[X86][SSE] Truncate with PACKSS any input with sufficient sign-bits

So far we've only been using PACKSS truncations with 'all-bits or zero-bits' patterns (vector comparison results etc.). When really we can safely use it for any case as long as the number of sign bits reach down to the last 16-bits (or 8-bits if we're truncating to bytes).

The next steps after this is add the equivalent support for PACKUS and to support packing to sub-128 bit vectors for truncating stores etc.

Differential Revision: https://reviews.llvm.org/D39476

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@317086 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 11fe0eae39a575ea32314fa9001fe36dbded2244..d243cf1072088dcd3471d8b274e42680b396ed66 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -16222,8 +16222,10 @@ SDValue X86TargetLowering::LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
     return DAG.getNode(X86ISD::VTRUNC, DL, VT, In);
   }
 
-  // Truncate with PACKSS if we are truncating a vector zero/all-bits result.
-  if (InVT.getScalarSizeInBits() == DAG.ComputeNumSignBits(In))
+  // Truncate with PACKSS if we are truncating a vector with sign-bits that
+  // extend all the way to the packed/truncated value.
+  unsigned NumPackedBits = std::min<unsigned>(VT.getScalarSizeInBits(), 16);
+  if ((InVT.getScalarSizeInBits() - NumPackedBits) < DAG.ComputeNumSignBits(In))
     if (SDValue V = truncateVectorWithPACKSS(VT, In, DL, DAG, Subtarget))
       return V;
 
@@ -34422,7 +34424,7 @@ static SDValue combineVectorTruncation(SDNode *N, SelectionDAG &DAG,
     return SDValue();
 }
 
-/// This function transforms vector truncation of 'all or none' bits values.
+/// This function transforms vector truncation of 'extended sign-bits' values.
 /// vXi16/vXi32/vXi64 to vXi8/vXi16/vXi32 into X86ISD::PACKSS operations.
 static SDValue combineVectorSignBitsTruncation(SDNode *N, SDLoc &DL,
                                                SelectionDAG &DAG,
@@ -34444,12 +34446,6 @@ static SDValue combineVectorSignBitsTruncation(SDNode *N, SDLoc &DL,
   MVT InVT = In.getValueType().getSimpleVT();
   MVT InSVT = InVT.getScalarType();
 
-  // Use PACKSS if the input is a splatted sign bit.
-  // e.g. Comparison result, sext_in_reg, etc.
-  unsigned NumSignBits = DAG.ComputeNumSignBits(In);
-  if (NumSignBits != InSVT.getSizeInBits())
-    return SDValue();
-
   // Check we have a truncation suited for PACKSS.
   if (!VT.is128BitVector() && !VT.is256BitVector())
     return SDValue();
@@ -34458,6 +34454,13 @@ static SDValue combineVectorSignBitsTruncation(SDNode *N, SDLoc &DL,
   if (InSVT != MVT::i16 && InSVT != MVT::i32 && InSVT != MVT::i64)
     return SDValue();
 
+  // Use PACKSS if the input has sign-bits that extend all the way to the
+  // packed/truncated value. e.g. Comparison result, sext_in_reg, etc.
+  unsigned NumSignBits = DAG.ComputeNumSignBits(In);
+  unsigned NumPackedBits = std::min<unsigned>(SVT.getSizeInBits(), 16);
+  if (NumSignBits <= (InSVT.getSizeInBits() - NumPackedBits))
+    return SDValue();
+
   return truncateVectorWithPACKSS(VT, In, DL, DAG, Subtarget);
 }
 

diff --git a/test/CodeGen/X86/avg.ll b/test/CodeGen/X86/avg.ll
index f672b3fdba00d7d05841a9b0938bfd056197da23..5b36ccf4a5417fb1761419b5e7dda9c525fb768e 100644 (file)
--- a/test/CodeGen/X86/avg.ll
+++ b/test/CodeGen/X86/avg.ll
@@ -2567,52 +2567,51 @@ define void @avg_v64i8_const(<64 x i8>* %a) nounwind {
 ; AVX2-NEXT:    vpaddd %ymm8, %ymm6, %ymm6
 ; AVX2-NEXT:    vpaddd %ymm8, %ymm5, %ymm5
 ; AVX2-NEXT:    vpaddd %ymm8, %ymm4, %ymm4
-; AVX2-NEXT:    vpaddd %ymm8, %ymm3, %ymm3
+; AVX2-NEXT:    vpaddd %ymm8, %ymm3, %ymm9
 ; AVX2-NEXT:    vpaddd %ymm8, %ymm2, %ymm2
 ; AVX2-NEXT:    vpaddd %ymm8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddd %ymm8, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrld $1, %ymm0, %ymm0
+; AVX2-NEXT:    vpsrld $1, %ymm0, %ymm10
 ; AVX2-NEXT:    vpsrld $1, %ymm1, %ymm1
-; AVX2-NEXT:    vpsrld $1, %ymm2, %ymm8
-; AVX2-NEXT:    vpsrld $1, %ymm3, %ymm9
+; AVX2-NEXT:    vpsrld $1, %ymm2, %ymm3
+; AVX2-NEXT:    vpsrld $1, %ymm9, %ymm8
 ; AVX2-NEXT:    vpsrld $1, %ymm4, %ymm4
 ; AVX2-NEXT:    vpsrld $1, %ymm5, %ymm5
 ; AVX2-NEXT:    vpsrld $1, %ymm6, %ymm6
-; AVX2-NEXT:    vpsrld $1, %ymm7, %ymm3
-; 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, %ymm3, %ymm3
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm3[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, %xmm7, %xmm7
-; AVX2-NEXT:    vpshufb %ymm2, %ymm6, %ymm6
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,3]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm6, %xmm6
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm7[0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm5, %ymm5
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm5, %xmm5
-; AVX2-NEXT:    vpshufb %ymm2, %ymm4, %ymm4
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm4, %xmm4
+; AVX2-NEXT:    vpsrld $1, %ymm7, %ymm2
+; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm7
+; AVX2-NEXT:    vpackssdw %xmm7, %xmm2, %xmm7
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
+; AVX2-NEXT:    vpshufb %xmm2, %xmm7, %xmm7
+; AVX2-NEXT:    vextracti128 $1, %ymm6, %xmm0
+; AVX2-NEXT:    vpackssdw %xmm0, %xmm6, %xmm0
+; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
+; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm6
+; AVX2-NEXT:    vpackssdw %xmm6, %xmm5, %xmm5
+; AVX2-NEXT:    vpshufb %xmm2, %xmm5, %xmm5
+; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm6
+; AVX2-NEXT:    vpackssdw %xmm6, %xmm4, %xmm4
+; AVX2-NEXT:    vpshufb %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm5[0]
-; AVX2-NEXT:    vinserti128 $1, %xmm6, %ymm4, %ymm4
-; AVX2-NEXT:    vpshufb %ymm2, %ymm9, %ymm5
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm5, %xmm5
-; AVX2-NEXT:    vpshufb %ymm2, %ymm8, %ymm6
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,3]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm6, %xmm6
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm5 = xmm6[0],xmm5[0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; 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:    vinserti128 $1, %xmm5, %ymm0, %ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm4, %ymm0
+; AVX2-NEXT:    vextracti128 $1, %ymm8, %xmm4
+; AVX2-NEXT:    vpackssdw %xmm4, %xmm8, %xmm4
+; AVX2-NEXT:    vpshufb %xmm2, %xmm4, %xmm4
+; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm5
+; AVX2-NEXT:    vpackssdw %xmm5, %xmm3, %xmm3
+; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm3
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm4
+; AVX2-NEXT:    vpackssdw %xmm4, %xmm1, %xmm1
+; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vextracti128 $1, %ymm10, %xmm4
+; AVX2-NEXT:    vpackssdw %xmm4, %xmm10, %xmm4
+; AVX2-NEXT:    vpshufb %xmm2, %xmm4, %xmm2
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; AVX2-NEXT:    vmovdqu %ymm1, (%rax)
 ; AVX2-NEXT:    vmovdqu %ymm0, (%rax)
-; AVX2-NEXT:    vmovdqu %ymm4, (%rax)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;

diff --git a/test/CodeGen/X86/avx2-shift.ll b/test/CodeGen/X86/avx2-shift.ll
index 594cfe61e04866d949a7ec199a548c2d420bc961..650fdd3af01b5bd6e7eaface1f4a46e67f875196 100644 (file)
--- a/test/CodeGen/X86/avx2-shift.ll
+++ b/test/CodeGen/X86/avx2-shift.ll
@@ -556,9 +556,8 @@ define <8 x i16> @variable_ashr16(<8 x i16> %lhs, <8  x i16> %rhs) {
 ; X32-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; X32-NEXT:    vpmovsxwd %xmm0, %ymm0
 ; X32-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; X32-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; X32-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; X32-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; X32-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; X32-NEXT:    vzeroupper
 ; X32-NEXT:    retl
 ;
@@ -567,9 +566,8 @@ define <8 x i16> @variable_ashr16(<8 x i16> %lhs, <8  x i16> %rhs) {
 ; X64-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; X64-NEXT:    vpmovsxwd %xmm0, %ymm0
 ; X64-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; X64-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; X64-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; X64-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; X64-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
   %res = ashr <8 x i16> %lhs, %rhs

diff --git a/test/CodeGen/X86/avx2-vector-shifts.ll b/test/CodeGen/X86/avx2-vector-shifts.ll
index dcce9b1051bfe0afea49d4728b78c6ea2d86a820..36ab1be8a4afda408f5e20aea3233ad5ce932fd6 100644 (file)
--- a/test/CodeGen/X86/avx2-vector-shifts.ll
+++ b/test/CodeGen/X86/avx2-vector-shifts.ll
@@ -499,9 +499,8 @@ define <8 x i16> @ashr_8i16(<8 x i16> %r, <8 x i16> %a) nounwind {
 ; X32-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; X32-NEXT:    vpmovsxwd %xmm0, %ymm0
 ; X32-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; X32-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; X32-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; X32-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; X32-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; X32-NEXT:    vzeroupper
 ; X32-NEXT:    retl
 ;
@@ -510,9 +509,8 @@ define <8 x i16> @ashr_8i16(<8 x i16> %r, <8 x i16> %a) nounwind {
 ; X64-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; X64-NEXT:    vpmovsxwd %xmm0, %ymm0
 ; X64-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; X64-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; X64-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; X64-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; X64-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
   %ashr = ashr <8 x i16> %r, %a

diff --git a/test/CodeGen/X86/vector-shift-ashr-128.ll b/test/CodeGen/X86/vector-shift-ashr-128.ll
index 55766daecff8b55a8e3687fa74ce2a230ffa7bb5..53a8961a2dd128d879609f63582a7694cf62085d 100644 (file)
--- a/test/CodeGen/X86/vector-shift-ashr-128.ll
+++ b/test/CodeGen/X86/vector-shift-ashr-128.ll
@@ -318,9 +318,8 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
 ; AVX2-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -1249,9 +1248,8 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) nounwind {
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
 ; AVX2-NEXT:    vpsravd {{.*}}(%rip), %ymm0, %ymm0
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;

diff --git a/test/CodeGen/X86/vector-trunc.ll b/test/CodeGen/X86/vector-trunc.ll
index 91d09831622331fac0da2f5aa22739772a7eabd4..e9da847d0ad7d14db845b044c3694cf3f480c681 100644 (file)
--- a/test/CodeGen/X86/vector-trunc.ll
+++ b/test/CodeGen/X86/vector-trunc.ll
@@ -404,9 +404,8 @@ define <8 x i16> @trunc8i32_8i16_ashr(<8 x i32> %a) {
 ; AVX2-LABEL: trunc8i32_8i16_ashr:
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpsrad $16, %ymm0, %ymm0
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -695,62 +694,28 @@ entry:
 }
 
 define void @trunc16i32_16i16_ashr(<16 x i32> %a) {
-; SSE2-LABEL: trunc16i32_16i16_ashr:
-; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    psrad $16, %xmm3
-; SSE2-NEXT:    psrad $16, %xmm2
-; SSE2-NEXT:    packssdw %xmm3, %xmm2
-; SSE2-NEXT:    psrad $16, %xmm1
-; SSE2-NEXT:    psrad $16, %xmm0
-; SSE2-NEXT:    packssdw %xmm1, %xmm0
-; SSE2-NEXT:    movdqu %xmm2, (%rax)
-; SSE2-NEXT:    movdqu %xmm0, (%rax)
-; SSE2-NEXT:    retq
-;
-; SSSE3-LABEL: trunc16i32_16i16_ashr:
-; SSSE3:       # BB#0: # %entry
-; SSSE3-NEXT:    psrad $16, %xmm3
-; SSSE3-NEXT:    psrad $16, %xmm2
-; SSSE3-NEXT:    packssdw %xmm3, %xmm2
-; SSSE3-NEXT:    psrad $16, %xmm1
-; SSSE3-NEXT:    psrad $16, %xmm0
-; SSSE3-NEXT:    packssdw %xmm1, %xmm0
-; SSSE3-NEXT:    movdqu %xmm2, (%rax)
-; SSSE3-NEXT:    movdqu %xmm0, (%rax)
-; SSSE3-NEXT:    retq
-;
-; SSE41-LABEL: trunc16i32_16i16_ashr:
-; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    psrad $16, %xmm2
-; SSE41-NEXT:    psrad $16, %xmm3
-; SSE41-NEXT:    psrad $16, %xmm0
-; SSE41-NEXT:    psrad $16, %xmm1
-; SSE41-NEXT:    pxor %xmm4, %xmm4
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0],xmm4[1],xmm1[2],xmm4[3],xmm1[4],xmm4[5],xmm1[6],xmm4[7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm4[1],xmm0[2],xmm4[3],xmm0[4],xmm4[5],xmm0[6],xmm4[7]
-; SSE41-NEXT:    packusdw %xmm1, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2],xmm4[3],xmm3[4],xmm4[5],xmm3[6],xmm4[7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0],xmm4[1],xmm2[2],xmm4[3],xmm2[4],xmm4[5],xmm2[6],xmm4[7]
-; SSE41-NEXT:    packusdw %xmm3, %xmm2
-; SSE41-NEXT:    movdqu %xmm2, (%rax)
-; SSE41-NEXT:    movdqu %xmm0, (%rax)
-; SSE41-NEXT:    retq
+; SSE-LABEL: trunc16i32_16i16_ashr:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    psrad $16, %xmm3
+; SSE-NEXT:    psrad $16, %xmm2
+; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    psrad $16, %xmm1
+; SSE-NEXT:    psrad $16, %xmm0
+; SSE-NEXT:    packssdw %xmm1, %xmm0
+; SSE-NEXT:    movdqu %xmm2, (%rax)
+; SSE-NEXT:    movdqu %xmm0, (%rax)
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc16i32_16i16_ashr:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vpsrad $16, %xmm0, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpsrad $16, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrad $16, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrad $16, %xmm1, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpsrad $16, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrad $16, %xmm1, %xmm1
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm4[1],xmm1[2],xmm4[3],xmm1[4],xmm4[5],xmm1[6],xmm4[7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2],xmm4[3],xmm3[4],xmm4[5],xmm3[6],xmm4[7]
-; AVX1-NEXT:    vpackusdw %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm4[1],xmm0[2],xmm4[3],xmm0[4],xmm4[5],xmm0[6],xmm4[7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0],xmm4[1],xmm2[2],xmm4[3],xmm2[4],xmm4[5],xmm2[6],xmm4[7]
-; AVX1-NEXT:    vpackusdw %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovups %ymm0, (%rax)
 ; AVX1-NEXT:    vzeroupper
@@ -760,11 +725,8 @@ define void @trunc16i32_16i16_ashr(<16 x i32> %a) {
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpsrad $16, %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrad $16, %ymm0, %ymm0
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm1, %ymm1
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    vmovdqu %ymm0, (%rax)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -935,53 +897,39 @@ entry:
 define void @trunc16i32_16i8_ashr(<16 x i32> %a) {
 ; SSE-LABEL: trunc16i32_16i8_ashr:
 ; SSE:       # BB#0: # %entry
-; SSE-NEXT:    psrad $24, %xmm0
 ; SSE-NEXT:    psrad $24, %xmm1
-; SSE-NEXT:    psrad $24, %xmm2
+; SSE-NEXT:    psrad $24, %xmm0
+; SSE-NEXT:    packssdw %xmm1, %xmm0
 ; SSE-NEXT:    psrad $24, %xmm3
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSE-NEXT:    pand %xmm4, %xmm3
-; SSE-NEXT:    pand %xmm4, %xmm2
-; SSE-NEXT:    packuswb %xmm3, %xmm2
-; SSE-NEXT:    pand %xmm4, %xmm1
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    packuswb %xmm1, %xmm0
-; SSE-NEXT:    packuswb %xmm2, %xmm0
+; SSE-NEXT:    psrad $24, %xmm2
+; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    packsswb %xmm2, %xmm0
 ; SSE-NEXT:    movdqu %xmm0, (%rax)
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc16i32_16i8_ashr:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vpsrad $24, %xmm0, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpsrad $24, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrad $24, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrad $24, %xmm1, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpsrad $24, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrad $24, %xmm1, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpackuswb %xmm0, %xmm2, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovdqu %xmm0, (%rax)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc16i32_16i8_ashr:
 ; AVX2:       # BB#0: # %entry
-; AVX2-NEXT:    vpsrad $24, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrad $24, %ymm1, %ymm1
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm1, %ymm1
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-NEXT:    vpsrad $24, %ymm0, %ymm0
+; AVX2-NEXT:    vpacksswb %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:    vmovdqu %xmm0, (%rax)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -1031,12 +979,12 @@ define void @trunc16i32_16i8_lshr(<16 x i32> %a) {
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpsrld $24, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrld $24, %ymm1, %ymm1
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm1, %ymm1
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpackssdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; AVX2-NEXT:    vpackssdw %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX2-NEXT:    vmovdqu %xmm0, (%rax)
@@ -1153,12 +1101,10 @@ define void @trunc16i16_16i8_ashr(<16 x i16> %a) {
 ;
 ; AVX1-LABEL: trunc16i16_16i8_ashr:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vpsraw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpsraw $8, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsraw $8, %xmm0, %xmm0
-; AVX1-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    vpacksswb %xmm0, %xmm1, %xmm1
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovdqu %xmm0, (%rax)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -1167,10 +1113,7 @@ define void @trunc16i16_16i8_ashr(<16 x i16> %a) {
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpsraw $8, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovdqu %xmm0, (%rax)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq