summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp18
-rw-r--r--llvm/test/CodeGen/X86/fp-logic.ll15
2 files changed, 22 insertions, 11 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index fbead736ec0..a19ffcfbe54 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -24447,6 +24447,24 @@ static SDValue PerformAndCombine(SDNode *N, SelectionDAG &DAG,
}
} // BEXTR
+ // If both input operands are being cast from floating point types,
+ // try to convert this into a floating point logic node to avoid
+ // unnecessary moves from SSE to integer registers.
+ // FIXME: Split this into a helper function, so it can also be used with
+ // or/xor combining.
+ if (N0.getOpcode() == ISD::BITCAST && N1.getOpcode() == ISD::BITCAST &&
+ ((Subtarget->hasSSE1() && VT == MVT::i32) ||
+ (Subtarget->hasSSE2() && VT == MVT::i64))) {
+ SDValue N00 = N0.getOperand(0);
+ SDValue N10 = N1.getOperand(0);
+ EVT N00Type = N00.getValueType();
+ EVT N10Type = N10.getValueType();
+ if (N00Type.isFloatingPoint() && N10Type.isFloatingPoint()) {
+ SDValue FLogic = DAG.getNode(X86ISD::FAND, DL, N00Type, N00, N10);
+ return DAG.getBitcast(VT, FLogic);
+ }
+ }
+
return SDValue();
}
diff --git a/llvm/test/CodeGen/X86/fp-logic.ll b/llvm/test/CodeGen/X86/fp-logic.ll
index d047f7742c8..0b714ffe36f 100644
--- a/llvm/test/CodeGen/X86/fp-logic.ll
+++ b/llvm/test/CodeGen/X86/fp-logic.ll
@@ -142,9 +142,8 @@ define float @f8(float %x) {
define i32 @f9(float %x, float %y) {
; CHECK-LABEL: f9:
; CHECK: # BB#0:
-; CHECK-NEXT: movd %xmm0, %ecx
-; CHECK-NEXT: movd %xmm1, %eax
-; CHECK-NEXT: andl %ecx, %eax
+; CHECK-NEXT: andps %xmm1, %xmm0
+; CHECK-NEXT: movd %xmm0, %eax
; CHECK-NEXT: retq
%bc1 = bitcast float %x to i32
@@ -158,10 +157,7 @@ define i32 @f9(float %x, float %y) {
define float @f10(float %x, float %y) {
; CHECK-LABEL: f10:
; CHECK: # BB#0:
-; CHECK-NEXT: movd %xmm0, %eax
-; CHECK-NEXT: movd %xmm1, %ecx
-; CHECK-NEXT: andl %eax, %ecx
-; CHECK-NEXT: movd %ecx, %xmm0
+; CHECK-NEXT: andps %xmm1, %xmm0
; CHECK-NEXT: retq
%bc1 = bitcast float %x to i32
@@ -208,10 +204,7 @@ define float @xor(float %x, float %y) {
define double @doubles(double %x, double %y) {
; CHECK-LABEL: doubles:
; CHECK: # BB#0:
-; CHECK-NEXT: movd %xmm0, %rax
-; CHECK-NEXT: movd %xmm1, %rcx
-; CHECK-NEXT: andq %rax, %rcx
-; CHECK-NEXT: movd %rcx, %xmm0
+; CHECK-NEXT: andpd %xmm1, %xmm0
; CHECK-NEXT: retq
%bc1 = bitcast double %x to i64
OpenPOWER on IntegriCloud