From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sat, 11 May 2019 16:07:12 +0000 (+0000)
Subject: [X86][SSE] Add SimplifyDemandedVectorElts HADD/HSUB handling.
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=6661d4fd0b9574f7e48db866bdb98a7aa88705ab;p=llvm

[X86][SSE] Add SimplifyDemandedVectorElts HADD/HSUB handling.

Still missing PHADDW/PHSUBW tests because PEXTRW doesn't call SimplifyDemandedVectorElts

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@360526 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 22f05a5f05a..a428f3fac0c 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -6152,6 +6152,34 @@ static void getPackDemandedElts(EVT VT, const APInt &DemandedElts,
   }
 }
 
+// Split the demanded elts of a HADD/HSUB node between its operands.
+static void getHorizDemandedElts(EVT VT, const APInt &DemandedElts,
+                                 APInt &DemandedLHS, APInt &DemandedRHS) {
+  int NumLanes = VT.getSizeInBits() / 128;
+  int NumElts = DemandedElts.getBitWidth();
+  int NumEltsPerLane = NumElts / NumLanes;
+  int HalfEltsPerLane = NumEltsPerLane / 2;
+
+  DemandedLHS = APInt::getNullValue(NumElts);
+  DemandedRHS = APInt::getNullValue(NumElts);
+
+  // Map DemandedElts to the horizontal operands.
+  for (int Idx = 0; Idx != NumElts; ++Idx) {
+    if (!DemandedElts[Idx])
+      continue;
+    int LaneIdx = (Idx / NumEltsPerLane) * NumEltsPerLane;
+    int LocalIdx = Idx % NumEltsPerLane;
+    if (LocalIdx < HalfEltsPerLane) {
+      DemandedLHS.setBit(LaneIdx + 2 * LocalIdx + 0);
+      DemandedLHS.setBit(LaneIdx + 2 * LocalIdx + 1);
+    } else {
+      LocalIdx -= HalfEltsPerLane;
+      DemandedRHS.setBit(LaneIdx + 2 * LocalIdx + 0);
+      DemandedRHS.setBit(LaneIdx + 2 * LocalIdx + 1);
+    }
+  }
+}
+
 /// Calculates the shuffle mask corresponding to the target-specific opcode.
 /// If the mask could be calculated, returns it in \p Mask, returns the shuffle
 /// operands in \p Ops, and returns true.
@@ -33433,6 +33461,23 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
       return true;
     break;
   }
+  case X86ISD::HADD:
+  case X86ISD::HSUB:
+  case X86ISD::FHADD:
+  case X86ISD::FHSUB: {
+    APInt DemandedLHS, DemandedRHS;
+    getHorizDemandedElts(VT, DemandedElts, DemandedLHS, DemandedRHS);
+
+    APInt LHSUndef, LHSZero;
+    if (SimplifyDemandedVectorElts(Op.getOperand(0), DemandedLHS, LHSUndef,
+                                   LHSZero, TLO, Depth + 1))
+      return true;
+    APInt RHSUndef, RHSZero;
+    if (SimplifyDemandedVectorElts(Op.getOperand(1), DemandedRHS, RHSUndef,
+                                   RHSZero, TLO, Depth + 1))
+      return true;
+    break;
+  }
   case X86ISD::VTRUNC:
   case X86ISD::VTRUNCS:
   case X86ISD::VTRUNCUS: {
diff --git a/test/CodeGen/X86/horizontal-shuffle-demanded.ll b/test/CodeGen/X86/horizontal-shuffle-demanded.ll
index aedfa19384b..31d217a9d3c 100644
--- a/test/CodeGen/X86/horizontal-shuffle-demanded.ll
+++ b/test/CodeGen/X86/horizontal-shuffle-demanded.ll
@@ -10,15 +10,13 @@ define void @test_demanded_haddps_128(<4 x float> %a0, <4 x float> %a1, float *%
 ; X86-LABEL: test_demanded_haddps_128:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vbroadcastss %xmm1, %xmm1
-; X86-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
 ; X86-NEXT:    vmovss %xmm0, (%eax)
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_demanded_haddps_128:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vbroadcastss %xmm1, %xmm1
-; X64-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
+; X64-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
 ; X64-NEXT:    vmovss %xmm0, (%rdi)
 ; X64-NEXT:    retq
   %1 = shufflevector <4 x float> %a1, <4 x float> undef, <4 x i32> zeroinitializer
@@ -32,14 +30,12 @@ define void @test_demanded_hsubps_128(<4 x float> %a0, <4 x float> %a1, float *%
 ; X86-LABEL: test_demanded_hsubps_128:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vbroadcastss %xmm0, %xmm0
 ; X86-NEXT:    vhsubps %xmm1, %xmm0, %xmm0
 ; X86-NEXT:    vextractps $2, %xmm0, (%eax)
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_demanded_hsubps_128:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vbroadcastss %xmm0, %xmm0
 ; X64-NEXT:    vhsubps %xmm1, %xmm0, %xmm0
 ; X64-NEXT:    vextractps $2, %xmm0, (%rdi)
 ; X64-NEXT:    retq
@@ -54,15 +50,13 @@ define void @test_demanded_haddpd_128(<2 x double> %a0, <2 x double> %a1, double
 ; X86-LABEL: test_demanded_haddpd_128:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vmovddup {{.*#+}} xmm1 = xmm1[0,0]
-; X86-NEXT:    vhaddpd %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
 ; X86-NEXT:    vmovlpd %xmm0, (%eax)
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_demanded_haddpd_128:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vmovddup {{.*#+}} xmm1 = xmm1[0,0]
-; X64-NEXT:    vhaddpd %xmm1, %xmm0, %xmm0
+; X64-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
 ; X64-NEXT:    vmovlpd %xmm0, (%rdi)
 ; X64-NEXT:    retq
   %1 = shufflevector <2 x double> %a1, <2 x double> undef, <2 x i32> zeroinitializer
@@ -76,15 +70,13 @@ define void @test_demanded_hsubpd_128(<2 x double> %a0, <2 x double> %a1, double
 ; X86-LABEL: test_demanded_hsubpd_128:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vmovddup {{.*#+}} xmm1 = xmm1[0,0]
-; X86-NEXT:    vhsubpd %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vhsubpd %xmm0, %xmm0, %xmm0
 ; X86-NEXT:    vmovlpd %xmm0, (%eax)
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_demanded_hsubpd_128:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vmovddup {{.*#+}} xmm1 = xmm1[0,0]
-; X64-NEXT:    vhsubpd %xmm1, %xmm0, %xmm0
+; X64-NEXT:    vhsubpd %xmm0, %xmm0, %xmm0
 ; X64-NEXT:    vmovlpd %xmm0, (%rdi)
 ; X64-NEXT:    retq
   %1 = shufflevector <2 x double> %a1, <2 x double> undef, <2 x i32> zeroinitializer
@@ -98,15 +90,13 @@ define void @test_demanded_phaddd_128(<4 x i32> %a0, <4 x i32> %a1, i32 *%a2) no
 ; X86-LABEL: test_demanded_phaddd_128:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vpbroadcastd %xmm1, %xmm1
-; X86-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vphaddd %xmm0, %xmm0, %xmm0
 ; X86-NEXT:    vmovd %xmm0, (%eax)
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_demanded_phaddd_128:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vpbroadcastd %xmm1, %xmm1
-; X64-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
+; X64-NEXT:    vphaddd %xmm0, %xmm0, %xmm0
 ; X64-NEXT:    vmovd %xmm0, (%rdi)
 ; X64-NEXT:    retq
   %1 = shufflevector <4 x i32> %a1, <4 x i32> undef, <4 x i32> zeroinitializer
@@ -120,15 +110,13 @@ define void @test_demanded_phsubd_128(<4 x i32> %a0, <4 x i32> %a1, i32 *%a2) no
 ; X86-LABEL: test_demanded_phsubd_128:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vpbroadcastd %xmm1, %xmm1
-; X86-NEXT:    vphsubd %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vphsubd %xmm0, %xmm0, %xmm0
 ; X86-NEXT:    vpextrd $1, %xmm0, (%eax)
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_demanded_phsubd_128:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vpbroadcastd %xmm1, %xmm1
-; X64-NEXT:    vphsubd %xmm1, %xmm0, %xmm0
+; X64-NEXT:    vphsubd %xmm0, %xmm0, %xmm0
 ; X64-NEXT:    vpextrd $1, %xmm0, (%rdi)
 ; X64-NEXT:    retq
   %1 = shufflevector <4 x i32> %a1, <4 x i32> undef, <4 x i32> zeroinitializer
@@ -190,8 +178,7 @@ define void @test_demanded_haddps_256(<8 x float> %a0, <8 x float> %a1, float *%
 ; X86-LABEL: test_demanded_haddps_256:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vbroadcastss %xmm1, %ymm1
-; X86-NEXT:    vhaddps %ymm1, %ymm0, %ymm0
+; X86-NEXT:    vhaddps %ymm0, %ymm0, %ymm0
 ; X86-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; X86-NEXT:    vmovss %xmm0, (%eax)
 ; X86-NEXT:    vzeroupper
@@ -199,8 +186,7 @@ define void @test_demanded_haddps_256(<8 x float> %a0, <8 x float> %a1, float *%
 ;
 ; X64-LABEL: test_demanded_haddps_256:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vbroadcastss %xmm1, %ymm1
-; X64-NEXT:    vhaddps %ymm1, %ymm0, %ymm0
+; X64-NEXT:    vhaddps %ymm0, %ymm0, %ymm0
 ; X64-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; X64-NEXT:    vmovss %xmm0, (%rdi)
 ; X64-NEXT:    vzeroupper
@@ -216,7 +202,6 @@ define void @test_demanded_hsubps_256(<8 x float> %a0, <8 x float> %a1, float *%
 ; X86-LABEL: test_demanded_hsubps_256:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vbroadcastss %xmm0, %ymm0
 ; X86-NEXT:    vhsubps %ymm1, %ymm0, %ymm0
 ; X86-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; X86-NEXT:    vextractps $3, %xmm0, (%eax)
@@ -225,7 +210,6 @@ define void @test_demanded_hsubps_256(<8 x float> %a0, <8 x float> %a1, float *%
 ;
 ; X64-LABEL: test_demanded_hsubps_256:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vbroadcastss %xmm0, %ymm0
 ; X64-NEXT:    vhsubps %ymm1, %ymm0, %ymm0
 ; X64-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; X64-NEXT:    vextractps $3, %xmm0, (%rdi)
@@ -242,8 +226,7 @@ define void @test_demanded_haddpd_256(<4 x double> %a0, <4 x double> %a1, double
 ; X86-LABEL: test_demanded_haddpd_256:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vbroadcastsd %xmm1, %ymm1
-; X86-NEXT:    vhaddpd %ymm1, %ymm0, %ymm0
+; X86-NEXT:    vhaddpd %ymm0, %ymm0, %ymm0
 ; X86-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; X86-NEXT:    vmovlpd %xmm0, (%eax)
 ; X86-NEXT:    vzeroupper
@@ -251,8 +234,7 @@ define void @test_demanded_haddpd_256(<4 x double> %a0, <4 x double> %a1, double
 ;
 ; X64-LABEL: test_demanded_haddpd_256:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vbroadcastsd %xmm1, %ymm1
-; X64-NEXT:    vhaddpd %ymm1, %ymm0, %ymm0
+; X64-NEXT:    vhaddpd %ymm0, %ymm0, %ymm0
 ; X64-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; X64-NEXT:    vmovlpd %xmm0, (%rdi)
 ; X64-NEXT:    vzeroupper
@@ -268,8 +250,7 @@ define void @test_demanded_hsubpd_256(<4 x double> %a0, <4 x double> %a1, double
 ; X86-LABEL: test_demanded_hsubpd_256:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vbroadcastsd %xmm1, %ymm1
-; X86-NEXT:    vhsubpd %ymm1, %ymm0, %ymm0
+; X86-NEXT:    vhsubpd %ymm0, %ymm0, %ymm0
 ; X86-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; X86-NEXT:    vmovlpd %xmm0, (%eax)
 ; X86-NEXT:    vzeroupper
@@ -277,8 +258,7 @@ define void @test_demanded_hsubpd_256(<4 x double> %a0, <4 x double> %a1, double
 ;
 ; X64-LABEL: test_demanded_hsubpd_256:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vbroadcastsd %xmm1, %ymm1
-; X64-NEXT:    vhsubpd %ymm1, %ymm0, %ymm0
+; X64-NEXT:    vhsubpd %ymm0, %ymm0, %ymm0
 ; X64-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; X64-NEXT:    vmovlpd %xmm0, (%rdi)
 ; X64-NEXT:    vzeroupper
@@ -294,7 +274,6 @@ define void @test_demanded_phaddd_256(<8 x i32> %a0, <8 x i32> %a1, i32 *%a2) no
 ; X86-LABEL: test_demanded_phaddd_256:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vpbroadcastd %xmm0, %ymm0
 ; X86-NEXT:    vphaddd %ymm1, %ymm0, %ymm0
 ; X86-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; X86-NEXT:    vpextrd $3, %xmm0, (%eax)
@@ -303,7 +282,6 @@ define void @test_demanded_phaddd_256(<8 x i32> %a0, <8 x i32> %a1, i32 *%a2) no
 ;
 ; X64-LABEL: test_demanded_phaddd_256:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vpbroadcastd %xmm0, %ymm0
 ; X64-NEXT:    vphaddd %ymm1, %ymm0, %ymm0
 ; X64-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; X64-NEXT:    vpextrd $3, %xmm0, (%rdi)
@@ -320,8 +298,7 @@ define void @test_demanded_phsubd_256(<8 x i32> %a0, <8 x i32> %a1, i32 *%a2) no
 ; X86-LABEL: test_demanded_phsubd_256:
 ; X86:       ## %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vpbroadcastd %xmm1, %ymm1
-; X86-NEXT:    vphsubd %ymm1, %ymm0, %ymm0
+; X86-NEXT:    vphsubd %ymm0, %ymm0, %ymm0
 ; X86-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; X86-NEXT:    vpextrd $1, %xmm0, (%eax)
 ; X86-NEXT:    vzeroupper
@@ -329,8 +306,7 @@ define void @test_demanded_phsubd_256(<8 x i32> %a0, <8 x i32> %a1, i32 *%a2) no
 ;
 ; X64-LABEL: test_demanded_phsubd_256:
 ; X64:       ## %bb.0:
-; X64-NEXT:    vpbroadcastd %xmm1, %ymm1
-; X64-NEXT:    vphsubd %ymm1, %ymm0, %ymm0
+; X64-NEXT:    vphsubd %ymm0, %ymm0, %ymm0
 ; X64-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; X64-NEXT:    vpextrd $1, %xmm0, (%rdi)
 ; X64-NEXT:    vzeroupper