[LegalizeVectorOps][X86] Add ISD::ABDS/ABSDU to the list of opcodes h…

…andled by LegalizeVectorOps. (#92332) The expand code is present, but we were missing the type query code so the nodes would be ignored until LegalizeDAG.
llvm · May 16, 2024 · f2d7400 · f2d7400
1 parent 245b7b6
commit f2d7400
Show file tree

Hide file tree

Showing 3 changed files with 34 additions and 32 deletions.
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
@@ -367,6 +367,8 @@ SDValue VectorLegalizer::LegalizeOp(SDValue Op) {
   case ISD::ROTL:
   case ISD::ROTR:
   case ISD::ABS:
+  case ISD::ABDS:
+  case ISD::ABDU:
   case ISD::BSWAP:
   case ISD::BITREVERSE:
   case ISD::CTLZ:

diff --git a/llvm/test/CodeGen/X86/midpoint-int-vec-128.ll b/llvm/test/CodeGen/X86/midpoint-int-vec-128.ll
@@ -1991,14 +1991,14 @@ define <8 x i16> @vec128_i16_unsigned_reg_reg(<8 x i16> %a1, <8 x i16> %a2) noun
 ;
 ; AVX512VL-FALLBACK-LABEL: vec128_i16_unsigned_reg_reg:
 ; AVX512VL-FALLBACK:       # %bb.0:
-; AVX512VL-FALLBACK-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm2
-; AVX512VL-FALLBACK-NEXT:    vpminuw %xmm1, %xmm0, %xmm1
-; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm1, %xmm2, %xmm2
-; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
-; AVX512VL-FALLBACK-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm1
-; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %xmm1, %xmm1, %xmm1
-; AVX512VL-FALLBACK-NEXT:    vpxor %xmm1, %xmm2, %xmm2
-; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpminuw %xmm1, %xmm0, %xmm2
+; AVX512VL-FALLBACK-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm2
+; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %xmm2, %xmm2, %xmm2
+; AVX512VL-FALLBACK-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
 ; AVX512VL-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
 ; AVX512VL-FALLBACK-NEXT:    retq
 ;
@@ -2784,14 +2784,14 @@ define <16 x i8> @vec128_i8_unsigned_reg_reg(<16 x i8> %a1, <16 x i8> %a2) nounw
 ;
 ; AVX512VL-FALLBACK-LABEL: vec128_i8_unsigned_reg_reg:
 ; AVX512VL-FALLBACK:       # %bb.0:
-; AVX512VL-FALLBACK-NEXT:    vpmaxub %xmm1, %xmm0, %xmm2
-; AVX512VL-FALLBACK-NEXT:    vpminub %xmm1, %xmm0, %xmm1
-; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
-; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
-; AVX512VL-FALLBACK-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm1
-; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %xmm1, %xmm1, %xmm1
-; AVX512VL-FALLBACK-NEXT:    vpternlogd $108, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm1, %xmm2
-; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpminub %xmm1, %xmm0, %xmm2
+; AVX512VL-FALLBACK-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm2
+; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %xmm2, %xmm2, %xmm2
+; AVX512VL-FALLBACK-NEXT:    vpternlogd $108, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm2, %xmm1
+; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
 ; AVX512VL-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
 ; AVX512VL-FALLBACK-NEXT:    retq
 ;

diff --git a/llvm/test/CodeGen/X86/midpoint-int-vec-256.ll b/llvm/test/CodeGen/X86/midpoint-int-vec-256.ll
@@ -1445,14 +1445,14 @@ define <16 x i16> @vec256_i16_unsigned_reg_reg(<16 x i16> %a1, <16 x i16> %a2) n
 ;
 ; AVX512VL-FALLBACK-LABEL: vec256_i16_unsigned_reg_reg:
 ; AVX512VL-FALLBACK:       # %bb.0:
-; AVX512VL-FALLBACK-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm2
-; AVX512VL-FALLBACK-NEXT:    vpminuw %ymm1, %ymm0, %ymm1
-; AVX512VL-FALLBACK-NEXT:    vpsubw %ymm1, %ymm2, %ymm2
-; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %ymm2, %ymm2
-; AVX512VL-FALLBACK-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm1
-; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %ymm1, %ymm1, %ymm1
-; AVX512VL-FALLBACK-NEXT:    vpxor %ymm1, %ymm2, %ymm2
-; AVX512VL-FALLBACK-NEXT:    vpsubw %ymm1, %ymm2, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpminuw %ymm1, %ymm0, %ymm2
+; AVX512VL-FALLBACK-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpsubw %ymm2, %ymm1, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm2
+; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %ymm2, %ymm2, %ymm2
+; AVX512VL-FALLBACK-NEXT:    vpxor %ymm2, %ymm1, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpsubw %ymm2, %ymm1, %ymm1
 ; AVX512VL-FALLBACK-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; AVX512VL-FALLBACK-NEXT:    retq
 ;
@@ -2210,14 +2210,14 @@ define <32 x i8> @vec256_i8_unsigned_reg_reg(<32 x i8> %a1, <32 x i8> %a2) nounw
 ;
 ; AVX512VL-FALLBACK-LABEL: vec256_i8_unsigned_reg_reg:
 ; AVX512VL-FALLBACK:       # %bb.0:
-; AVX512VL-FALLBACK-NEXT:    vpmaxub %ymm1, %ymm0, %ymm2
-; AVX512VL-FALLBACK-NEXT:    vpminub %ymm1, %ymm0, %ymm1
-; AVX512VL-FALLBACK-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
-; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %ymm2, %ymm2
-; AVX512VL-FALLBACK-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %ymm1, %ymm1, %ymm1
-; AVX512VL-FALLBACK-NEXT:    vpternlogd $108, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
-; AVX512VL-FALLBACK-NEXT:    vpsubb %ymm1, %ymm2, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpminub %ymm1, %ymm0, %ymm2
+; AVX512VL-FALLBACK-NEXT:    vpmaxub %ymm1, %ymm0, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpsubb %ymm2, %ymm1, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm2
+; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %ymm2, %ymm2, %ymm2
+; AVX512VL-FALLBACK-NEXT:    vpternlogd $108, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm2, %ymm1
+; AVX512VL-FALLBACK-NEXT:    vpsubb %ymm2, %ymm1, %ymm1
 ; AVX512VL-FALLBACK-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
 ; AVX512VL-FALLBACK-NEXT:    retq
 ;