[llvm] r302546 - VX512] Only look at lower bit in constant scalar masks
Guy Blank via llvm-commits
llvm-commits at lists.llvm.org
Tue May 9 09:16:48 PDT 2017
Author: guyblank
Date: Tue May 9 11:16:48 2017
New Revision: 302546
URL: http://llvm.org/viewvc/llvm-project?rev=302546&view=rev
Log:
VX512] Only look at lower bit in constant scalar masks
for scalar masked instructions only the lower bit of the mask is relevant. so for constant masks we should either do an unmasked operation or no operation, depending on the value of the lower bit.
This patch handles cases where the lower bit is '1'.
Differential Revision: https://reviews.llvm.org/D32805
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
llvm/trunk/test/CodeGen/X86/avx512-scalar_mask.ll
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=302546&r1=302545&r2=302546&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Tue May 9 11:16:48 2017
@@ -19021,8 +19021,10 @@ static SDValue getScalarMaskingNode(SDVa
SDValue PreservedSrc,
const X86Subtarget &Subtarget,
SelectionDAG &DAG) {
- if (isAllOnesConstant(Mask))
- return Op;
+
+ if (auto *MaskConst = dyn_cast<ConstantSDNode>(Mask))
+ if (MaskConst->getZExtValue() & 0x1)
+ return Op;
MVT VT = Op.getSimpleValueType();
SDLoc dl(Op);
Modified: llvm/trunk/test/CodeGen/X86/avx512-scalar_mask.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-scalar_mask.ll?rev=302546&r1=302545&r2=302546&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/avx512-scalar_mask.ll (original)
+++ llvm/trunk/test/CodeGen/X86/avx512-scalar_mask.ll Tue May 9 11:16:48 2017
@@ -26,6 +26,7 @@ define <4 x float>@test_var_maskz(<4 x f
ret < 4 x float> %res
}
+; FIXME: we should just return %xmm0 here.
define <4 x float>@test_const0_mask(<4 x float> %v0, <4 x float> %v1, <4 x float> %v2) {
; CHECK-LABEL: test_const0_mask:
; CHECK: ## BB#0:
@@ -36,6 +37,7 @@ define <4 x float>@test_const0_mask(<4 x
ret < 4 x float> %res
}
+; FIXME: we should zero the lower element of xmm0 and return it.
define <4 x float>@test_const0_maskz(<4 x float> %v0, <4 x float> %v1, <4 x float> %v2) {
; CHECK-LABEL: test_const0_maskz:
; CHECK: ## BB#0:
@@ -46,6 +48,7 @@ define <4 x float>@test_const0_maskz(<4
ret < 4 x float> %res
}
+; FIXME: we should just return %xmm0 here.
define <4 x float>@test_const2_mask(<4 x float> %v0, <4 x float> %v1, <4 x float> %v2) {
; CHECK-LABEL: test_const2_mask:
; CHECK: ## BB#0:
@@ -56,6 +59,7 @@ define <4 x float>@test_const2_mask(<4 x
ret < 4 x float> %res
}
+; FIXME: we should zero the lower element of xmm0 and return it.
define <4 x float>@test_const2_maskz(<4 x float> %v0, <4 x float> %v1, <4 x float> %v2) {
; CHECK-LABEL: test_const2_maskz:
; CHECK: ## BB#0:
@@ -87,9 +91,7 @@ define <4 x float>@test_const_allone_mas
define <4 x float>@test_const_3_mask(<4 x float> %v0, <4 x float> %v1, <4 x float> %v2) {
; CHECK-LABEL: test_const_3_mask:
; CHECK: ## BB#0:
-; CHECK-NEXT: kxnorw %k0, %k0, %k0
-; CHECK-NEXT: kshiftrw $15, %k0, %k1
-; CHECK-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 {%k1}
+; CHECK-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
%res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %v0,<4 x float> %v1, <4 x float> %v2, i8 3, i32 4)
ret < 4 x float> %res
@@ -98,9 +100,7 @@ define <4 x float>@test_const_3_mask(<4
define <4 x float>@test_const_3_maskz(<4 x float> %v0, <4 x float> %v1, <4 x float> %v2) {
; CHECK-LABEL: test_const_3_maskz:
; CHECK: ## BB#0:
-; CHECK-NEXT: kxnorw %k0, %k0, %k0
-; CHECK-NEXT: kshiftrw $15, %k0, %k1
-; CHECK-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 {%k1} {z}
+; CHECK-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0
; CHECK-NEXT: retq
%res = call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %v0,<4 x float> %v1, <4 x float> %v2, i8 3, i32 4)
ret < 4 x float> %res
More information about the llvm-commits
mailing list