[X86][SSE] combineTruncateWithSat - use truncateVectorWithPACK to chain PACKSS vXi32-vXi8 saturated truncation

We can use PACKSS to saturate each stage of the chain: PACKSSDW down to [-32768,32767] and then PACKSSWB to [-128,127].

PACKUS is a little trickier and will be handled in a separate patch.

llvm-svn: 325235

2 files changed, 26 insertions, 170 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 6a82de1d7fe..d9b7f192b51 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -34212,7 +34212,9 @@ static SDValue detectAVX512USatPattern(SDValue In, EVT VT,
 static SDValue combineTruncateWithSat(SDValue In, EVT VT, const SDLoc &DL,
                                       SelectionDAG &DAG,
                                       const X86Subtarget &Subtarget) {
+  EVT SVT = VT.getScalarType();
   EVT InVT = In.getValueType();
+  EVT InSVT = InVT.getScalarType();
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   if (TLI.isTypeLegal(InVT) && TLI.isTypeLegal(VT) &&
       isSATValidOnAVX512Subtarget(InVT, VT, Subtarget)) {
@@ -34221,16 +34223,19 @@ static SDValue combineTruncateWithSat(SDValue In, EVT VT, const SDLoc &DL,
     if (auto USatVal = detectUSatPattern(In, VT))
       return DAG.getNode(X86ISD::VTRUNCUS, DL, VT, USatVal);
   }
-  if (VT.isVector() && isPowerOf2_32(VT.getVectorNumElements()) &&
-      ((VT.getScalarType() == MVT::i8 && InVT.getScalarType() == MVT::i16) ||
-       (VT.getScalarType() == MVT::i16 && InVT.getScalarType() == MVT::i32))) {
-    if (auto SSatVal = detectSSatPattern(In, VT))
-      return truncateVectorWithPACK(X86ISD::PACKSS, VT, SSatVal, DL, DAG,
-                                    Subtarget);
-    if (Subtarget.hasSSE41() || VT.getScalarType() == MVT::i8)
+  if (VT.isVector() && isPowerOf2_32(VT.getVectorNumElements())) {
+    if ((SVT == MVT::i8 || SVT == MVT::i16) &&
+        (InSVT == MVT::i16 || InSVT == MVT::i32)) {
+      if (auto SSatVal = detectSSatPattern(In, VT))
+        return truncateVectorWithPACK(X86ISD::PACKSS, VT, SSatVal, DL, DAG,
+                                      Subtarget);
+    }
+    if ((SVT == MVT::i8 && InSVT == MVT::i16) ||
+        (SVT == MVT::i16 && InSVT == MVT::i32 && Subtarget.hasSSE41())) {
       if (auto USatVal = detectSSatPattern(In, VT, true))
         return truncateVectorWithPACK(X86ISD::PACKUS, VT, USatVal, DL, DAG,
                                       Subtarget);
+    }
   }
   return SDValue();
 }
diff --git a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
index 08c51c89ddc..3fe7c333328 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
@@ -2723,178 +2723,29 @@ define <8 x i8> @trunc_ssat_v8i32_v8i8(<8 x i32> %a0) {
 }
 
 define <16 x i8> @trunc_ssat_v16i32_v16i8(<16 x i32> %a0) {
-; SSE2-LABEL: trunc_ssat_v16i32_v16i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [127,127,127,127]
-; SSE2-NEXT:    movdqa %xmm5, %xmm4
-; SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
-; SSE2-NEXT:    pand %xmm4, %xmm3
-; SSE2-NEXT:    pandn %xmm5, %xmm4
-; SSE2-NEXT:    por %xmm3, %xmm4
-; SSE2-NEXT:    movdqa %xmm5, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    pandn %xmm5, %xmm3
-; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    movdqa %xmm5, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm5, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm5, %xmm1
-; SSE2-NEXT:    pcmpgtd %xmm0, %xmm1
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    pandn %xmm5, %xmm1
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [4294967168,4294967168,4294967168,4294967168]
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    pcmpgtd %xmm5, %xmm0
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pandn %xmm5, %xmm0
-; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm2, %xmm1
-; SSE2-NEXT:    pcmpgtd %xmm5, %xmm1
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    pandn %xmm5, %xmm1
-; SSE2-NEXT:    por %xmm2, %xmm1
-; SSE2-NEXT:    movdqa %xmm3, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm5, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm3
-; SSE2-NEXT:    pandn %xmm5, %xmm2
-; SSE2-NEXT:    por %xmm3, %xmm2
-; SSE2-NEXT:    movdqa %xmm4, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm5, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm4
-; SSE2-NEXT:    pandn %xmm5, %xmm3
-; SSE2-NEXT:    por %xmm4, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSE2-NEXT:    pand %xmm4, %xmm3
-; SSE2-NEXT:    pand %xmm4, %xmm2
-; SSE2-NEXT:    packuswb %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm4, %xmm1
-; SSE2-NEXT:    pand %xmm4, %xmm0
-; SSE2-NEXT:    packuswb %xmm1, %xmm0
-; SSE2-NEXT:    packuswb %xmm2, %xmm0
-; SSE2-NEXT:    retq
-;
-; SSSE3-LABEL: trunc_ssat_v16i32_v16i8:
-; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [127,127,127,127]
-; SSSE3-NEXT:    movdqa %xmm5, %xmm4
-; SSSE3-NEXT:    pcmpgtd %xmm3, %xmm4
-; SSSE3-NEXT:    pand %xmm4, %xmm3
-; SSSE3-NEXT:    pandn %xmm5, %xmm4
-; SSSE3-NEXT:    por %xmm3, %xmm4
-; SSSE3-NEXT:    movdqa %xmm5, %xmm3
-; SSSE3-NEXT:    pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT:    pand %xmm3, %xmm2
-; SSSE3-NEXT:    pandn %xmm5, %xmm3
-; SSSE3-NEXT:    por %xmm2, %xmm3
-; SSSE3-NEXT:    movdqa %xmm5, %xmm2
-; SSSE3-NEXT:    pcmpgtd %xmm1, %xmm2
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    pandn %xmm5, %xmm2
-; SSSE3-NEXT:    por %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa %xmm5, %xmm1
-; SSSE3-NEXT:    pcmpgtd %xmm0, %xmm1
-; SSSE3-NEXT:    pand %xmm1, %xmm0
-; SSSE3-NEXT:    pandn %xmm5, %xmm1
-; SSSE3-NEXT:    por %xmm0, %xmm1
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [4294967168,4294967168,4294967168,4294967168]
-; SSSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSSE3-NEXT:    pcmpgtd %xmm5, %xmm0
-; SSSE3-NEXT:    pand %xmm0, %xmm1
-; SSSE3-NEXT:    pandn %xmm5, %xmm0
-; SSSE3-NEXT:    por %xmm1, %xmm0
-; SSSE3-NEXT:    movdqa %xmm2, %xmm1
-; SSSE3-NEXT:    pcmpgtd %xmm5, %xmm1
-; SSSE3-NEXT:    pand %xmm1, %xmm2
-; SSSE3-NEXT:    pandn %xmm5, %xmm1
-; SSSE3-NEXT:    por %xmm2, %xmm1
-; SSSE3-NEXT:    movdqa %xmm3, %xmm2
-; SSSE3-NEXT:    pcmpgtd %xmm5, %xmm2
-; SSSE3-NEXT:    pand %xmm2, %xmm3
-; SSSE3-NEXT:    pandn %xmm5, %xmm2
-; SSSE3-NEXT:    por %xmm3, %xmm2
-; SSSE3-NEXT:    movdqa %xmm4, %xmm3
-; SSSE3-NEXT:    pcmpgtd %xmm5, %xmm3
-; SSSE3-NEXT:    pand %xmm3, %xmm4
-; SSSE3-NEXT:    pandn %xmm5, %xmm3
-; SSSE3-NEXT:    por %xmm4, %xmm3
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSSE3-NEXT:    pand %xmm4, %xmm3
-; SSSE3-NEXT:    pand %xmm4, %xmm2
-; SSSE3-NEXT:    packuswb %xmm3, %xmm2
-; SSSE3-NEXT:    pand %xmm4, %xmm1
-; SSSE3-NEXT:    pand %xmm4, %xmm0
-; SSSE3-NEXT:    packuswb %xmm1, %xmm0
-; SSSE3-NEXT:    packuswb %xmm2, %xmm0
-; SSSE3-NEXT:    retq
-;
-; SSE41-LABEL: trunc_ssat_v16i32_v16i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [127,127,127,127]
-; SSE41-NEXT:    pminsd %xmm4, %xmm3
-; SSE41-NEXT:    pminsd %xmm4, %xmm2
-; SSE41-NEXT:    pminsd %xmm4, %xmm1
-; SSE41-NEXT:    pminsd %xmm4, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [4294967168,4294967168,4294967168,4294967168]
-; SSE41-NEXT:    pmaxsd %xmm4, %xmm0
-; SSE41-NEXT:    pmaxsd %xmm4, %xmm1
-; SSE41-NEXT:    pmaxsd %xmm4, %xmm2
-; SSE41-NEXT:    pmaxsd %xmm4, %xmm3
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSE41-NEXT:    pand %xmm4, %xmm3
-; SSE41-NEXT:    pand %xmm4, %xmm2
-; SSE41-NEXT:    packuswb %xmm3, %xmm2
-; SSE41-NEXT:    pand %xmm4, %xmm1
-; SSE41-NEXT:    pand %xmm4, %xmm0
-; SSE41-NEXT:    packuswb %xmm1, %xmm0
-; SSE41-NEXT:    packuswb %xmm2, %xmm0
-; SSE41-NEXT:    retq
+; SSE-LABEL: trunc_ssat_v16i32_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    packssdw %xmm1, %xmm0
+; SSE-NEXT:    packsswb %xmm2, %xmm0
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_ssat_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [127,127,127,127]
-; AVX1-NEXT:    vpminsd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpminsd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX1-NEXT:    vpminsd %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpminsd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [4294967168,4294967168,4294967168,4294967168]
-; AVX1-NEXT:    vpmaxsd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpmaxsd %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpmaxsd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpmaxsd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_ssat_v16i32_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [127,127,127,127,127,127,127,127]
-; AVX2-NEXT:    vpminsd %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vpminsd %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [4294967168,4294967168,4294967168,4294967168,4294967168,4294967168,4294967168,4294967168]
-; AVX2-NEXT:    vpmaxsd %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpmaxsd %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;