[X86] Disable VPBLENDVB formation in combineLogicBlendIntoPBLENDV if VPTERNLOG is supported.
VPBLENDVB is multiple uops while VPTERNLOG is a single uop. So
we should use that instead.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D83155
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index edc6ce1..83ad55a 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -42936,6 +42936,10 @@
if (!Subtarget.hasSSE41())
return SDValue();
+ // If we have VPTERNLOG we should prefer that since PBLENDVB is multiple uops.
+ if (Subtarget.hasVLX())
+ return SDValue();
+
MVT BlendVT = VT.is256BitVector() ? MVT::v32i8 : MVT::v16i8;
X = DAG.getBitcast(BlendVT, X);
diff --git a/llvm/test/CodeGen/X86/vselect-packss.ll b/llvm/test/CodeGen/X86/vselect-packss.ll
index 966642d54..157245c 100644
--- a/llvm/test/CodeGen/X86/vselect-packss.ll
+++ b/llvm/test/CodeGen/X86/vselect-packss.ll
@@ -52,14 +52,23 @@
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; AVX512NOBW-LABEL: vselect_packss_v16i16:
-; AVX512NOBW: # %bb.0:
-; AVX512NOBW-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
-; AVX512NOBW-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512NOBW-NEXT: vpmovdb %zmm0, %xmm0
-; AVX512NOBW-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
-; AVX512NOBW-NEXT: vzeroupper
-; AVX512NOBW-NEXT: retq
+; AVX512F-LABEL: vselect_packss_v16i16:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
+; AVX512F-NEXT: vzeroupper
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: vselect_packss_v16i16:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512VL-NEXT: vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT: vpternlogq $202, %xmm3, %xmm2, %xmm0
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
;
; AVX512BWNOVL-LABEL: vselect_packss_v16i16:
; AVX512BWNOVL: # %bb.0:
@@ -73,7 +82,7 @@
; AVX512BWVL: # %bb.0:
; AVX512BWVL-NEXT: vpcmpeqw %ymm1, %ymm0, %k0
; AVX512BWVL-NEXT: vpmovm2b %k0, %xmm0
-; AVX512BWVL-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
+; AVX512BWVL-NEXT: vpternlogq $202, %xmm3, %xmm2, %xmm0
; AVX512BWVL-NEXT: vzeroupper
; AVX512BWVL-NEXT: retq
%1 = icmp eq <16 x i16> %a0, %a1
@@ -144,14 +153,23 @@
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; AVX512NOBW-LABEL: vselect_packss_v16i32:
-; AVX512NOBW: # %bb.0:
-; AVX512NOBW-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
-; AVX512NOBW-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512NOBW-NEXT: vpmovdb %zmm0, %xmm0
-; AVX512NOBW-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
-; AVX512NOBW-NEXT: vzeroupper
-; AVX512NOBW-NEXT: retq
+; AVX512F-LABEL: vselect_packss_v16i32:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
+; AVX512F-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
+; AVX512F-NEXT: vzeroupper
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: vselect_packss_v16i32:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
+; AVX512VL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT: vpternlogq $202, %xmm3, %xmm2, %xmm0
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
;
; AVX512BWNOVL-LABEL: vselect_packss_v16i32:
; AVX512BWNOVL: # %bb.0:
@@ -165,7 +183,7 @@
; AVX512BWVL: # %bb.0:
; AVX512BWVL-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
; AVX512BWVL-NEXT: vpmovm2b %k0, %xmm0
-; AVX512BWVL-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
+; AVX512BWVL-NEXT: vpternlogq $202, %xmm3, %xmm2, %xmm0
; AVX512BWVL-NEXT: vzeroupper
; AVX512BWVL-NEXT: retq
%1 = icmp eq <16 x i32> %a0, %a1
@@ -292,16 +310,27 @@
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; AVX512NOBW-LABEL: vselect_packss_v16i64:
-; AVX512NOBW: # %bb.0:
-; AVX512NOBW-NEXT: vpcmpeqq %zmm2, %zmm0, %k0
-; AVX512NOBW-NEXT: vpcmpeqq %zmm3, %zmm1, %k1
-; AVX512NOBW-NEXT: kunpckbw %k0, %k1, %k1
-; AVX512NOBW-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512NOBW-NEXT: vpmovdb %zmm0, %xmm0
-; AVX512NOBW-NEXT: vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
-; AVX512NOBW-NEXT: vzeroupper
-; AVX512NOBW-NEXT: retq
+; AVX512F-LABEL: vselect_packss_v16i64:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vpcmpeqq %zmm2, %zmm0, %k0
+; AVX512F-NEXT: vpcmpeqq %zmm3, %zmm1, %k1
+; AVX512F-NEXT: kunpckbw %k0, %k1, %k1
+; AVX512F-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT: vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
+; AVX512F-NEXT: vzeroupper
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: vselect_packss_v16i64:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vpcmpeqq %zmm2, %zmm0, %k0
+; AVX512VL-NEXT: vpcmpeqq %zmm3, %zmm1, %k1
+; AVX512VL-NEXT: kunpckbw %k0, %k1, %k1
+; AVX512VL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT: vpternlogq $202, %xmm5, %xmm4, %xmm0
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
;
; AVX512BWNOVL-LABEL: vselect_packss_v16i64:
; AVX512BWNOVL: # %bb.0:
@@ -319,7 +348,7 @@
; AVX512BWVL-NEXT: vpcmpeqq %zmm3, %zmm1, %k1
; AVX512BWVL-NEXT: kunpckbw %k0, %k1, %k0
; AVX512BWVL-NEXT: vpmovm2b %k0, %xmm0
-; AVX512BWVL-NEXT: vpblendvb %xmm0, %xmm4, %xmm5, %xmm0
+; AVX512BWVL-NEXT: vpternlogq $202, %xmm5, %xmm4, %xmm0
; AVX512BWVL-NEXT: vzeroupper
; AVX512BWVL-NEXT: retq
%1 = icmp eq <16 x i64> %a0, %a1
@@ -375,14 +404,41 @@
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; AVX512-LABEL: vselect_packss:
-; AVX512: # %bb.0:
-; AVX512-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
-; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX512-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
-; AVX512-NEXT: vzeroupper
-; AVX512-NEXT: retq
+; AVX512F-LABEL: vselect_packss:
+; AVX512F: # %bb.0:
+; AVX512F-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX512F-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
+; AVX512F-NEXT: vzeroupper
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: vselect_packss:
+; AVX512VL: # %bb.0:
+; AVX512VL-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX512VL-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT: vpternlogq $202, %xmm3, %xmm2, %xmm0
+; AVX512VL-NEXT: vzeroupper
+; AVX512VL-NEXT: retq
+;
+; AVX512BWNOVL-LABEL: vselect_packss:
+; AVX512BWNOVL: # %bb.0:
+; AVX512BWNOVL-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
+; AVX512BWNOVL-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX512BWNOVL-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
+; AVX512BWNOVL-NEXT: vpblendvb %xmm0, %xmm2, %xmm3, %xmm0
+; AVX512BWNOVL-NEXT: vzeroupper
+; AVX512BWNOVL-NEXT: retq
+;
+; AVX512BWVL-LABEL: vselect_packss:
+; AVX512BWVL: # %bb.0:
+; AVX512BWVL-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
+; AVX512BWVL-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX512BWVL-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
+; AVX512BWVL-NEXT: vpternlogq $202, %xmm3, %xmm2, %xmm0
+; AVX512BWVL-NEXT: vzeroupper
+; AVX512BWVL-NEXT: retq
%1 = icmp eq <16 x i16> %a0, %a1
%2 = sext <16 x i1> %1 to <16 x i16>
%3 = shufflevector <16 x i16> %2, <16 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
