<div dir="ltr">On Wed, Jul 23, 2014 at 1:41 PM, Jim Grosbach <span dir="ltr"><<a href="mailto:grosbach@apple.com" target="_blank">grosbach@apple.com</a>></span> wrote:<br><div class="gmail_extra"><div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Author: grosbach<br>
Date: Wed Jul 23 15:41:38 2014<br>
New Revision: 213799<br>
<br>
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=213799&view=rev" target="_blank">http://llvm.org/viewvc/llvm-project?rev=213799&view=rev</a><br>
Log:<br>
X86: restrict combine to when type sizes are safe.<br>
<br>
The folding of unary operations through a vector compare and mask operation<br>
is only safe if the unary operation result is of the same size as its input.<br>
For example, it's not safe for [su]itofp from v4i32 to v4f64.<br>
<br>
Modified:<br>
llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp<br>
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp<br>
llvm/trunk/test/CodeGen/AArch64/arm64-setcc-int-to-fp-combine.ll<br>
llvm/trunk/test/CodeGen/X86/x86-setcc-int-to-fp-combine.ll<br></blockquote><div><br></div><div>Should this get pulled into the 3.5 branch?</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Modified: llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp?rev=213799&r1=213798&r2=213799&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp?rev=213799&r1=213798&r2=213799&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp (original)<br>
+++ llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp Wed Jul 23 15:41:38 2014<br>
@@ -6492,11 +6492,13 @@ static SDValue performVectorCompareAndMa<br>
// AND(VECTOR_CMP(x,y), constant2)<br>
// constant2 = UNARYOP(constant)<br>
<br>
- // Early exit if this isn't a vector operation or if the operand of the<br>
- // unary operation isn't a bitwise AND.<br>
+ // Early exit if this isn't a vector operation, the operand of the<br>
+ // unary operation isn't a bitwise AND, or if the sizes of the operations<br>
+ // aren't the same.<br>
EVT VT = N->getValueType(0);<br>
if (!VT.isVector() || N->getOperand(0)->getOpcode() != ISD::AND ||<br>
- N->getOperand(0)->getOperand(0)->getOpcode() != ISD::SETCC)<br>
+ N->getOperand(0)->getOperand(0)->getOpcode() != ISD::SETCC ||<br>
+ VT.getSizeInBits() != N->getOperand(0)->getValueType(0).getSizeInBits())<br>
return SDValue();<br>
<br>
// Now check that the other operand of the AND is a constant splat. We could<br>
<br>
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=213799&r1=213798&r2=213799&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=213799&r1=213798&r2=213799&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)<br>
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Wed Jul 23 15:41:38 2014<br>
@@ -21806,11 +21806,13 @@ static SDValue performVectorCompareAndMa<br>
// AND(VECTOR_CMP(x,y), constant2)<br>
// constant2 = UNARYOP(constant)<br>
<br>
- // Early exit if this isn't a vector operation or if the operand of the<br>
- // unary operation isn't a bitwise AND.<br>
+ // Early exit if this isn't a vector operation, the operand of the<br>
+ // unary operation isn't a bitwise AND, or if the sizes of the operations<br>
+ // aren't the same.<br>
EVT VT = N->getValueType(0);<br>
if (!VT.isVector() || N->getOperand(0)->getOpcode() != ISD::AND ||<br>
- N->getOperand(0)->getOperand(0)->getOpcode() != ISD::SETCC)<br>
+ N->getOperand(0)->getOperand(0)->getOpcode() != ISD::SETCC ||<br>
+ VT.getSizeInBits() != N->getOperand(0)->getValueType(0).getSizeInBits())<br>
return SDValue();<br>
<br>
// Now check that the other operand of the AND is a constant splat. We could<br>
<br>
Modified: llvm/trunk/test/CodeGen/AArch64/arm64-setcc-int-to-fp-combine.ll<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/arm64-setcc-int-to-fp-combine.ll?rev=213799&r1=213798&r2=213799&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/arm64-setcc-int-to-fp-combine.ll?rev=213799&r1=213798&r2=213799&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/test/CodeGen/AArch64/arm64-setcc-int-to-fp-combine.ll (original)<br>
+++ llvm/trunk/test/CodeGen/AArch64/arm64-setcc-int-to-fp-combine.ll Wed Jul 23 15:41:38 2014<br>
@@ -11,3 +11,17 @@ define <4 x float> @foo(<4 x float> %val<br>
%result = sitofp <4 x i32> %ext to <4 x float><br>
ret <4 x float> %result<br>
}<br>
+; Make sure the operation doesn't try to get folded when the sizes don't match,<br>
+; as that ends up crashing later when trying to form a bitcast operation for<br>
+; the folded nodes.<br>
+define void @foo1(<4 x float> %val, <4 x float> %test, <4 x double>* %p) nounwind {<br>
+; CHECK-LABEL: foo1:<br>
+; CHECK: movi.4s<br>
+; CHECK: scvtf.2d<br>
+; CHECK: scvtf.2d<br>
+ %cmp = fcmp oeq <4 x float> %val, %test<br>
+ %ext = zext <4 x i1> %cmp to <4 x i32><br>
+ %result = sitofp <4 x i32> %ext to <4 x double><br>
+ store <4 x double> %result, <4 x double>* %p<br>
+ ret void<br>
+}<br>
<br>
Modified: llvm/trunk/test/CodeGen/X86/x86-setcc-int-to-fp-combine.ll<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/x86-setcc-int-to-fp-combine.ll?rev=213799&r1=213798&r2=213799&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/x86-setcc-int-to-fp-combine.ll?rev=213799&r1=213798&r2=213799&view=diff</a><br>
==============================================================================<br>
--- llvm/trunk/test/CodeGen/X86/x86-setcc-int-to-fp-combine.ll (original)<br>
+++ llvm/trunk/test/CodeGen/X86/x86-setcc-int-to-fp-combine.ll Wed Jul 23 15:41:38 2014<br>
@@ -17,14 +17,38 @@ define <4 x float> @foo(<4 x float> %val<br>
ret <4 x float> %result<br>
}<br>
<br>
-define void @bar(<4 x float>* noalias %result) nounwind {<br>
+; Make sure the operation doesn't try to get folded when the sizes don't match,<br>
+; as that ends up crashing later when trying to form a bitcast operation for<br>
+; the folded nodes.<br>
+define void @foo1(<4 x float> %val, <4 x float> %test, <4 x double>* %p) nounwind {<br>
; CHECK-LABEL: LCPI1_0:<br>
+; CHECK-NEXT: .long 1 ## 0x1<br>
+; CHECK-NEXT: .long 1 ## 0x1<br>
+; CHECK-NEXT: .long 1 ## 0x1<br>
+; CHECK-NEXT: .long 1 ## 0x1<br>
+; CHECK-LABEL: foo1:<br>
+; FIXME: The operation gets scalarized. If/when the compiler learns to better<br>
+; use [V]CVTDQ2PD, this will need updated.<br>
+; CHECK: cvtsi2sdq<br>
+; CHECK: cvtsi2sdq<br>
+; CHECK: cvtsi2sdq<br>
+; CHECK: cvtsi2sdq<br>
+ %cmp = fcmp oeq <4 x float> %val, %test<br>
+ %ext = zext <4 x i1> %cmp to <4 x i32><br>
+ %result = sitofp <4 x i32> %ext to <4 x double><br>
+ store <4 x double> %result, <4 x double>* %p<br>
+ ret void<br>
+}<br>
+<br>
+; Also test the general purpose constant folding of int->fp.<br>
+define void @foo2(<4 x float>* noalias %result) nounwind {<br>
+; CHECK-LABEL: LCPI2_0:<br>
; CHECK-NEXT: .long 1082130432 ## float 4.000000e+00<br>
; CHECK-NEXT: .long 1084227584 ## float 5.000000e+00<br>
; CHECK-NEXT: .long 1086324736 ## float 6.000000e+00<br>
; CHECK-NEXT: .long 1088421888 ## float 7.000000e+00<br>
-; CHECK-LABEL: bar:<br>
-; CHECK: movaps LCPI1_0(%rip), %xmm0<br>
+; CHECK-LABEL: foo2:<br>
+; CHECK: movaps LCPI2_0(%rip), %xmm0<br>
<br>
%val = uitofp <4 x i32> <i32 4, i32 5, i32 6, i32 7> to <4 x float><br>
store <4 x float> %val, <4 x float>* %result<br>
<br>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
</blockquote></div><br><br clear="all"><div><br></div>-- <br>Saleem Abdulrasool<br>compnerd (at) compnerd (dot) org
</div></div>