defm : avx512_min_max_lowering<"VPMAXUQZ", umax>; + defm : avx512_min_max_lowering<"VPMINUQZ", umin>; + defm : avx512_min_max_lowering<"VPMAXSQZ", smax>; + defm : avx512_min_max_lowering<"VPMINSQZ", smin>; } //===----------------------------------------------------------------------===// diff --git a/test/CodeGen/X86/avx512-arith.ll b/test/CodeGen/X86/avx512-arith.ll index 29793a7e0bc..be88e3530a3 100644 --- a/test/CodeGen/X86/avx512-arith.ll +++ b/test/CodeGen/X86/avx512-arith.ll @@ -190,6 +190,52 @@ define <4 x i64> @imulq256(<4 x i64> %y, <4 x i64> %x) { ret <4 x i64>%z } +define <4 x i64> @imulq256_bcast(<4 x i64> %x) { +; AVX512F-LABEL: imulq256_bcast: +; AVX512F: # %bb.0: +; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [1337,1337,1337,1337] +; AVX512F-NEXT: vpmuludq %ymm1, %ymm0, %ymm2 +; AVX512F-NEXT: vpsrlq $32, %ymm0, %ymm0 +; AVX512F-NEXT: vpmuludq %ymm1, %ymm0, %ymm0 +; AVX512F-NEXT: vpsllq $32, %ymm0, %ymm0 +; AVX512F-NEXT: vpaddq %ymm0, %ymm2, %ymm0 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: imulq256_bcast: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpbroadcastq {{.*#+}} ymm1 = [1337,1337,1337,1337] +; AVX512VL-NEXT: vpmuludq %ymm1, %ymm0, %ymm2 +; AVX512VL-NEXT: vpsrlq $32, %ymm0, %ymm0 +; AVX512VL-NEXT: vpmuludq %ymm1, %ymm0, %ymm0 +; AVX512VL-NEXT: vpsllq $32, %ymm0, %ymm0 +; AVX512VL-NEXT: vpaddq %ymm0, %ymm2, %ymm0 +; AVX512VL-NEXT: retq +; +; AVX512BW-LABEL: imulq256_bcast: +; AVX512BW: # %bb.0: +; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [1337,1337,1337,1337] +; AVX512BW-NEXT: vpmuludq %ymm1, %ymm0, %ymm2 +; AVX512BW-NEXT: vpsrlq $32, %ymm0, %ymm0 +; AVX512BW-NEXT: vpmuludq %ymm1, %ymm0, %ymm0 +; AVX512BW-NEXT: vpsllq $32, %ymm0, %ymm0 +; AVX512BW-NEXT: vpaddq %ymm0, %ymm2, %ymm0 +; AVX512BW-NEXT: retq +; +; AVX512DQ-LABEL: imulq256_bcast: +; AVX512DQ: # %bb.0: +; AVX512DQ-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 +; AVX512DQ-NEXT: vpmullq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512DQ-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0 +; AVX512DQ-NEXT: retq +; +; SKX-LABEL: imulq256_bcast: +; SKX: # %bb.0: +; SKX-NEXT: vpmullq {{.*}}(%rip){1to4}, %ymm0, %ymm0 +; SKX-NEXT: retq + %z = mul <4 x i64> %x, + ret <4 x i64>%z +} + define <2 x i64> @imulq128(<2 x i64> %y, <2 x i64> %x) { ; AVX512F-LABEL: imulq128: ; AVX512F: # %bb.0: @@ -244,6 +290,54 @@ define <2 x i64> @imulq128(<2 x i64> %y, <2 x i64> %x) { ret <2 x i64>%z } +define <2 x i64> @imulq128_bcast(<2 x i64> %x) { +; AVX512F-LABEL: imulq128_bcast: +; AVX512F: # %bb.0: +; AVX512F-NEXT: vmovdqa {{.*#+}} xmm1 = [8086,8086] +; AVX512F-NEXT: vpmuludq %xmm1, %xmm0, %xmm2 +; AVX512F-NEXT: vpsrlq $32, %xmm0, %xmm0 +; AVX512F-NEXT: vpmuludq %xmm1, %xmm0, %xmm0 +; AVX512F-NEXT: vpsllq $32, %xmm0, %xmm0 +; AVX512F-NEXT: vpaddq %xmm0, %xmm2, %xmm0 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: imulq128_bcast: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [8086,8086] +; AVX512VL-NEXT: vpmuludq %xmm1, %xmm0, %xmm2 +; AVX512VL-NEXT: vpsrlq $32, %xmm0, %xmm0 +; AVX512VL-NEXT: vpmuludq %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: vpsllq $32, %xmm0, %xmm0 +; AVX512VL-NEXT: vpaddq %xmm0, %xmm2, %xmm0 +; AVX512VL-NEXT: retq +; +; AVX512BW-LABEL: imulq128_bcast: +; AVX512BW: # %bb.0: +; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = [8086,8086] +; AVX512BW-NEXT: vpmuludq %xmm1, %xmm0, %xmm2 +; AVX512BW-NEXT: vpsrlq $32, %xmm0, %xmm0 +; AVX512BW-NEXT: vpmuludq %xmm1, %xmm0, %xmm0 +; AVX512BW-NEXT: vpsllq $32, %xmm0, %xmm0 +; AVX512BW-NEXT: vpaddq %xmm0, %xmm2, %xmm0 +; AVX512BW-NEXT: retq +; +; AVX512DQ-LABEL: imulq128_bcast: +; AVX512DQ: # %bb.0: +; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512DQ-NEXT: vmovdqa {{.*#+}} xmm1 = [8086,8086] +; AVX512DQ-NEXT: vpmullq %zmm1, %zmm0, %zmm0 +; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 +; AVX512DQ-NEXT: vzeroupper +; AVX512DQ-NEXT: retq +; +; SKX-LABEL: imulq128_bcast: +; SKX: # %bb.0: +; SKX-NEXT: vpmullq {{.*}}(%rip), %xmm0, %xmm0 +; SKX-NEXT: retq + %z = mul <2 x i64> %x, + ret <2 x i64>%z +} + define <8 x double> @mulpd512(<8 x double> %y, <8 x double> %x) { ; CHECK-LABEL: mulpd512: ; CHECK: # %bb.0: # %entry diff --git a/test/CodeGen/X86/masked_store_trunc_ssat.ll b/test/CodeGen/X86/masked_store_trunc_ssat.ll index 777d4d14e4e..af413e85be3 100644 --- a/test/CodeGen/X86/masked_store_trunc_ssat.ll +++ b/test/CodeGen/X86/masked_store_trunc_ssat.ll @@ -1719,10 +1719,8 @@ define void @truncstore_v4i64_v4i32(<4 x i64> %x, <4 x i32>* %p, <4 x i32> %mask ; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 ; AVX512F-NEXT: kshiftlw $12, %k0, %k0 ; AVX512F-NEXT: kshiftrw $12, %k0, %k1 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [2147483647,2147483647,2147483647,2147483647] -; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968] -; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 +; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512F-NEXT: vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512F-NEXT: vpmovqd %zmm0, %ymm0 ; AVX512F-NEXT: vmovdqu32 %zmm0, (%rdi) {%k1} ; AVX512F-NEXT: vzeroupper @@ -1744,10 +1742,8 @@ define void @truncstore_v4i64_v4i32(<4 x i64> %x, <4 x i32>* %p, <4 x i32> %mask ; AVX512BW-NEXT: vptestmd %zmm1, %zmm1, %k0 ; AVX512BW-NEXT: kshiftlw $12, %k0, %k0 ; AVX512BW-NEXT: kshiftrw $12, %k0, %k1 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [2147483647,2147483647,2147483647,2147483647] -; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968] -; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512BW-NEXT: vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovqd %zmm0, %ymm0 ; AVX512BW-NEXT: vmovdqu32 %zmm0, (%rdi) {%k1} ; AVX512BW-NEXT: vzeroupper @@ -2027,10 +2023,8 @@ define void @truncstore_v4i64_v4i16(<4 x i64> %x, <4 x i16>* %p, <4 x i32> %mask ; AVX512F-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 ; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [32767,32767,32767,32767] -; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [18446744073709518848,18446744073709518848,18446744073709518848,18446744073709518848] -; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 +; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512F-NEXT: vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512F-NEXT: vpmovqw %zmm0, %xmm0 ; AVX512F-NEXT: kmovw %k0, %eax ; AVX512F-NEXT: testb $1, %al @@ -2071,10 +2065,8 @@ define void @truncstore_v4i64_v4i16(<4 x i64> %x, <4 x i16>* %p, <4 x i32> %mask ; AVX512BW-NEXT: vptestmd %zmm1, %zmm1, %k0 ; AVX512BW-NEXT: kshiftld $28, %k0, %k0 ; AVX512BW-NEXT: kshiftrd $28, %k0, %k1 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [32767,32767,32767,32767] -; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [18446744073709518848,18446744073709518848,18446744073709518848,18446744073709518848] -; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512BW-NEXT: vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovqw %zmm0, %xmm0 ; AVX512BW-NEXT: vmovdqu16 %zmm0, (%rdi) {%k1} ; AVX512BW-NEXT: vzeroupper @@ -2361,10 +2353,8 @@ define void @truncstore_v4i64_v4i8(<4 x i64> %x, <4 x i8>* %p, <4 x i32> %mask) ; AVX512F-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 ; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [127,127,127,127] -; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [18446744073709551488,18446744073709551488,18446744073709551488,18446744073709551488] -; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 +; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512F-NEXT: vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512F-NEXT: vpmovqb %zmm0, %xmm0 ; AVX512F-NEXT: kmovw %k0, %eax ; AVX512F-NEXT: testb $1, %al @@ -2405,10 +2395,8 @@ define void @truncstore_v4i64_v4i8(<4 x i64> %x, <4 x i8>* %p, <4 x i32> %mask) ; AVX512BW-NEXT: vptestmd %zmm1, %zmm1, %k0 ; AVX512BW-NEXT: kshiftlq $60, %k0, %k0 ; AVX512BW-NEXT: kshiftrq $60, %k0, %k1 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [127,127,127,127] -; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [18446744073709551488,18446744073709551488,18446744073709551488,18446744073709551488] -; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512BW-NEXT: vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovqb %zmm0, %xmm0 ; AVX512BW-NEXT: vmovdqu8 %zmm0, (%rdi) {%k1} ; AVX512BW-NEXT: vzeroupper diff --git a/test/CodeGen/X86/masked_store_trunc_usat.ll b/test/CodeGen/X86/masked_store_trunc_usat.ll index 254f0cda48f..1f47ed5d156 100644 --- a/test/CodeGen/X86/masked_store_trunc_usat.ll +++ b/test/CodeGen/X86/masked_store_trunc_usat.ll @@ -1465,8 +1465,7 @@ define void @truncstore_v4i64_v4i32(<4 x i64> %x, <4 x i32>* %p, <4 x i32> %mask ; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 ; AVX512F-NEXT: kshiftlw $12, %k0, %k0 ; AVX512F-NEXT: kshiftrw $12, %k0, %k1 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [4294967295,4294967295,4294967295,4294967295] -; AVX512F-NEXT: vpminuq %zmm1, %zmm0, %zmm0 +; AVX512F-NEXT: vpminuq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512F-NEXT: vpmovqd %zmm0, %ymm0 ; AVX512F-NEXT: vmovdqu32 %zmm0, (%rdi) {%k1} ; AVX512F-NEXT: vzeroupper @@ -1487,8 +1486,7 @@ define void @truncstore_v4i64_v4i32(<4 x i64> %x, <4 x i32>* %p, <4 x i32> %mask ; AVX512BW-NEXT: vptestmd %zmm1, %zmm1, %k0 ; AVX512BW-NEXT: kshiftlw $12, %k0, %k0 ; AVX512BW-NEXT: kshiftrw $12, %k0, %k1 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [4294967295,4294967295,4294967295,4294967295] -; AVX512BW-NEXT: vpminuq %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpminuq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovqd %zmm0, %ymm0 ; AVX512BW-NEXT: vmovdqu32 %zmm0, (%rdi) {%k1} ; AVX512BW-NEXT: vzeroupper @@ -1734,8 +1732,7 @@ define void @truncstore_v4i64_v4i16(<4 x i64> %x, <4 x i16>* %p, <4 x i32> %mask ; AVX512F-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 ; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [65535,65535,65535,65535] -; AVX512F-NEXT: vpminuq %zmm1, %zmm0, %zmm0 +; AVX512F-NEXT: vpminuq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512F-NEXT: vpmovqw %zmm0, %xmm0 ; AVX512F-NEXT: kmovw %k0, %eax ; AVX512F-NEXT: testb $1, %al @@ -1776,8 +1773,7 @@ define void @truncstore_v4i64_v4i16(<4 x i64> %x, <4 x i16>* %p, <4 x i32> %mask ; AVX512BW-NEXT: vptestmd %zmm1, %zmm1, %k0 ; AVX512BW-NEXT: kshiftld $28, %k0, %k0 ; AVX512BW-NEXT: kshiftrd $28, %k0, %k1 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [65535,65535,65535,65535] -; AVX512BW-NEXT: vpminuq %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpminuq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovqw %zmm0, %xmm0 ; AVX512BW-NEXT: vmovdqu16 %zmm0, (%rdi) {%k1} ; AVX512BW-NEXT: vzeroupper @@ -2028,8 +2024,7 @@ define void @truncstore_v4i64_v4i8(<4 x i64> %x, <4 x i8>* %p, <4 x i32> %mask) ; AVX512F-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 ; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [255,255,255,255] -; AVX512F-NEXT: vpminuq %zmm1, %zmm0, %zmm0 +; AVX512F-NEXT: vpminuq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512F-NEXT: vpmovqb %zmm0, %xmm0 ; AVX512F-NEXT: kmovw %k0, %eax ; AVX512F-NEXT: testb $1, %al @@ -2070,8 +2065,7 @@ define void @truncstore_v4i64_v4i8(<4 x i64> %x, <4 x i8>* %p, <4 x i32> %mask) ; AVX512BW-NEXT: vptestmd %zmm1, %zmm1, %k0 ; AVX512BW-NEXT: kshiftlq $60, %k0, %k0 ; AVX512BW-NEXT: kshiftrq $60, %k0, %k1 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [255,255,255,255] -; AVX512BW-NEXT: vpminuq %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpminuq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovqb %zmm0, %xmm0 ; AVX512BW-NEXT: vmovdqu8 %zmm0, (%rdi) {%k1} ; AVX512BW-NEXT: vzeroupper diff --git a/test/CodeGen/X86/vector-trunc-packus.ll b/test/CodeGen/X86/vector-trunc-packus.ll index a0306dc1cd4..a5fa41c521f 100644 --- a/test/CodeGen/X86/vector-trunc-packus.ll +++ b/test/CodeGen/X86/vector-trunc-packus.ll @@ -223,8 +223,7 @@ define <4 x i32> @trunc_packus_v4i64_v4i32(<4 x i64> %a0) { ; AVX512F-LABEL: trunc_packus_v4i64_v4i32: ; AVX512F: # %bb.0: ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [4294967295,4294967295,4294967295,4294967295] -; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0 +; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 ; AVX512F-NEXT: vpmovqd %zmm0, %ymm0 @@ -243,8 +242,7 @@ define <4 x i32> @trunc_packus_v4i64_v4i32(<4 x i64> %a0) { ; AVX512BW-LABEL: trunc_packus_v4i64_v4i32: ; AVX512BW: # %bb.0: ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [4294967295,4294967295,4294967295,4294967295] -; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovqd %zmm0, %ymm0 diff --git a/test/CodeGen/X86/vector-trunc-ssat.ll b/test/CodeGen/X86/vector-trunc-ssat.ll index bb734bb8e32..a298382cfb0 100644 --- a/test/CodeGen/X86/vector-trunc-ssat.ll +++ b/test/CodeGen/X86/vector-trunc-ssat.ll @@ -233,10 +233,8 @@ define <4 x i32> @trunc_ssat_v4i64_v4i32(<4 x i64> %a0) { ; AVX512F-LABEL: trunc_ssat_v4i64_v4i32: ; AVX512F: # %bb.0: ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [2147483647,2147483647,2147483647,2147483647] -; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm1 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968] -; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 +; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512F-NEXT: vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512F-NEXT: vpmovqd %zmm0, %ymm0 ; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 ; AVX512F-NEXT: vzeroupper @@ -251,10 +249,8 @@ define <4 x i32> @trunc_ssat_v4i64_v4i32(<4 x i64> %a0) { ; AVX512BW-LABEL: trunc_ssat_v4i64_v4i32: ; AVX512BW: # %bb.0: ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [2147483647,2147483647,2147483647,2147483647] -; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpbroadcastq {{.*#+}} ymm1 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968] -; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; AVX512BW-NEXT: vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovqd %zmm0, %ymm0 ; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 ; AVX512BW-NEXT: vzeroupper