setTargetDAGCombine(ISD::VECTOR_SHUFFLE);
setTargetDAGCombine(ISD::EXTRACT_VECTOR_ELT);
setTargetDAGCombine(ISD::BUILD_VECTOR);
+ setTargetDAGCombine(ISD::VSELECT);
setTargetDAGCombine(ISD::SELECT);
setTargetDAGCombine(ISD::SHL);
setTargetDAGCombine(ISD::SRA);
return SDValue();
}
-/// PerformSELECTCombine - Do target-specific dag combines on SELECT nodes.
+/// PerformSELECTCombine - Do target-specific dag combines on SELECT and VSELECT
+/// nodes.
static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG,
const X86Subtarget *Subtarget) {
DebugLoc DL = N->getDebugLoc();
// instructions match the semantics of the common C idiom x<y?x:y but not
// x<=y?x:y, because of how they handle negative zero (which can be
// ignored in unsafe-math mode).
- if (Subtarget->hasXMMInt() &&
- (LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64) &&
- Cond.getOpcode() == ISD::SETCC) {
+ if (Subtarget->hasXMMInt() && Cond.getOpcode() == ISD::SETCC &&
+ (LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64 ||
+ LHS.getValueType() == MVT::v4f32 || LHS.getValueType() == MVT::v2f64)) {
ISD::CondCode CC = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
unsigned Opcode = 0;
default: break;
case ISD::EXTRACT_VECTOR_ELT:
return PerformEXTRACT_VECTOR_ELTCombine(N, DAG, *this);
+ case ISD::VSELECT:
case ISD::SELECT: return PerformSELECTCombine(N, DAG, Subtarget);
case X86ISD::CMOV: return PerformCMOVCombine(N, DAG, DCI);
case ISD::ADD: return OptimizeConditionalInDecrement(N, DAG);
-; RUN: llc < %s -march=x86-64 -asm-verbose=false -join-physregs | FileCheck %s
-; RUN: llc < %s -march=x86-64 -asm-verbose=false -join-physregs -enable-unsafe-fp-math -enable-no-nans-fp-math | FileCheck -check-prefix=UNSAFE %s
-; RUN: llc < %s -march=x86-64 -asm-verbose=false -join-physregs -enable-no-nans-fp-math | FileCheck -check-prefix=FINITE %s
+; RUN: llc < %s -march=x86-64 -asm-verbose=false -join-physregs -promote-elements | FileCheck %s
+; RUN: llc < %s -march=x86-64 -asm-verbose=false -join-physregs -enable-unsafe-fp-math -enable-no-nans-fp-math -promote-elements | FileCheck -check-prefix=UNSAFE %s
+; RUN: llc < %s -march=x86-64 -asm-verbose=false -join-physregs -enable-no-nans-fp-math -promote-elements | FileCheck -check-prefix=FINITE %s
; Some of these patterns can be matched as SSE min or max. Some of
; then can be matched provided that the operands are swapped.
%x_addr.0 = select i1 %0, double 3.000000e+03, double %x ; <double> [#uses=1]
ret double %x_addr.0
}
+
+; UNSAFE: maxpd:
+; UNSAFE: maxpd
+define <2 x double> @maxpd(<2 x double> %x, <2 x double> %y) {
+ %max_is_x = fcmp oge <2 x double> %x, %y
+ %max = select <2 x i1> %max_is_x, <2 x double> %x, <2 x double> %y
+ ret <2 x double> %max
+}
+
+; UNSAFE: minpd:
+; UNSAFE: minpd
+define <2 x double> @minpd(<2 x double> %x, <2 x double> %y) {
+ %min_is_x = fcmp ole <2 x double> %x, %y
+ %min = select <2 x i1> %min_is_x, <2 x double> %x, <2 x double> %y
+ ret <2 x double> %min
+}
+
+; UNSAFE: maxps:
+; UNSAFE: maxps
+define <4 x float> @maxps(<4 x float> %x, <4 x float> %y) {
+ %max_is_x = fcmp oge <4 x float> %x, %y
+ %max = select <4 x i1> %max_is_x, <4 x float> %x, <4 x float> %y
+ ret <4 x float> %max
+}
+
+; UNSAFE: minps:
+; UNSAFE: minps
+define <4 x float> @minps(<4 x float> %x, <4 x float> %y) {
+ %min_is_x = fcmp ole <4 x float> %x, %y
+ %min = select <4 x i1> %min_is_x, <4 x float> %x, <4 x float> %y
+ ret <4 x float> %min
+}
; CHECK: B
define <2 x double> @B(<2 x double> %x, <2 x double> %y) {
- ; CHECK: cmpltpd
+ ; CHECK: cmpnlepd
; CHECK: blendvpd
- %max_is_x = fcmp ogt <2 x double> %x, %y
- %max = select <2 x i1> %max_is_x, <2 x double> %x, <2 x double> %y
- ret <2 x double> %max
+ %min_is_x = fcmp ult <2 x double> %x, %y
+ %min = select <2 x i1> %min_is_x, <2 x double> %x, <2 x double> %y
+ ret <2 x double> %min
}