[X86][AVX512] Improve support for signed saturation truncation stores

Matches what we already manage for unsigned saturation truncation stores

Differential Revision: https://reviews.llvm.org/D43629

llvm-svn: 326372

2 files changed, 23 insertions, 9 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index e4f86127a06..57e50d30af8 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -34349,6 +34349,20 @@ static SDValue detectSSatPattern(SDValue In, EVT VT, bool MatchPackUS = false) {
   return SDValue();
 }
 
+/// Detect a pattern of truncation with signed saturation.
+/// The types should allow to use VPMOVSS* instruction on AVX512.
+/// Return the source value to be truncated or SDValue() if the pattern was not
+/// matched.
+static SDValue detectAVX512SSatPattern(SDValue In, EVT VT,
+                                       const X86Subtarget &Subtarget,
+                                       const TargetLowering &TLI) {
+  if (!TLI.isTypeLegal(In.getValueType()))
+    return SDValue();
+  if (!isSATValidOnAVX512Subtarget(In.getValueType(), VT, Subtarget))
+    return SDValue();
+  return detectSSatPattern(In, VT);
+}
+
 /// Detect a pattern of truncation with saturation:
 /// (truncate (umin (x, unsigned_max_of_dest_type)) to dest_type).
 /// The types should allow to use VPMOVUS* instruction on AVX512.
@@ -34988,6 +35002,12 @@ static SDValue combineStore(SDNode *N, SelectionDAG &DAG,
 
     const TargetLowering &TLI = DAG.getTargetLoweringInfo();
     if (SDValue Val =
+        detectAVX512SSatPattern(St->getValue(), St->getMemoryVT(), Subtarget,
+                                TLI))
+      return EmitTruncSStore(true /* Signed saturation */, St->getChain(),
+                             dl, Val, St->getBasePtr(),
+                             St->getMemoryVT(), St->getMemOperand(), DAG);
+    if (SDValue Val =
         detectAVX512USatPattern(St->getValue(), St->getMemoryVT(), Subtarget,
                                 TLI))
       return EmitTruncSStore(false /* Unsigned saturation */, St->getChain(),
diff --git a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
index 2ad9b8de43c..4ba6693ac52 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
@@ -2042,9 +2042,7 @@ define void @trunc_ssat_v8i64_v8i8_store(<8 x i64> %a0, <8 x i8> *%p1) {
 ;
 ; AVX512-LABEL: trunc_ssat_v8i64_v8i8_store:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
-; AVX512-NEXT:    vpmaxsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
-; AVX512-NEXT:    vpmovqb %zmm0, (%rdi)
+; AVX512-NEXT:    vpmovsqb %zmm0, (%rdi)
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = icmp slt <8 x i64> %a0, <i64 127, i64 127, i64 127, i64 127, i64 127, i64 127, i64 127, i64 127>
@@ -2998,9 +2996,7 @@ define void @trunc_ssat_v8i32_v8i8_store(<8 x i32> %a0, <8 x i8> *%p1) {
 ;
 ; AVX512VL-LABEL: trunc_ssat_v8i32_v8i8_store:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpminsd {{.*}}(%rip){1to8}, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmaxsd {{.*}}(%rip){1to8}, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovdb %ymm0, (%rdi)
+; AVX512VL-NEXT:    vpmovsdb %ymm0, (%rdi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -3018,9 +3014,7 @@ define void @trunc_ssat_v8i32_v8i8_store(<8 x i32> %a0, <8 x i8> *%p1) {
 ;
 ; AVX512BWVL-LABEL: trunc_ssat_v8i32_v8i8_store:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpminsd {{.*}}(%rip){1to8}, %ymm0, %ymm0
-; AVX512BWVL-NEXT:    vpmaxsd {{.*}}(%rip){1to8}, %ymm0, %ymm0
-; AVX512BWVL-NEXT:    vpmovdb %ymm0, (%rdi)
+; AVX512BWVL-NEXT:    vpmovsdb %ymm0, (%rdi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %1 = icmp slt <8 x i32> %a0, <i32 127, i32 127, i32 127, i32 127, i32 127, i32 127, i32 127, i32 127>