[AVX-512] Update lowering for gather/scatter prefetch intrinsics to match the immedia...

Our _MM_HINT_T0/T1 constant values are 3/2 which matches gcc, but not icc or Intel documentation. Interestingly gcc had this same bug on their implementation of the gather/scatter builtins at one point too.

Fixes PR32411.

llvm-svn: 299234

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index aa74133..11b3c2e 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -20433,9 +20433,9 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget &Subtarget,
   case PREFETCH: {
     SDValue Hint = Op.getOperand(6);
     unsigned HintVal = cast<ConstantSDNode>(Hint)->getZExtValue();
-    assert((HintVal == 1 || HintVal == 2) &&
-           "Wrong prefetch hint in intrinsic: should be 1 or 2");
-    unsigned Opcode = (HintVal != 1 ? IntrData->Opc1 : IntrData->Opc0);
+    assert((HintVal == 2 || HintVal == 3) &&
+           "Wrong prefetch hint in intrinsic: should be 2 or 3");
+    unsigned Opcode = (HintVal == 2 ? IntrData->Opc1 : IntrData->Opc0);
     SDValue Chain = Op.getOperand(0);
     SDValue Mask  = Op.getOperand(2);
     SDValue Index = Op.getOperand(3);

diff --git a/llvm/test/CodeGen/X86/avx512-gather-scatter-intrin.ll b/llvm/test/CodeGen/X86/avx512-gather-scatter-intrin.ll
index 3eaee2d..4890afe 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512-gather-scatter-intrin.ll
+++ b/llvm/test/CodeGen/X86/avx512-gather-scatter-intrin.ll
@@ -285,9 +285,9 @@ define void @prefetch(<8 x i64> %ind, i8* %base) {
 ; CHECK-NEXT:    vscatterpf1qps (%rdi,%zmm0,2) {%k1}
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
-  call void @llvm.x86.avx512.gatherpf.qps.512(i8 -1, <8 x i64> %ind, i8* %base, i32 4, i32 1)
+  call void @llvm.x86.avx512.gatherpf.qps.512(i8 -1, <8 x i64> %ind, i8* %base, i32 4, i32 3)
   call void @llvm.x86.avx512.gatherpf.qps.512(i8 0, <8 x i64> %ind, i8* %base, i32 4, i32 2)
-  call void @llvm.x86.avx512.scatterpf.qps.512(i8 1, <8 x i64> %ind, i8* %base, i32 2, i32 1)
+  call void @llvm.x86.avx512.scatterpf.qps.512(i8 1, <8 x i64> %ind, i8* %base, i32 2, i32 3)
   call void @llvm.x86.avx512.scatterpf.qps.512(i8 120, <8 x i64> %ind, i8* %base, i32 2, i32 2)
   ret void
 }