Recommit r358887 "[TargetLowering][AMDGPU][X86] Improve SimplifyDemandedBits bitcast...

I've included a new fix in X86RegisterInfo to prevent PR41619 without
reintroducing r359392. We might be able to improve that in the base class
implementation of shouldRewriteCopySrc somehow. But this hopefully enables
forward progress on SimplifyDemandedBits improvements for now.

Original commit message:

This patch adds support for BigBitWidth -> SmallBitWidth bitcasts, splitting the DemandedBits/Elts accordingly.

The AMDGPU backend needed an extra  (srl (and x, c1 << c2), c2) -> (and (srl(x, c2), c1) combine to encourage BFE creation, I investigated putting this in DAGComb
but it caused a lot of noise on other targets - some improvements, some regressions.

The X86 changes are all definite wins.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@360552 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 502913886e338aca7fbe7970399cdc79b481de17..fc7e7d9c3ce460876ec0dc6a5d3491ff82e21d36 100644 (file)
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -1577,6 +1577,30 @@ bool TargetLowering::SimplifyDemandedBits(
                                      KnownSrcZero, TLO, Depth + 1))
         return true;
 
+      KnownBits KnownSrcBits;
+      if (SimplifyDemandedBits(Src, DemandedSrcBits, DemandedSrcElts,
+                               KnownSrcBits, TLO, Depth + 1))
+        return true;
+    } else if ((NumSrcEltBits % BitWidth) == 0 &&
+               TLO.DAG.getDataLayout().isLittleEndian()) {
+      unsigned Scale = NumSrcEltBits / BitWidth;
+      unsigned NumSrcElts = SrcVT.isVector() ? SrcVT.getVectorNumElements() : 1;
+      APInt DemandedSrcBits = APInt::getNullValue(NumSrcEltBits);
+      APInt DemandedSrcElts = APInt::getNullValue(NumSrcElts);
+      for (unsigned i = 0; i != NumElts; ++i)
+        if (DemandedElts[i]) {
+          unsigned Offset = (i % Scale) * BitWidth;
+          DemandedSrcBits.insertBits(DemandedBits, Offset);
+          DemandedSrcElts.setBit(i / Scale);
+        }
+
+      if (SrcVT.isVector()) {
+        APInt KnownSrcUndef, KnownSrcZero;
+        if (SimplifyDemandedVectorElts(Src, DemandedSrcElts, KnownSrcUndef,
+                                       KnownSrcZero, TLO, Depth + 1))
+          return true;
+      }
+
       KnownBits KnownSrcBits;
       if (SimplifyDemandedBits(Src, DemandedSrcBits, DemandedSrcElts,
                                KnownSrcBits, TLO, Depth + 1))
@@ -1586,7 +1610,7 @@ bool TargetLowering::SimplifyDemandedBits(
     // If this is a bitcast, let computeKnownBits handle it.  Only do this on a
     // recursive call where Known may be useful to the caller.
     if (Depth > 0) {
-      Known = TLO.DAG.computeKnownBits(Op, Depth);
+      Known = TLO.DAG.computeKnownBits(Op, DemandedElts, Depth);
       return false;
     }
     break;

diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp
index 7dec87cdcb03302128cb1342224d3f1012b60fa8..1e62958e722a1f10fc6097acd38b80b605a09120 100644 (file)
--- a/lib/Target/X86/X86RegisterInfo.cpp
+++ b/lib/Target/X86/X86RegisterInfo.cpp
@@ -216,6 +216,21 @@ X86RegisterInfo::getPointerRegClass(const MachineFunction &MF,
   }
 }
 
+bool X86RegisterInfo::shouldRewriteCopySrc(const TargetRegisterClass *DefRC,
+                                           unsigned DefSubReg,
+                                           const TargetRegisterClass *SrcRC,
+                                           unsigned SrcSubReg) const {
+  // Prevent rewriting a copy where the destination size is larger than the
+  // input size. See PR41619.
+  // FIXME: Should this be factored into the base implementation somehow.
+  if (DefRC->hasSuperClassEq(&X86::GR64RegClass) && DefSubReg == 0 &&
+      SrcRC->hasSuperClassEq(&X86::GR64RegClass) && SrcSubReg == X86::sub_32bit)
+    return false;
+
+  return TargetRegisterInfo::shouldRewriteCopySrc(DefRC, DefSubReg,
+                                                  SrcRC, SrcSubReg);
+}
+
 const TargetRegisterClass *
 X86RegisterInfo::getGPRsForTailCall(const MachineFunction &MF) const {
   const Function &F = MF.getFunction();

diff --git a/lib/Target/X86/X86RegisterInfo.h b/lib/Target/X86/X86RegisterInfo.h
index d362f20b4727f8291a0e47a29a467353614163e9..81e4920f6a5cedadb000f22615046fe77f15bfb6 100644 (file)
--- a/lib/Target/X86/X86RegisterInfo.h
+++ b/lib/Target/X86/X86RegisterInfo.h
@@ -74,6 +74,11 @@ public:
   getLargestLegalSuperClass(const TargetRegisterClass *RC,
                             const MachineFunction &MF) const override;
 
+  bool shouldRewriteCopySrc(const TargetRegisterClass *DefRC,
+                            unsigned DefSubReg,
+                            const TargetRegisterClass *SrcRC,
+                            unsigned SrcSubReg) const override;
+
   /// getPointerRegClass - Returns a TargetRegisterClass used for pointer
   /// values.
   const TargetRegisterClass *

diff --git a/test/CodeGen/X86/bitcast-setcc-256.ll b/test/CodeGen/X86/bitcast-setcc-256.ll
index 41635f375286ea2d364ad0719436c08df7d4e4ac..f9a233a583bdddc2cfa89ab54c4f2dc1e259705d 100644 (file)
--- a/test/CodeGen/X86/bitcast-setcc-256.ll
+++ b/test/CodeGen/X86/bitcast-setcc-256.ll
@@ -448,22 +448,6 @@ define void @bitcast_8i32_store(i8* %p, <8 x i32> %a0) {
 define void @bitcast_4i64_store(i4* %p, <4 x i64> %a0) {
 ; SSE2-SSSE3-LABEL: bitcast_4i64_store:
 ; SSE2-SSSE3:       # %bb.0:
-; SSE2-SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-SSSE3-NEXT:    pxor %xmm2, %xmm1
-; SSE2-SSSE3-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-SSSE3-NEXT:    pcmpeqd %xmm1, %xmm3
-; SSE2-SSSE3-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-SSSE3-NEXT:    pcmpgtd %xmm1, %xmm4
-; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[0,0,2,2]
-; SSE2-SSSE3-NEXT:    pand %xmm3, %xmm1
-; SSE2-SSSE3-NEXT:    por %xmm4, %xmm1
-; SSE2-SSSE3-NEXT:    pxor %xmm2, %xmm0
-; SSE2-SSSE3-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-SSSE3-NEXT:    pcmpeqd %xmm0, %xmm3
-; SSE2-SSSE3-NEXT:    pcmpgtd %xmm0, %xmm2
-; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,0,2,2]
-; SSE2-SSSE3-NEXT:    pand %xmm3, %xmm0
-; SSE2-SSSE3-NEXT:    por %xmm2, %xmm0
 ; SSE2-SSSE3-NEXT:    packssdw %xmm1, %xmm0
 ; SSE2-SSSE3-NEXT:    movmskps %xmm0, %eax
 ; SSE2-SSSE3-NEXT:    movb %al, (%rdi)

diff --git a/test/CodeGen/X86/bitcast-setcc-512.ll b/test/CodeGen/X86/bitcast-setcc-512.ll
index 3c294345dd50a7061c75d19269030225b3889f23..177be1fd6a69543fc18aac88bb3dad6eb51ae1ec 100644 (file)
--- a/test/CodeGen/X86/bitcast-setcc-512.ll
+++ b/test/CodeGen/X86/bitcast-setcc-512.ll
@@ -609,15 +609,13 @@ define void @bitcast_8i64_store(i8* %p, <8 x i64> %a0) {
 ;
 ; AVX1-LABEL: bitcast_8i64_store:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    movb %al, (%rdi)

diff --git a/test/CodeGen/X86/bitcast-vector-bool.ll b/test/CodeGen/X86/bitcast-vector-bool.ll
index 0914f2bff9f64761ba2129abcd14ab8dff2c3ae2..adcee2abe33b6eff4f1342bbef6043a75a2e2539 100644 (file)
--- a/test/CodeGen/X86/bitcast-vector-bool.ll
+++ b/test/CodeGen/X86/bitcast-vector-bool.ll
@@ -208,22 +208,6 @@ define i8 @bitcast_v16i8_to_v2i8(<16 x i8> %a0) nounwind {
 define i2 @bitcast_v4i64_to_v2i2(<4 x i64> %a0) nounwind {
 ; SSE2-SSSE3-LABEL: bitcast_v4i64_to_v2i2:
 ; SSE2-SSSE3:       # %bb.0:
-; SSE2-SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-SSSE3-NEXT:    pxor %xmm2, %xmm1
-; SSE2-SSSE3-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-SSSE3-NEXT:    pcmpeqd %xmm1, %xmm3
-; SSE2-SSSE3-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-SSSE3-NEXT:    pcmpgtd %xmm1, %xmm4
-; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[0,0,2,2]
-; SSE2-SSSE3-NEXT:    pand %xmm3, %xmm1
-; SSE2-SSSE3-NEXT:    por %xmm4, %xmm1
-; SSE2-SSSE3-NEXT:    pxor %xmm2, %xmm0
-; SSE2-SSSE3-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-SSSE3-NEXT:    pcmpeqd %xmm0, %xmm3
-; SSE2-SSSE3-NEXT:    pcmpgtd %xmm0, %xmm2
-; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,0,2,2]
-; SSE2-SSSE3-NEXT:    pand %xmm3, %xmm0
-; SSE2-SSSE3-NEXT:    por %xmm2, %xmm0
 ; SSE2-SSSE3-NEXT:    packssdw %xmm1, %xmm0
 ; SSE2-SSSE3-NEXT:    movmskps %xmm0, %eax
 ; SSE2-SSSE3-NEXT:    movl %eax, %ecx
@@ -532,15 +516,13 @@ define i4 @bitcast_v8i64_to_v2i4(<8 x i64> %a0) nounwind {
 ;
 ; AVX1-LABEL: bitcast_v8i64_to_v2i4:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    movl %eax, %ecx

diff --git a/test/CodeGen/X86/dagcombine-cse.ll b/test/CodeGen/X86/dagcombine-cse.ll
index bf1dab35875f6b0ebd0c9b141e195a0cc1611941..a532d87170d80f17c6338ac074b18de3f3c97eab 100644 (file)
--- a/test/CodeGen/X86/dagcombine-cse.ll
+++ b/test/CodeGen/X86/dagcombine-cse.ll
@@ -14,18 +14,11 @@ define i32 @t(i8* %ref_frame_ptr, i32 %ref_frame_stride, i32 %idxX, i32 %idxY) n
 ;
 ; X64-LABEL: t:
 ; X64:       ## %bb.0: ## %entry
-; X64-NEXT:    ## kill: def $edx killed $edx def $rdx
-; X64-NEXT:    ## kill: def $esi killed $esi def $rsi
 ; X64-NEXT:    imull %ecx, %esi
-; X64-NEXT:    leal (%rsi,%rdx), %eax
-; X64-NEXT:    cltq
+; X64-NEXT:    addl %edx, %esi
+; X64-NEXT:    movslq %esi, %rax
 ; X64-NEXT:    movl (%rdi,%rax), %eax
-; X64-NEXT:    leal 4(%rsi,%rdx), %ecx
-; X64-NEXT:    movslq %ecx, %rcx
-; X64-NEXT:    movzwl (%rdi,%rcx), %ecx
-; X64-NEXT:    shlq $32, %rcx
-; X64-NEXT:    orq %rax, %rcx
-; X64-NEXT:    movq %rcx, %xmm0
+; X64-NEXT:    movq %rax, %xmm0
 ; X64-NEXT:    movd %xmm0, %eax
 ; X64-NEXT:    retq
 entry:

diff --git a/test/CodeGen/X86/movmsk-cmp.ll b/test/CodeGen/X86/movmsk-cmp.ll
index 2e6123ff014e5ac2b4cd2b2d561cb7981a6a27df..c8b3488af1679daca196c1ca0165fa1c3ff90b4d 100644 (file)
--- a/test/CodeGen/X86/movmsk-cmp.ll
+++ b/test/CodeGen/X86/movmsk-cmp.ll
@@ -929,22 +929,6 @@ define i1 @allzeros_v16i32_sign(<16 x i32> %arg) {
 define i1 @allones_v4i64_sign(<4 x i64> %arg) {
 ; SSE2-LABEL: allones_v4i64_sign:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm1, %xmm3
-; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[0,0,2,2]
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    por %xmm4, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,0,2,2]
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    packssdw %xmm1, %xmm0
 ; SSE2-NEXT:    movmskps %xmm0, %eax
 ; SSE2-NEXT:    cmpb $15, %al
@@ -989,22 +973,6 @@ define i1 @allones_v4i64_sign(<4 x i64> %arg) {
 define i1 @allzeros_v4i64_sign(<4 x i64> %arg) {
 ; SSE2-LABEL: allzeros_v4i64_sign:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm1, %xmm3
-; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[0,0,2,2]
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    por %xmm4, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,0,2,2]
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    packssdw %xmm1, %xmm0
 ; SSE2-NEXT:    movmskps %xmm0, %eax
 ; SSE2-NEXT:    testb %al, %al
@@ -1095,15 +1063,13 @@ define i1 @allones_v8i64_sign(<8 x i64> %arg) {
 ;
 ; AVX1-LABEL: allones_v8i64_sign:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    cmpb $-1, %al
@@ -1198,15 +1164,13 @@ define i1 @allzeros_v8i64_sign(<8 x i64> %arg) {
 ;
 ; AVX1-LABEL: allzeros_v8i64_sign:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    testb %al, %al
@@ -2539,19 +2503,17 @@ define i1 @allones_v8i64_and1(<8 x i64> %arg) {
 ;
 ; AVX1-LABEL: allones_v8i64_and1:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpsllq $63, %xmm2, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsllq $63, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsllq $63, %xmm2, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpsllq $63, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllq $63, %xmm1, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    cmpb $-1, %al
@@ -2615,19 +2577,17 @@ define i1 @allzeros_v8i64_and1(<8 x i64> %arg) {
 ;
 ; AVX1-LABEL: allzeros_v8i64_and1:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpsllq $63, %xmm2, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsllq $63, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsllq $63, %xmm2, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpsllq $63, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllq $63, %xmm1, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    testb %al, %al
@@ -3962,19 +3922,17 @@ define i1 @allones_v8i64_and4(<8 x i64> %arg) {
 ;
 ; AVX1-LABEL: allones_v8i64_and4:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpsllq $61, %xmm2, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsllq $61, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsllq $61, %xmm2, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpsllq $61, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpsllq $61, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllq $61, %xmm1, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    cmpb $-1, %al
@@ -4038,19 +3996,17 @@ define i1 @allzeros_v8i64_and4(<8 x i64> %arg) {
 ;
 ; AVX1-LABEL: allzeros_v8i64_and4:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpsllq $61, %xmm2, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsllq $61, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsllq $61, %xmm2, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpsllq $61, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpsllq $61, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllq $61, %xmm1, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    testb %al, %al
@@ -4170,22 +4126,6 @@ define i32 @movmskps(<4 x float> %x) {
 define i32 @movmskpd256(<4 x double> %x) {
 ; SSE2-LABEL: movmskpd256:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm1, %xmm3
-; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[0,0,2,2]
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    por %xmm4, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,0,2,2]
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    packssdw %xmm1, %xmm0
 ; SSE2-NEXT:    movmskps %xmm0, %eax
 ; SSE2-NEXT:    retq

diff --git a/test/CodeGen/X86/pr41619.ll b/test/CodeGen/X86/pr41619.ll
new file mode 100644 (file)
index 0000000..7c71f2c
--- /dev/null
+++ b/test/CodeGen/X86/pr41619.ll
@@ -0,0 +1,27 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-macosx10.14.0 -mattr=avx2 | FileCheck %s
+
+define void @foo(double %arg) {
+; CHECK-LABEL: foo:
+; CHECK:       ## %bb.0: ## %bb
+; CHECK-NEXT:    vmovq %xmm0, %rax
+; CHECK-NEXT:    vmovd %eax, %xmm0
+; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vmovq %xmm0, %rax
+; CHECK-NEXT:    movl %eax, (%rax)
+; CHECK-NEXT:    vmovlps %xmm1, (%rax)
+; CHECK-NEXT:    retq
+bb:
+  %tmp = bitcast double %arg to i64
+  %tmp1 = trunc i64 %tmp to i32
+  %tmp2 = bitcast i32 %tmp1 to float
+  %tmp3 = insertelement <4 x float> zeroinitializer, float %tmp2, i32 2
+  %tmp4 = bitcast <4 x float> %tmp3 to <2 x double>
+  %tmp5 = extractelement <2 x double> %tmp4, i32 0
+  %tmp6 = extractelement <2 x double> %tmp4, i32 1
+  %tmp7 = bitcast double %tmp6 to i64
+  %tmp8 = trunc i64 %tmp7 to i32
+  store i32 %tmp8, i32* undef, align 4
+  store double %tmp5, double* undef, align 16
+  ret void
+}