[DAG] Move integer setcc %x, %x folding into FoldSetCC
First step towards PR40800 - I intend to move the float case in a separate future patch.
I had to tweak the (overly reduced) thumb2 test and the x86 widening test change is annoying (no longer rematerializable) but we should address this separately.
Differential Revision: https://reviews.llvm.org/D59244
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356040 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index a87883b..fe3f2b9 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -1974,6 +1974,10 @@
break;
}
+ // We can always fold X == X for integer setcc's.
+ if (N1 == N2 && OpVT.isInteger())
+ return getBoolConstant(ISD::isTrueWhenEqual(Cond), dl, VT, OpVT);
+
if (ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2)) {
const APInt &C2 = N2C->getAPIntValue();
if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1)) {
diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 06596db..2a1b974 100644
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -3004,13 +3004,10 @@
if (N0 == N1) {
// The sext(setcc()) => setcc() optimization relies on the appropriate
// constant being emitted.
+ assert(!N0.getValueType().isInteger() &&
+ "Integer types should be handled by FoldSetCC");
bool EqTrue = ISD::isTrueWhenEqual(Cond);
-
- // We can always fold X == X for integer setcc's.
- if (N0.getValueType().isInteger())
- return DAG.getBoolConstant(EqTrue, dl, VT, OpVT);
-
unsigned UOF = ISD::getUnorderedFlavor(Cond);
if (UOF == 2) // FP operators that are undefined on NaNs.
return DAG.getBoolConstant(EqTrue, dl, VT, OpVT);
diff --git a/test/CodeGen/AArch64/fast-isel-cmp-vec.ll b/test/CodeGen/AArch64/fast-isel-cmp-vec.ll
index 4211206..6e532fd 100644
--- a/test/CodeGen/AArch64/fast-isel-cmp-vec.ll
+++ b/test/CodeGen/AArch64/fast-isel-cmp-vec.ll
@@ -24,10 +24,8 @@
define <2 x i32> @icmp_constfold_v2i32(<2 x i32> %a) {
; CHECK-LABEL: icmp_constfold_v2i32:
; CHECK: ; %bb.0:
-; CHECK-NEXT: movi.2d v[[CMP:[0-9]+]], #0xffffffffffffffff
-; CHECK-NEXT: ; %bb.1:
; CHECK-NEXT: movi.2s [[MASK:v[0-9]+]], #1
-; CHECK-NEXT: and.8b v0, v[[CMP]], [[MASK]]
+; CHECK-NEXT: and.8b v0, [[MASK]], [[MASK]]
; CHECK-NEXT: ret
%1 = icmp eq <2 x i32> %a, %a
br label %bb2
@@ -56,10 +54,9 @@
define <4 x i32> @icmp_constfold_v4i32(<4 x i32> %a) {
; CHECK-LABEL: icmp_constfold_v4i32:
; CHECK: ; %bb.0:
-; CHECK-NEXT: movi.2d v[[CMP:[0-9]+]], #0xffffffffffffffff
-; CHECK-NEXT: ; %bb.1:
; CHECK-NEXT: movi.4h [[MASK:v[0-9]+]], #1
-; CHECK-NEXT: and.8b [[ZEXT:v[0-9]+]], v[[CMP]], [[MASK]]
+; CHECK-NEXT: ; %bb.1:
+; CHECK-NEXT: and.8b [[ZEXT:v[0-9]+]], [[MASK]], [[MASK]]
; CHECK-NEXT: ushll.4s v0, [[ZEXT]], #0
; CHECK-NEXT: ret
%1 = icmp eq <4 x i32> %a, %a
@@ -87,10 +84,8 @@
define <16 x i8> @icmp_constfold_v16i8(<16 x i8> %a) {
; CHECK-LABEL: icmp_constfold_v16i8:
; CHECK: ; %bb.0:
-; CHECK-NEXT: movi.2d [[CMP:v[0-9]+]], #0xffffffffffffffff
-; CHECK-NEXT: ; %bb.1:
; CHECK-NEXT: movi.16b [[MASK:v[0-9]+]], #1
-; CHECK-NEXT: and.16b v0, [[CMP]], [[MASK]]
+; CHECK-NEXT: and.16b v0, [[MASK]], [[MASK]]
; CHECK-NEXT: ret
%1 = icmp eq <16 x i8> %a, %a
br label %bb2
diff --git a/test/CodeGen/Thumb2/cross-rc-coalescing-2.ll b/test/CodeGen/Thumb2/cross-rc-coalescing-2.ll
index 3a22c4a..b9ec0b4 100644
--- a/test/CodeGen/Thumb2/cross-rc-coalescing-2.ll
+++ b/test/CodeGen/Thumb2/cross-rc-coalescing-2.ll
@@ -64,7 +64,7 @@
%32 = fsub float %30, %31 ; <float> [#uses=1]
%33 = fsub float %10, %32 ; <float> [#uses=1]
store float %33, float* undef, align 4
- %34 = icmp slt i32 undef, undef ; <i1> [#uses=1]
+ %34 = icmp slt i32 %tmp54, %tmp80 ; <i1> [#uses=1]
br i1 %34, label %bb8, label %bb9
bb9: ; preds = %bb8
diff --git a/test/CodeGen/X86/widen_compare-1.ll b/test/CodeGen/X86/widen_compare-1.ll
index 8e20057..f37e764 100644
--- a/test/CodeGen/X86/widen_compare-1.ll
+++ b/test/CodeGen/X86/widen_compare-1.ll
@@ -7,12 +7,12 @@
define <2 x i16> @compare_v2i64_to_v2i16_unary(<2 x i16>* %src) nounwind {
; X86-LABEL: compare_v2i64_to_v2i16_unary:
; X86: # %bb.0:
-; X86-NEXT: pcmpeqd %xmm0, %xmm0
+; X86-NEXT: movaps {{.*#+}} xmm0 = [65535,0,65535,0]
; X86-NEXT: retl
;
; X64-LABEL: compare_v2i64_to_v2i16_unary:
; X64: # %bb.0:
-; X64-NEXT: pcmpeqd %xmm0, %xmm0
+; X64-NEXT: movaps {{.*#+}} xmm0 = [65535,65535]
; X64-NEXT: retq
%val = load <2 x i16>, <2 x i16>* %src, align 4
%cmp = icmp uge <2 x i16> %val, %val