[llvm] 17fc78e - [DAGCombiner] Change foldAndOrOfSETCC() to optimize and/or patterns with floating points.
Konstantina Mitropoulou via llvm-commits
llvm-commits at lists.llvm.org
Thu Aug 31 11:37:09 PDT 2023
Author: Konstantina Mitropoulou
Date: 2023-08-31T11:36:50-07:00
New Revision: 17fc78e7a4e178482650f2a1c18954823c5cd4eb
URL: https://github.com/llvm/llvm-project/commit/17fc78e7a4e178482650f2a1c18954823c5cd4eb
DIFF: https://github.com/llvm/llvm-project/commit/17fc78e7a4e178482650f2a1c18954823c5cd4eb.diff
LOG: [DAGCombiner] Change foldAndOrOfSETCC() to optimize and/or patterns with floating points.
This reverts commit 48fa79a503a7cf380f98b6335fbd349afae1bd86.
Reviewed By: brooksmoses
Differential Revision: https://reviews.llvm.org/D159240
Added:
llvm/test/CodeGen/AArch64/combine_andor_with_cmps.ll
Modified:
llvm/include/llvm/CodeGen/ISDOpcodes.h
llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/test/CodeGen/AMDGPU/combine_andor_with_cmps.ll
llvm/test/CodeGen/AMDGPU/fma.f16.ll
llvm/test/CodeGen/AMDGPU/or.ll
Removed:
################################################################################
diff --git a/llvm/include/llvm/CodeGen/ISDOpcodes.h b/llvm/include/llvm/CodeGen/ISDOpcodes.h
index 12457c09fdb8bc..6da28ee93dc83d 100644
--- a/llvm/include/llvm/CodeGen/ISDOpcodes.h
+++ b/llvm/include/llvm/CodeGen/ISDOpcodes.h
@@ -1544,6 +1544,12 @@ inline bool isIntEqualitySetCC(CondCode Code) {
return Code == SETEQ || Code == SETNE;
}
+/// Return true if this is a setcc instruction that performs an equality
+/// comparison when used with floating point operands.
+inline bool isFPEqualitySetCC(CondCode Code) {
+ return Code == SETOEQ || Code == SETONE || Code == SETUEQ || Code == SETUNE;
+}
+
/// Return true if the specified condition returns true if the two operands to
/// the condition are equal. Note that if one of the two operands is a NaN,
/// this value is meaningless.
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 822351b549360c..8e896dbe7962ed 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -6042,6 +6042,72 @@ SDValue DAGCombiner::foldLogicOfSetCCs(bool IsAnd, SDValue N0, SDValue N1,
return SDValue();
}
+static bool arebothOperandsNotSNan(SDValue Operand1, SDValue Operand2,
+ SelectionDAG &DAG) {
+ return DAG.isKnownNeverSNaN(Operand2) && DAG.isKnownNeverSNaN(Operand1);
+}
+
+static bool arebothOperandsNotNan(SDValue Operand1, SDValue Operand2,
+ SelectionDAG &DAG) {
+ return DAG.isKnownNeverNaN(Operand2) && DAG.isKnownNeverNaN(Operand1);
+}
+
+static unsigned getMinMaxOpcodeForFP(SDValue Operand1, SDValue Operand2,
+ ISD::CondCode CC, unsigned OrAndOpcode,
+ SelectionDAG &DAG,
+ bool isFMAXNUMFMINNUM_IEEE,
+ bool isFMAXNUMFMINNUM) {
+ // The optimization cannot be applied for all the predicates because
+ // of the way FMINNUM/FMAXNUM and FMINNUM_IEEE/FMAXNUM_IEEE handle
+ // NaNs. For FMINNUM_IEEE/FMAXNUM_IEEE, the optimization cannot be
+ // applied at all if one of the operands is a signaling NaN.
+
+ // It is safe to use FMINNUM_IEEE/FMAXNUM_IEEE if all the operands
+ // are non NaN values.
+ if (((CC == ISD::SETLT || CC == ISD::SETLE) && (OrAndOpcode == ISD::OR)) ||
+ ((CC == ISD::SETGT || CC == ISD::SETGE) && (OrAndOpcode == ISD::AND)))
+ return arebothOperandsNotNan(Operand1, Operand2, DAG) &&
+ isFMAXNUMFMINNUM_IEEE
+ ? ISD::FMINNUM_IEEE
+ : ISD::DELETED_NODE;
+ else if (((CC == ISD::SETGT || CC == ISD::SETGE) &&
+ (OrAndOpcode == ISD::OR)) ||
+ ((CC == ISD::SETLT || CC == ISD::SETLE) &&
+ (OrAndOpcode == ISD::AND)))
+ return arebothOperandsNotNan(Operand1, Operand2, DAG) &&
+ isFMAXNUMFMINNUM_IEEE
+ ? ISD::FMAXNUM_IEEE
+ : ISD::DELETED_NODE;
+ // Both FMINNUM/FMAXNUM and FMINNUM_IEEE/FMAXNUM_IEEE handle quiet
+ // NaNs in the same way. But, FMINNUM/FMAXNUM and FMINNUM_IEEE/
+ // FMAXNUM_IEEE handle signaling NaNs
diff erently. If we cannot prove
+ // that there are not any sNaNs, then the optimization is not valid
+ // for FMINNUM_IEEE/FMAXNUM_IEEE. In the presence of sNaNs, we apply
+ // the optimization using FMINNUM/FMAXNUM for the following cases. If
+ // we can prove that we do not have any sNaNs, then we can do the
+ // optimization using FMINNUM_IEEE/FMAXNUM_IEEE for the following
+ // cases.
+ else if (((CC == ISD::SETOLT || CC == ISD::SETOLE) &&
+ (OrAndOpcode == ISD::OR)) ||
+ ((CC == ISD::SETUGT || CC == ISD::SETUGE) &&
+ (OrAndOpcode == ISD::AND)))
+ return isFMAXNUMFMINNUM ? ISD::FMINNUM
+ : arebothOperandsNotSNan(Operand1, Operand2, DAG) &&
+ isFMAXNUMFMINNUM_IEEE
+ ? ISD::FMINNUM_IEEE
+ : ISD::DELETED_NODE;
+ else if (((CC == ISD::SETOGT || CC == ISD::SETOGE) &&
+ (OrAndOpcode == ISD::OR)) ||
+ ((CC == ISD::SETULT || CC == ISD::SETULE) &&
+ (OrAndOpcode == ISD::AND)))
+ return isFMAXNUMFMINNUM ? ISD::FMAXNUM
+ : arebothOperandsNotSNan(Operand1, Operand2, DAG) &&
+ isFMAXNUMFMINNUM_IEEE
+ ? ISD::FMAXNUM_IEEE
+ : ISD::DELETED_NODE;
+ return ISD::DELETED_NODE;
+}
+
static SDValue foldAndOrOfSETCC(SDNode *LogicOp, SelectionDAG &DAG) {
using AndOrSETCCFoldKind = TargetLowering::AndOrSETCCFoldKind;
assert(
@@ -6083,12 +6149,21 @@ static SDValue foldAndOrOfSETCC(SDNode *LogicOp, SelectionDAG &DAG) {
// The optimization does not work for `==` or `!=` .
// The two comparisons should have either the same predicate or the
// predicate of one of the comparisons is the opposite of the other one.
- if (OpVT.isInteger() && !ISD::isIntEqualitySetCC(CCL) &&
- (CCL == CCR || CCL == ISD::getSetCCSwappedOperands(CCR)) &&
- TLI.isOperationLegal(ISD::UMAX, OpVT) &&
- TLI.isOperationLegal(ISD::SMAX, OpVT) &&
- TLI.isOperationLegal(ISD::UMIN, OpVT) &&
- TLI.isOperationLegal(ISD::SMIN, OpVT)) {
+ bool isFMAXNUMFMINNUM_IEEE = TLI.isOperationLegal(ISD::FMAXNUM_IEEE, OpVT) &&
+ TLI.isOperationLegal(ISD::FMINNUM_IEEE, OpVT);
+ bool isFMAXNUMFMINNUM = TLI.isOperationLegalOrCustom(ISD::FMAXNUM, OpVT) &&
+ TLI.isOperationLegalOrCustom(ISD::FMINNUM, OpVT);
+ if (((OpVT.isInteger() && TLI.isOperationLegal(ISD::UMAX, OpVT) &&
+ TLI.isOperationLegal(ISD::SMAX, OpVT) &&
+ TLI.isOperationLegal(ISD::UMIN, OpVT) &&
+ TLI.isOperationLegal(ISD::SMIN, OpVT)) ||
+ (OpVT.isFloatingPoint() &&
+ (isFMAXNUMFMINNUM_IEEE || isFMAXNUMFMINNUM))) &&
+ !ISD::isIntEqualitySetCC(CCL) && !ISD::isFPEqualitySetCC(CCL) &&
+ CCL != ISD::SETFALSE && CCL != ISD::SETO && CCL != ISD::SETUO &&
+ CCL != ISD::SETTRUE &&
+ (CCL == CCR || CCL == ISD::getSetCCSwappedOperands(CCR))) {
+
SDValue CommonValue, Operand1, Operand2;
ISD::CondCode CC = ISD::SETCC_INVALID;
if (CCL == CCR) {
@@ -6126,19 +6201,26 @@ static SDValue foldAndOrOfSETCC(SDNode *LogicOp, SelectionDAG &DAG) {
CC = ISD::SETCC_INVALID;
if (CC != ISD::SETCC_INVALID) {
- unsigned NewOpcode;
+ unsigned NewOpcode = ISD::DELETED_NODE;
bool IsSigned = isSignedIntSetCC(CC);
- bool IsLess = (CC == ISD::SETLE || CC == ISD::SETULE ||
- CC == ISD::SETLT || CC == ISD::SETULT);
- bool IsOr = (LogicOp->getOpcode() == ISD::OR);
- if (IsLess == IsOr)
- NewOpcode = IsSigned ? ISD::SMIN : ISD::UMIN;
- else
- NewOpcode = IsSigned ? ISD::SMAX : ISD::UMAX;
-
- SDValue MinMaxValue =
- DAG.getNode(NewOpcode, DL, OpVT, Operand1, Operand2);
- return DAG.getSetCC(DL, VT, MinMaxValue, CommonValue, CC);
+ if (OpVT.isInteger()) {
+ bool IsLess = (CC == ISD::SETLE || CC == ISD::SETULE ||
+ CC == ISD::SETLT || CC == ISD::SETULT);
+ bool IsOr = (LogicOp->getOpcode() == ISD::OR);
+ if (IsLess == IsOr)
+ NewOpcode = IsSigned ? ISD::SMIN : ISD::UMIN;
+ else
+ NewOpcode = IsSigned ? ISD::SMAX : ISD::UMAX;
+ } else if (OpVT.isFloatingPoint())
+ NewOpcode =
+ getMinMaxOpcodeForFP(Operand1, Operand2, CC, LogicOp->getOpcode(),
+ DAG, isFMAXNUMFMINNUM_IEEE, isFMAXNUMFMINNUM);
+
+ if (NewOpcode != ISD::DELETED_NODE) {
+ SDValue MinMaxValue =
+ DAG.getNode(NewOpcode, DL, OpVT, Operand1, Operand2);
+ return DAG.getSetCC(DL, VT, MinMaxValue, CommonValue, CC);
+ }
}
}
diff --git a/llvm/test/CodeGen/AArch64/combine_andor_with_cmps.ll b/llvm/test/CodeGen/AArch64/combine_andor_with_cmps.ll
new file mode 100644
index 00000000000000..783683cf7e8442
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/combine_andor_with_cmps.ll
@@ -0,0 +1,74 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mcpu=neoverse-n1 -verify-machineinstrs < %s | FileCheck %s
+
+; The tests check the following optimization of DAGCombiner for AArch64:
+; CMP(A,C)||CMP(B,C) => CMP(MIN/MAX(A,B), C)
+; CMP(A,C)&&CMP(B,C) => CMP(MIN/MAX(A,B), C)
+
+define i1 @test1(float %arg1, float %arg2, float %arg3) #0 {
+; CHECK-LABEL: test1:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fminnm s0, s0, s1
+; CHECK-NEXT: fcmp s0, s2
+; CHECK-NEXT: cset w0, mi
+; CHECK-NEXT: ret
+ %cmp1 = fcmp olt float %arg1, %arg3
+ %cmp2 = fcmp olt float %arg2, %arg3
+ %or1 = or i1 %cmp1, %cmp2
+ ret i1 %or1
+}
+
+define i1 @test2(double %arg1, double %arg2, double %arg3) #0 {
+; CHECK-LABEL: test2:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmaxnm d0, d0, d1
+; CHECK-NEXT: fcmp d0, d2
+; CHECK-NEXT: cset w0, gt
+; CHECK-NEXT: ret
+ %cmp1 = fcmp ogt double %arg1, %arg3
+ %cmp2 = fcmp ogt double %arg2, %arg3
+ %or1 = or i1 %cmp1, %cmp2
+ ret i1 %or1
+}
+
+; It is illegal to apply the optimization in the following two test cases
+; because FMINNUM_IEEE and FMAXNUM_IEEE are not supported.
+
+define i1 @test3(float %arg1, float %arg2, float %arg3) {
+; CHECK-LABEL: test3:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmov s3, #1.00000000
+; CHECK-NEXT: fadd s0, s0, s3
+; CHECK-NEXT: fmov s3, #2.00000000
+; CHECK-NEXT: fadd s1, s1, s3
+; CHECK-NEXT: fcmp s1, s2
+; CHECK-NEXT: fccmp s0, s2, #0, lt
+; CHECK-NEXT: cset w0, lt
+; CHECK-NEXT: ret
+ %add1 = fadd nnan float %arg1, 1.0
+ %add2 = fadd nnan float %arg2, 2.0
+ %cmp1 = fcmp nnan olt float %add1, %arg3
+ %cmp2 = fcmp nnan olt float %add2, %arg3
+ %or1 = and i1 %cmp1, %cmp2
+ ret i1 %or1
+}
+
+define i1 @test4(float %arg1, float %arg2, float %arg3) {
+; CHECK-LABEL: test4:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmov s3, #1.00000000
+; CHECK-NEXT: fadd s0, s0, s3
+; CHECK-NEXT: fmov s3, #2.00000000
+; CHECK-NEXT: fadd s1, s1, s3
+; CHECK-NEXT: fcmp s1, s2
+; CHECK-NEXT: fccmp s0, s2, #4, gt
+; CHECK-NEXT: cset w0, gt
+; CHECK-NEXT: ret
+ %add1 = fadd nnan float %arg1, 1.0
+ %add2 = fadd nnan float %arg2, 2.0
+ %cmp1 = fcmp nnan ogt float %add1, %arg3
+ %cmp2 = fcmp nnan ogt float %add2, %arg3
+ %or1 = and i1 %cmp1, %cmp2
+ ret i1 %or1
+}
+
diff --git a/llvm/test/CodeGen/AMDGPU/combine_andor_with_cmps.ll b/llvm/test/CodeGen/AMDGPU/combine_andor_with_cmps.ll
index 0926a2295f0a57..6efbd6ce87385e 100644
--- a/llvm/test/CodeGen/AMDGPU/combine_andor_with_cmps.ll
+++ b/llvm/test/CodeGen/AMDGPU/combine_andor_with_cmps.ll
@@ -1,18 +1,19 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
-; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -amdgpu-enable-delay-alu=0 < %s | FileCheck %s
+; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -amdgpu-enable-delay-alu=0 < %s | FileCheck %s -check-prefixes=GCN,GFX11
+; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -amdgpu-enable-delay-alu=0 -enable-no-nans-fp-math < %s | FileCheck %s -check-prefixes=GCN,GFX11NONANS
; The tests check the following optimization of DAGCombiner:
; CMP(A,C)||CMP(B,C) => CMP(MIN/MAX(A,B), C)
; CMP(A,C)&&CMP(B,C) => CMP(MIN/MAX(A,B), C)
define i1 @test1(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test1:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test1:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg1, 1000
%cmp2 = icmp slt i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -20,13 +21,13 @@ define i1 @test1(i32 %arg1, i32 %arg2) {
}
define i1 @test2(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test2:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test2:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, 1000
%cmp2 = icmp ult i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -34,13 +35,13 @@ define i1 @test2(i32 %arg1, i32 %arg2) {
}
define i1 @test3(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test3:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, 0x3e9, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test3:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, 0x3e9, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sle i32 %arg1, 1000
%cmp2 = icmp sle i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -48,13 +49,13 @@ define i1 @test3(i32 %arg1, i32 %arg2) {
}
define i1 @test4(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test4:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x3e9, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test4:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x3e9, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ule i32 %arg1, 1000
%cmp2 = icmp ule i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -62,13 +63,13 @@ define i1 @test4(i32 %arg1, i32 %arg2) {
}
define i1 @test5(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test5:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test5:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg1, 1000
%cmp2 = icmp sgt i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -76,13 +77,13 @@ define i1 @test5(i32 %arg1, i32 %arg2) {
}
define i1 @test6(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test6:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test6:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ugt i32 %arg1, 1000
%cmp2 = icmp ugt i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -90,13 +91,13 @@ define i1 @test6(i32 %arg1, i32 %arg2) {
}
define i1 @test7(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test7:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, 0x3e7, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test7:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, 0x3e7, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sge i32 %arg1, 1000
%cmp2 = icmp sge i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -104,13 +105,13 @@ define i1 @test7(i32 %arg1, i32 %arg2) {
}
define i1 @test8(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test8:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, 0x3e7, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test8:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, 0x3e7, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp uge i32 %arg1, 1000
%cmp2 = icmp uge i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -118,13 +119,13 @@ define i1 @test8(i32 %arg1, i32 %arg2) {
}
define i1 @test9(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test9:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test9:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg1, %arg3
%cmp2 = icmp slt i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -132,13 +133,13 @@ define i1 @test9(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test10(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test10:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test10:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %arg3
%cmp2 = icmp ult i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -146,13 +147,13 @@ define i1 @test10(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test11(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test11:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_le_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test11:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_le_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sle i32 %arg1, %arg3
%cmp2 = icmp sle i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -160,13 +161,13 @@ define i1 @test11(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test12(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test12:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_le_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test12:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_le_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ule i32 %arg1, %arg3
%cmp2 = icmp ule i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -174,13 +175,13 @@ define i1 @test12(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test13(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test13:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test13:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg1, %arg3
%cmp2 = icmp sgt i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -188,13 +189,13 @@ define i1 @test13(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test14(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test14:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test14:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ugt i32 %arg1, %arg3
%cmp2 = icmp ugt i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -202,13 +203,13 @@ define i1 @test14(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test15(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test15:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_ge_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test15:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_ge_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sge i32 %arg1, %arg3
%cmp2 = icmp sge i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -216,13 +217,13 @@ define i1 @test15(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test16(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test16:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_ge_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test16:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_ge_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp uge i32 %arg1, %arg3
%cmp2 = icmp uge i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -230,13 +231,13 @@ define i1 @test16(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test17(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test17:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test17:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg1, 1000
%cmp2 = icmp slt i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -244,13 +245,13 @@ define i1 @test17(i32 %arg1, i32 %arg2) {
}
define i1 @test18(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test18:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test18:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, 1000
%cmp2 = icmp ult i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -258,13 +259,13 @@ define i1 @test18(i32 %arg1, i32 %arg2) {
}
define i1 @test19(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test19:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, 0x3e9, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test19:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, 0x3e9, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sle i32 %arg1, 1000
%cmp2 = icmp sle i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -272,13 +273,13 @@ define i1 @test19(i32 %arg1, i32 %arg2) {
}
define i1 @test20(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test20:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x3e9, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test20:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x3e9, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ule i32 %arg1, 1000
%cmp2 = icmp ule i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -286,13 +287,13 @@ define i1 @test20(i32 %arg1, i32 %arg2) {
}
define i1 @test21(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test21:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test21:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg1, 1000
%cmp2 = icmp sgt i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -300,13 +301,13 @@ define i1 @test21(i32 %arg1, i32 %arg2) {
}
define i1 @test22(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test22:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test22:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ugt i32 %arg1, 1000
%cmp2 = icmp ugt i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -314,13 +315,13 @@ define i1 @test22(i32 %arg1, i32 %arg2) {
}
define i1 @test23(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test23:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, 0x3e7, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test23:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, 0x3e7, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sge i32 %arg1, 1000
%cmp2 = icmp sge i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -328,13 +329,13 @@ define i1 @test23(i32 %arg1, i32 %arg2) {
}
define i1 @test24(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test24:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, 0x3e7, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test24:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, 0x3e7, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp uge i32 %arg1, 1000
%cmp2 = icmp uge i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -342,13 +343,13 @@ define i1 @test24(i32 %arg1, i32 %arg2) {
}
define i1 @test25(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test25:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test25:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg1, %arg3
%cmp2 = icmp slt i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -356,13 +357,13 @@ define i1 @test25(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test26(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test26:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test26:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %arg3
%cmp2 = icmp ult i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -370,13 +371,13 @@ define i1 @test26(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test27(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test27:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_le_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test27:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_le_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sle i32 %arg1, %arg3
%cmp2 = icmp sle i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -384,13 +385,13 @@ define i1 @test27(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test28(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test28:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_le_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test28:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_le_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ule i32 %arg1, %arg3
%cmp2 = icmp ule i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -398,13 +399,13 @@ define i1 @test28(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test29(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test29:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test29:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg1, %arg3
%cmp2 = icmp sgt i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -412,13 +413,13 @@ define i1 @test29(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test30(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test30:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test30:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ugt i32 %arg1, %arg3
%cmp2 = icmp ugt i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -426,13 +427,13 @@ define i1 @test30(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test31(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test31:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_ge_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test31:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_ge_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sge i32 %arg1, %arg3
%cmp2 = icmp sge i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -440,13 +441,13 @@ define i1 @test31(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test32(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test32:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_ge_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test32:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_ge_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp uge i32 %arg1, %arg3
%cmp2 = icmp uge i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -454,13 +455,13 @@ define i1 @test32(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test33(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test33:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v1, 0x3e8, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, v1, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test33:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v1, 0x3e8, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, v1, v0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg1, %arg2
%cmp2 = icmp slt i32 %arg1, 1000
%or = or i1 %cmp1, %cmp2
@@ -468,18 +469,18 @@ define i1 @test33(i32 %arg1, i32 %arg2) {
}
define amdgpu_gfx void @test34(i32 inreg %arg1, i32 inreg %arg2) {
-; CHECK-LABEL: test34:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_min_i32 s0, s4, s5
-; CHECK-NEXT: v_mov_b32_e32 v0, 0
-; CHECK-NEXT: s_cmpk_lt_i32 s0, 0x3e9
-; CHECK-NEXT: v_mov_b32_e32 v1, 0
-; CHECK-NEXT: s_cselect_b32 s0, -1, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
-; CHECK-NEXT: global_store_b8 v[0:1], v2, off dlc
-; CHECK-NEXT: s_waitcnt_vscnt null, 0x0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test34:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_min_i32 s0, s4, s5
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: s_cmpk_lt_i32 s0, 0x3e9
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cselect_b32 s0, -1, 0
+; GCN-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
+; GCN-NEXT: global_store_b8 v[0:1], v2, off dlc
+; GCN-NEXT: s_waitcnt_vscnt null, 0x0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sle i32 %arg1, 1000
%cmp2 = icmp sle i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -488,18 +489,18 @@ define amdgpu_gfx void @test34(i32 inreg %arg1, i32 inreg %arg2) {
}
define amdgpu_gfx void @test35(i32 inreg %arg1, i32 inreg %arg2) {
-; CHECK-LABEL: test35:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_max_i32 s0, s4, s5
-; CHECK-NEXT: v_mov_b32_e32 v0, 0
-; CHECK-NEXT: s_cmpk_gt_i32 s0, 0x3e8
-; CHECK-NEXT: v_mov_b32_e32 v1, 0
-; CHECK-NEXT: s_cselect_b32 s0, -1, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
-; CHECK-NEXT: global_store_b8 v[0:1], v2, off dlc
-; CHECK-NEXT: s_waitcnt_vscnt null, 0x0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test35:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_max_i32 s0, s4, s5
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: s_cmpk_gt_i32 s0, 0x3e8
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cselect_b32 s0, -1, 0
+; GCN-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
+; GCN-NEXT: global_store_b8 v[0:1], v2, off dlc
+; GCN-NEXT: s_waitcnt_vscnt null, 0x0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg1, 1000
%cmp2 = icmp sgt i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
@@ -508,18 +509,18 @@ define amdgpu_gfx void @test35(i32 inreg %arg1, i32 inreg %arg2) {
}
define amdgpu_gfx void @test36(i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3) {
-; CHECK-LABEL: test36:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_min_u32 s0, s4, s5
-; CHECK-NEXT: v_mov_b32_e32 v0, 0
-; CHECK-NEXT: s_cmp_lt_u32 s0, s6
-; CHECK-NEXT: v_mov_b32_e32 v1, 0
-; CHECK-NEXT: s_cselect_b32 s0, -1, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
-; CHECK-NEXT: global_store_b8 v[0:1], v2, off dlc
-; CHECK-NEXT: s_waitcnt_vscnt null, 0x0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test36:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_min_u32 s0, s4, s5
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: s_cmp_lt_u32 s0, s6
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cselect_b32 s0, -1, 0
+; GCN-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
+; GCN-NEXT: global_store_b8 v[0:1], v2, off dlc
+; GCN-NEXT: s_waitcnt_vscnt null, 0x0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %arg3
%cmp2 = icmp ult i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -528,18 +529,18 @@ define amdgpu_gfx void @test36(i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3
}
define amdgpu_gfx void @test37(i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3) {
-; CHECK-LABEL: test37:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_max_i32 s0, s4, s5
-; CHECK-NEXT: v_mov_b32_e32 v0, 0
-; CHECK-NEXT: s_cmp_ge_i32 s0, s6
-; CHECK-NEXT: v_mov_b32_e32 v1, 0
-; CHECK-NEXT: s_cselect_b32 s0, -1, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
-; CHECK-NEXT: global_store_b8 v[0:1], v2, off dlc
-; CHECK-NEXT: s_waitcnt_vscnt null, 0x0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test37:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_max_i32 s0, s4, s5
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: s_cmp_ge_i32 s0, s6
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cselect_b32 s0, -1, 0
+; GCN-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
+; GCN-NEXT: global_store_b8 v[0:1], v2, off dlc
+; GCN-NEXT: s_waitcnt_vscnt null, 0x0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sge i32 %arg1, %arg3
%cmp2 = icmp sge i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -548,18 +549,18 @@ define amdgpu_gfx void @test37(i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3
}
define amdgpu_gfx void @test38(i32 inreg %arg1, i32 inreg %arg2) {
-; CHECK-LABEL: test38:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_max_u32 s0, s4, s5
-; CHECK-NEXT: v_mov_b32_e32 v0, 0
-; CHECK-NEXT: s_cmpk_lt_u32 s0, 0x3e9
-; CHECK-NEXT: v_mov_b32_e32 v1, 0
-; CHECK-NEXT: s_cselect_b32 s0, -1, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
-; CHECK-NEXT: global_store_b8 v[0:1], v2, off dlc
-; CHECK-NEXT: s_waitcnt_vscnt null, 0x0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test38:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_max_u32 s0, s4, s5
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: s_cmpk_lt_u32 s0, 0x3e9
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cselect_b32 s0, -1, 0
+; GCN-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
+; GCN-NEXT: global_store_b8 v[0:1], v2, off dlc
+; GCN-NEXT: s_waitcnt_vscnt null, 0x0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ule i32 %arg1, 1000
%cmp2 = icmp ule i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -568,18 +569,18 @@ define amdgpu_gfx void @test38(i32 inreg %arg1, i32 inreg %arg2) {
}
define amdgpu_gfx void @test39(i32 inreg %arg1, i32 inreg %arg2) {
-; CHECK-LABEL: test39:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_min_i32 s0, s4, s5
-; CHECK-NEXT: v_mov_b32_e32 v0, 0
-; CHECK-NEXT: s_cmpk_gt_i32 s0, 0x3e7
-; CHECK-NEXT: v_mov_b32_e32 v1, 0
-; CHECK-NEXT: s_cselect_b32 s0, -1, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
-; CHECK-NEXT: global_store_b8 v[0:1], v2, off dlc
-; CHECK-NEXT: s_waitcnt_vscnt null, 0x0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test39:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_min_i32 s0, s4, s5
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: s_cmpk_gt_i32 s0, 0x3e7
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cselect_b32 s0, -1, 0
+; GCN-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
+; GCN-NEXT: global_store_b8 v[0:1], v2, off dlc
+; GCN-NEXT: s_waitcnt_vscnt null, 0x0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sge i32 %arg1, 1000
%cmp2 = icmp sge i32 %arg2, 1000
%and = and i1 %cmp1, %cmp2
@@ -588,18 +589,18 @@ define amdgpu_gfx void @test39(i32 inreg %arg1, i32 inreg %arg2) {
}
define amdgpu_gfx void @test40(i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3) {
-; CHECK-LABEL: test40:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_max_i32 s0, s4, s5
-; CHECK-NEXT: v_mov_b32_e32 v0, 0
-; CHECK-NEXT: s_cmp_le_i32 s0, s6
-; CHECK-NEXT: v_mov_b32_e32 v1, 0
-; CHECK-NEXT: s_cselect_b32 s0, -1, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
-; CHECK-NEXT: global_store_b8 v[0:1], v2, off dlc
-; CHECK-NEXT: s_waitcnt_vscnt null, 0x0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test40:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_max_i32 s0, s4, s5
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: s_cmp_le_i32 s0, s6
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cselect_b32 s0, -1, 0
+; GCN-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
+; GCN-NEXT: global_store_b8 v[0:1], v2, off dlc
+; GCN-NEXT: s_waitcnt_vscnt null, 0x0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sle i32 %arg1, %arg3
%cmp2 = icmp sle i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -608,18 +609,18 @@ define amdgpu_gfx void @test40(i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3
}
define amdgpu_gfx void @test41(i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3) {
-; CHECK-LABEL: test41:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_min_u32 s0, s4, s5
-; CHECK-NEXT: v_mov_b32_e32 v0, 0
-; CHECK-NEXT: s_cmp_ge_u32 s0, s6
-; CHECK-NEXT: v_mov_b32_e32 v1, 0
-; CHECK-NEXT: s_cselect_b32 s0, -1, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
-; CHECK-NEXT: global_store_b8 v[0:1], v2, off dlc
-; CHECK-NEXT: s_waitcnt_vscnt null, 0x0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test41:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_min_u32 s0, s4, s5
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: s_cmp_ge_u32 s0, s6
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: s_cselect_b32 s0, -1, 0
+; GCN-NEXT: v_cndmask_b32_e64 v2, 0, 1, s0
+; GCN-NEXT: global_store_b8 v[0:1], v2, off dlc
+; GCN-NEXT: s_waitcnt_vscnt null, 0x0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp uge i32 %arg1, %arg3
%cmp2 = icmp uge i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -628,13 +629,13 @@ define amdgpu_gfx void @test41(i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3
}
define i1 @test42(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test42:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test42:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg3, %arg1
%cmp2 = icmp ult i32 %arg3, %arg2
%or = and i1 %cmp1, %cmp2
@@ -642,13 +643,13 @@ define i1 @test42(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test43(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test43:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test43:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg3, %arg1
%cmp2 = icmp ult i32 %arg3, %arg2
%or = or i1 %cmp1, %cmp2
@@ -656,13 +657,13 @@ define i1 @test43(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test44(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test44:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test44:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ugt i32 %arg3, %arg1
%cmp2 = icmp ugt i32 %arg3, %arg2
%or = and i1 %cmp1, %cmp2
@@ -670,13 +671,13 @@ define i1 @test44(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test45(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test45:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test45:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ugt i32 %arg3, %arg1
%cmp2 = icmp ugt i32 %arg3, %arg2
%or = or i1 %cmp1, %cmp2
@@ -684,13 +685,13 @@ define i1 @test45(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test46(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test46:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test46:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg3, %arg1
%cmp2 = icmp sgt i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -698,13 +699,13 @@ define i1 @test46(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test47(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test47:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test47:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg1, %arg3
%cmp2 = icmp slt i32 %arg3, %arg2
%or = or i1 %cmp1, %cmp2
@@ -712,13 +713,13 @@ define i1 @test47(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test48(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test48:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test48:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg1, %arg3
%cmp2 = icmp sgt i32 %arg3, %arg2
%or = or i1 %cmp1, %cmp2
@@ -726,13 +727,13 @@ define i1 @test48(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test49(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test49:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test49:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg3, %arg1
%cmp2 = icmp slt i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
@@ -740,13 +741,13 @@ define i1 @test49(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test50(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test50:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test50:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg3, %arg1
%cmp2 = icmp sgt i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -754,13 +755,13 @@ define i1 @test50(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test51(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test51:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test51:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_gt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg1, %arg3
%cmp2 = icmp slt i32 %arg3, %arg2
%and = and i1 %cmp1, %cmp2
@@ -768,13 +769,13 @@ define i1 @test51(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test52(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test52:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test52:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg1, %arg3
%cmp2 = icmp sgt i32 %arg3, %arg2
%and = and i1 %cmp1, %cmp2
@@ -782,13 +783,13 @@ define i1 @test52(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test53(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test53:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_i32_e32 v0, v0, v1
-; CHECK-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test53:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_i32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_i32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp sgt i32 %arg3, %arg1
%cmp2 = icmp slt i32 %arg2, %arg3
%and = and i1 %cmp1, %cmp2
@@ -796,14 +797,13 @@ define i1 @test53(i32 %arg1, i32 %arg2, i32 %arg3) {
}
define i1 @test54(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test54:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test54:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_f32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %arg3
%cmp2 = fcmp olt float %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
@@ -811,14 +811,13 @@ define i1 @test54(float %arg1, float %arg2, float %arg3) #0 {
}
define i1 @test55(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test55:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_le_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test55:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ole double %arg1, %arg3
%cmp2 = fcmp ole double %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
@@ -826,14 +825,13 @@ define i1 @test55(double %arg1, double %arg2, double %arg3) #0 {
}
define i1 @test56(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test56:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_gt_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test56:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ogt double %arg1, %arg3
%cmp2 = fcmp ogt double %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
@@ -841,14 +839,13 @@ define i1 @test56(double %arg1, double %arg2, double %arg3) #0 {
}
define i1 @test57(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test57:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ge_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test57:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_f32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp oge float %arg1, %arg3
%cmp2 = fcmp oge float %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
@@ -856,14 +853,21 @@ define i1 @test57(float %arg1, float %arg2, float %arg3) #0 {
}
define i1 @test58(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test58:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nle_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_nle_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test58:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: v_cmp_nle_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test58:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ugt double %arg1, %arg3
%cmp2 = fcmp ugt double %arg2, %arg3
%and1 = and i1 %cmp1, %cmp2
@@ -871,14 +875,21 @@ define i1 @test58(double %arg1, double %arg2, double %arg3) #0 {
}
define i1 @test59(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test59:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nlt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test59:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test59:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp uge float %arg1, %arg3
%cmp2 = fcmp uge float %arg2, %arg3
%and1 = and i1 %cmp1, %cmp2
@@ -886,14 +897,21 @@ define i1 @test59(float %arg1, float %arg2, float %arg3) #0 {
}
define i1 @test60(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test60:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ngt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test60:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test60:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ule float %arg1, %arg3
%cmp2 = fcmp ule float %arg2, %arg3
%and1 = and i1 %cmp1, %cmp2
@@ -901,14 +919,21 @@ define i1 @test60(float %arg1, float %arg2, float %arg3) #0 {
}
define i1 @test61(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test61:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nge_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_nge_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test61:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: v_cmp_nge_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test61:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_lt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ult double %arg1, %arg3
%cmp2 = fcmp ult double %arg2, %arg3
%and1 = and i1 %cmp1, %cmp2
@@ -916,15 +941,14 @@ define i1 @test61(double %arg1, double %arg2, double %arg3) #0 {
}
define i1 @test62(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test62:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test62:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
+; GCN-NEXT: v_min_f32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, 1.0
%add2 = fadd nnan float %arg2, 2.0
%cmp1 = fcmp nnan olt float %add1, %arg3
@@ -934,16 +958,15 @@ define i1 @test62(float %arg1, float %arg2, float %arg3) {
}
define i1 @test63(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test63:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
-; CHECK-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
-; CHECK-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_le_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test63:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
+; GCN-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
+; GCN-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan double %arg1, 1.0
%add2 = fadd nnan double %arg2, 2.0
%cmp1 = fcmp nnan ole double %add1, %arg3
@@ -953,16 +976,15 @@ define i1 @test63(double %arg1, double %arg2, double %arg3) #0 {
}
define i1 @test64(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test64:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
-; CHECK-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
-; CHECK-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_gt_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test64:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
+; GCN-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
+; GCN-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan double %arg1, 1.0
%add2 = fadd nnan double %arg2, 2.0
%cmp1 = fcmp nnan ogt double %add1, %arg3
@@ -972,15 +994,14 @@ define i1 @test64(double %arg1, double %arg2, double %arg3) #0 {
}
define i1 @test65(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test65:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
-; CHECK-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ge_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test65:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
+; GCN-NEXT: v_max_f32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, 1.0
%add2 = fadd nnan float %arg2, 2.0
%cmp1 = fcmp nnan oge float %add1, %arg3
@@ -990,16 +1011,15 @@ define i1 @test65(float %arg1, float %arg2, float %arg3) {
}
define i1 @test66(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test66:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
-; CHECK-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
-; CHECK-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_gt_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test66:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
+; GCN-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
+; GCN-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan double %arg1, 1.0
%add2 = fadd nnan double %arg2, 2.0
%cmp1 = fcmp nnan ugt double %add1, %arg3
@@ -1009,15 +1029,14 @@ define i1 @test66(double %arg1, double %arg2, double %arg3) {
}
define i1 @test67(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test67:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
-; CHECK-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ge_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test67:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
+; GCN-NEXT: v_min_f32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, 1.0
%add2 = fadd nnan float %arg2, 2.0
%cmp1 = fcmp nnan uge float %add1, %arg3
@@ -1027,15 +1046,14 @@ define i1 @test67(float %arg1, float %arg2, float %arg3) #0 {
}
define i1 @test68(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test68:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
-; CHECK-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_le_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test68:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
+; GCN-NEXT: v_max_f32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, 1.0
%add2 = fadd nnan float %arg2, 2.0
%cmp1 = fcmp nnan ule float %add1, %arg3
@@ -1045,16 +1063,15 @@ define i1 @test68(float %arg1, float %arg2, float %arg3) #0 {
}
define i1 @test69(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test69:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
-; CHECK-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
-; CHECK-NEXT: v_cmp_lt_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_lt_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test69:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
+; GCN-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
+; GCN-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_lt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan double %arg1, 1.0
%add2 = fadd nnan double %arg2, 2.0
%cmp1 = fcmp nnan ult double %add1, %arg3
@@ -1064,15 +1081,22 @@ define i1 @test69(double %arg1, double %arg2, double %arg3) {
}
define i1 @test70(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test70:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test70:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test70:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp olt float %var1, %arg3
@@ -1082,16 +1106,15 @@ define i1 @test70(float %arg1, float %arg2, float %arg3) {
}
define i1 @test71(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test71:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_le_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test71:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GCN-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GCN-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ole double %var1, %arg3
@@ -1101,16 +1124,15 @@ define i1 @test71(double %arg1, double %arg2, double %arg3) {
}
define i1 @test72(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test72:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_gt_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test72:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GCN-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GCN-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ogt double %var1, %arg3
@@ -1120,15 +1142,22 @@ define i1 @test72(double %arg1, double %arg2, double %arg3) {
}
define i1 @test73(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test73:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ge_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test73:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test73:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp oge float %var1, %arg3
@@ -1138,16 +1167,25 @@ define i1 @test73(float %arg1, float %arg2, float %arg3) {
}
define i1 @test74(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test74:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_nle_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_nle_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test74:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: v_cmp_nle_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test74:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11NONANS-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11NONANS-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ugt double %var1, %arg3
@@ -1157,15 +1195,22 @@ define i1 @test74(double %arg1, double %arg2, double %arg3) {
}
define i1 @test75(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test75:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nlt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test75:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test75:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp uge float %var1, %arg3
@@ -1175,15 +1220,22 @@ define i1 @test75(float %arg1, float %arg2, float %arg3) {
}
define i1 @test76(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test76:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ngt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test76:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test76:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp ule float %var1, %arg3
@@ -1193,16 +1245,25 @@ define i1 @test76(float %arg1, float %arg2, float %arg3) {
}
define i1 @test77(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test77:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_nge_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_nge_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test77:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: v_cmp_nge_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test77:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11NONANS-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_lt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ult double %var1, %arg3
@@ -1212,14 +1273,13 @@ define i1 @test77(double %arg1, double %arg2, double %arg3) {
}
define i1 @test78(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test78:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_gt_f32_e64 s0, v2, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test78:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_f32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %arg3
%cmp2 = fcmp ogt float %arg3, %arg2
%or1 = or i1 %cmp1, %cmp2
@@ -1227,14 +1287,21 @@ define i1 @test78(float %arg1, float %arg2, float %arg3) #0 {
}
define i1 @test79(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test79:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nle_f32_e64 s0, v2, v1
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test79:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test79:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ult float %arg1, %arg3
%cmp2 = fcmp ugt float %arg3, %arg2
%and1 = and i1 %cmp1, %cmp2
@@ -1242,15 +1309,14 @@ define i1 @test79(float %arg1, float %arg2, float %arg3) #0 {
}
define i1 @test80(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test80:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
-; CHECK-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_le_f32_e64 s0, v2, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test80:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v0, 1.0, v0 :: v_dual_add_f32 v1, 2.0, v1
+; GCN-NEXT: v_max_f32_e32 v0, v0, v1
+; GCN-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, 1.0
%add2 = fadd nnan float %arg2, 2.0
%cmp1 = fcmp nnan oge float %add1, %arg3
@@ -1260,16 +1326,15 @@ define i1 @test80(float %arg1, float %arg2, float %arg3) {
}
define i1 @test81(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test81:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
-; CHECK-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
-; CHECK-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_lt_f64_e64 s0, v[4:5], v[2:3]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test81:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_add_f64 v[0:1], v[0:1], 1.0
+; GCN-NEXT: v_add_f64 v[2:3], v[2:3], 2.0
+; GCN-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan double %arg1, 1.0
%add2 = fadd nnan double %arg2, 2.0
%cmp1 = fcmp nnan ugt double %add1, %arg3
@@ -1279,16 +1344,15 @@ define i1 @test81(double %arg1, double %arg2, double %arg3) {
}
define i1 @test82(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test82:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_ge_f64_e64 s0, v[4:5], v[2:3]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test82:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GCN-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GCN-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ole double %var1, %arg3
@@ -1298,15 +1362,22 @@ define i1 @test82(double %arg1, double %arg2, double %arg3) {
}
define i1 @test83(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test83:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nlt_f32_e64 s0, v2, v1
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test83:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test83:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp ule float %var1, %arg3
@@ -1316,16 +1387,23 @@ define i1 @test83(float %arg1, float %arg2, float %arg3) {
}
define i1 @test84(half %arg1, half %arg2, half %arg3) {
-; CHECK-LABEL: test84:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f16_e32 v0, v0, v0
-; CHECK-NEXT: v_max_f16_e32 v1, v1, v1
-; CHECK-NEXT: v_cmp_lt_f16_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_lt_f16_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test84:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f16_e32 v0, v0, v0
+; GFX11-NEXT: v_max_f16_e32 v1, v1, v1
+; GFX11-NEXT: v_min_f16_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_lt_f16_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test84:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f16_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f16_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call half @llvm.canonicalize.f16(half %arg1)
%var2 = call half @llvm.canonicalize.f16(half %arg2)
%cmp1 = fcmp olt half %var1, %arg3
@@ -1335,23 +1413,31 @@ define i1 @test84(half %arg1, half %arg2, half %arg3) {
}
define <2 x i1> @test85(<2 x half> %arg1, <2 x half> %arg2, <2 x half> %arg3) {
-; CHECK-LABEL: test85:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_pk_max_f16 v0, v0, v0
-; CHECK-NEXT: v_pk_max_f16 v1, v1, v1
-; CHECK-NEXT: v_lshrrev_b32_e32 v3, 16, v2
-; CHECK-NEXT: v_lshrrev_b32_e32 v4, 16, v0
-; CHECK-NEXT: v_lshrrev_b32_e32 v5, 16, v1
-; CHECK-NEXT: v_cmp_le_f16_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_le_f16_e64 s0, v1, v2
-; CHECK-NEXT: v_cmp_le_f16_e64 s1, v4, v3
-; CHECK-NEXT: v_cmp_le_f16_e64 s2, v5, v3
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_or_b32 s0, s1, s2
-; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test85:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX11-NEXT: v_pk_min_f16 v0, v0, v1
+; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v2
+; GFX11-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX11-NEXT: v_cmp_le_f16_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: v_cmp_le_f16_e32 vcc_lo, v3, v1
+; GFX11-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test85:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_pk_min_f16 v0, v0, v1
+; GFX11NONANS-NEXT: v_lshrrev_b32_e32 v1, 16, v2
+; GFX11NONANS-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX11NONANS-NEXT: v_cmp_le_f16_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: v_cmp_le_f16_e32 vcc_lo, v3, v1
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %arg1)
%var2 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %arg2)
%cmp1 = fcmp ole <2 x half> %var1, %arg3
@@ -1361,23 +1447,31 @@ define <2 x i1> @test85(<2 x half> %arg1, <2 x half> %arg2, <2 x half> %arg3) {
}
define <2 x i1> @test86(<2 x half> %arg1, <2 x half> %arg2, <2 x half> %arg3) {
-; CHECK-LABEL: test86:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_pk_max_f16 v0, v0, v0
-; CHECK-NEXT: v_pk_max_f16 v1, v1, v1
-; CHECK-NEXT: v_lshrrev_b32_e32 v3, 16, v2
-; CHECK-NEXT: v_lshrrev_b32_e32 v4, 16, v0
-; CHECK-NEXT: v_lshrrev_b32_e32 v5, 16, v1
-; CHECK-NEXT: v_cmp_gt_f16_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_gt_f16_e64 s0, v1, v2
-; CHECK-NEXT: v_cmp_gt_f16_e64 s1, v4, v3
-; CHECK-NEXT: v_cmp_gt_f16_e64 s2, v5, v3
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_or_b32 s0, s1, s2
-; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test86:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX11-NEXT: v_pk_max_f16 v0, v0, v1
+; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v2
+; GFX11-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX11-NEXT: v_cmp_gt_f16_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: v_cmp_gt_f16_e32 vcc_lo, v3, v1
+; GFX11-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test86:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_pk_max_f16 v0, v0, v1
+; GFX11NONANS-NEXT: v_lshrrev_b32_e32 v1, 16, v2
+; GFX11NONANS-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX11NONANS-NEXT: v_cmp_gt_f16_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: v_cmp_gt_f16_e32 vcc_lo, v3, v1
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %arg1)
%var2 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %arg2)
%cmp1 = fcmp ogt <2 x half> %var1, %arg3
@@ -1387,16 +1481,23 @@ define <2 x i1> @test86(<2 x half> %arg1, <2 x half> %arg2, <2 x half> %arg3) {
}
define i1 @test87(half %arg1, half %arg2, half %arg3) {
-; CHECK-LABEL: test87:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f16_e32 v0, v0, v0
-; CHECK-NEXT: v_max_f16_e32 v1, v1, v1
-; CHECK-NEXT: v_cmp_ge_f16_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ge_f16_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test87:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f16_e32 v0, v0, v0
+; GFX11-NEXT: v_max_f16_e32 v1, v1, v1
+; GFX11-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_ge_f16_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test87:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f16_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call half @llvm.canonicalize.f16(half %arg1)
%var2 = call half @llvm.canonicalize.f16(half %arg2)
%cmp1 = fcmp oge half %var1, %arg3
@@ -1406,23 +1507,31 @@ define i1 @test87(half %arg1, half %arg2, half %arg3) {
}
define <2 x i1> @test88(<2 x half> %arg1, <2 x half> %arg2, <2 x half> %arg3) {
-; CHECK-LABEL: test88:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_pk_max_f16 v0, v0, v0
-; CHECK-NEXT: v_pk_max_f16 v1, v1, v1
-; CHECK-NEXT: v_lshrrev_b32_e32 v3, 16, v2
-; CHECK-NEXT: v_lshrrev_b32_e32 v4, 16, v0
-; CHECK-NEXT: v_lshrrev_b32_e32 v5, 16, v1
-; CHECK-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nle_f16_e64 s0, v1, v2
-; CHECK-NEXT: v_cmp_nle_f16_e64 s1, v4, v3
-; CHECK-NEXT: v_cmp_nle_f16_e64 s2, v5, v3
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_and_b32 s0, s1, s2
-; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test88:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX11-NEXT: v_pk_min_f16 v0, v0, v1
+; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v2
+; GFX11-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX11-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: v_cmp_nle_f16_e32 vcc_lo, v3, v1
+; GFX11-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test88:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_pk_min_f16 v0, v0, v1
+; GFX11NONANS-NEXT: v_lshrrev_b32_e32 v1, 16, v2
+; GFX11NONANS-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX11NONANS-NEXT: v_cmp_gt_f16_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: v_cmp_gt_f16_e32 vcc_lo, v3, v1
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %arg1)
%var2 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %arg2)
%cmp1 = fcmp ugt <2 x half> %var1, %arg3
@@ -1432,16 +1541,23 @@ define <2 x i1> @test88(<2 x half> %arg1, <2 x half> %arg2, <2 x half> %arg3) {
}
define i1 @test89(half %arg1, half %arg2, half %arg3) {
-; CHECK-LABEL: test89:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f16_e32 v0, v0, v0
-; CHECK-NEXT: v_max_f16_e32 v1, v1, v1
-; CHECK-NEXT: v_cmp_nlt_f16_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nlt_f16_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test89:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f16_e32 v0, v0, v0
+; GFX11-NEXT: v_max_f16_e32 v1, v1, v1
+; GFX11-NEXT: v_min_f16_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_nlt_f16_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test89:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f16_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f16_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call half @llvm.canonicalize.f16(half %arg1)
%var2 = call half @llvm.canonicalize.f16(half %arg2)
%cmp1 = fcmp uge half %var1, %arg3
@@ -1451,16 +1567,23 @@ define i1 @test89(half %arg1, half %arg2, half %arg3) {
}
define i1 @test90(half %arg1, half %arg2, half %arg3) {
-; CHECK-LABEL: test90:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f16_e32 v0, v0, v0
-; CHECK-NEXT: v_max_f16_e32 v1, v1, v1
-; CHECK-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ngt_f16_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test90:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f16_e32 v0, v0, v0
+; GFX11-NEXT: v_max_f16_e32 v1, v1, v1
+; GFX11-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test90:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f16_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_le_f16_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call half @llvm.canonicalize.f16(half %arg1)
%var2 = call half @llvm.canonicalize.f16(half %arg2)
%cmp1 = fcmp ule half %var1, %arg3
@@ -1470,23 +1593,31 @@ define i1 @test90(half %arg1, half %arg2, half %arg3) {
}
define <2 x i1> @test91(<2 x half> %arg1, <2 x half> %arg2, <2 x half> %arg3) {
-; CHECK-LABEL: test91:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_pk_max_f16 v0, v0, v0
-; CHECK-NEXT: v_pk_max_f16 v1, v1, v1
-; CHECK-NEXT: v_lshrrev_b32_e32 v3, 16, v2
-; CHECK-NEXT: v_lshrrev_b32_e32 v4, 16, v0
-; CHECK-NEXT: v_lshrrev_b32_e32 v5, 16, v1
-; CHECK-NEXT: v_cmp_nge_f16_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nge_f16_e64 s0, v1, v2
-; CHECK-NEXT: v_cmp_nge_f16_e64 s1, v4, v3
-; CHECK-NEXT: v_cmp_nge_f16_e64 s2, v5, v3
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_and_b32 s0, s1, s2
-; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test91:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX11-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX11-NEXT: v_pk_max_f16 v0, v0, v1
+; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v2
+; GFX11-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX11-NEXT: v_cmp_nge_f16_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: v_cmp_nge_f16_e32 vcc_lo, v3, v1
+; GFX11-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test91:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_pk_max_f16 v0, v0, v1
+; GFX11NONANS-NEXT: v_lshrrev_b32_e32 v1, 16, v2
+; GFX11NONANS-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX11NONANS-NEXT: v_cmp_lt_f16_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: v_cmp_lt_f16_e32 vcc_lo, v3, v1
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %arg1)
%var2 = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %arg2)
%cmp1 = fcmp ult <2 x half> %var1, %arg3
@@ -1496,13 +1627,13 @@ define <2 x i1> @test91(<2 x half> %arg1, <2 x half> %arg2, <2 x half> %arg3) {
}
define i1 @test92(i32 %arg1, i32 %arg2, i32 %arg3, i32 %C) {
-; CHECK-LABEL: test92:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min3_u32 v0, v0, v1, v2
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v3
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test92:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min3_u32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v3
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1512,16 +1643,16 @@ define i1 @test92(i32 %arg1, i32 %arg2, i32 %arg3, i32 %C) {
}
define i1 @test93(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
-; CHECK-LABEL: test93:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_max_u32_e32 v1, v2, v3
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cmp_gt_u32_e64 s0, v1, v4
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test93:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_max_u32_e32 v1, v2, v3
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
+; GCN-NEXT: v_cmp_gt_u32_e64 s0, v1, v4
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ugt i32 %arg3, %C
@@ -1533,16 +1664,16 @@ define i1 @test93(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
}
define i1 @test94(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %arg7, i32 %arg8, i32 %C) {
-; CHECK-LABEL: test94:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v2, v2, v3
-; CHECK-NEXT: v_min3_u32 v0, v0, v1, v2
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v4
-; CHECK-NEXT: v_min3_u32 v0, v5, v6, v0
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v8
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test94:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v2, v2, v3
+; GCN-NEXT: v_min3_u32 v0, v0, v1, v2
+; GCN-NEXT: v_min_u32_e32 v0, v0, v4
+; GCN-NEXT: v_min3_u32 v0, v5, v6, v0
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v8
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%or1 = or i1 %cmp1, %cmp2
@@ -1562,13 +1693,13 @@ define i1 @test94(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %ar
}
define i1 @test95(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
-; CHECK-LABEL: test95:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_maxmin_u32 v0, v0, v1, v2
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test95:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_maxmin_u32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1578,13 +1709,13 @@ define i1 @test95(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
}
define i1 @test96(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
-; CHECK-LABEL: test96:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_minmax_u32 v0, v0, v1, v2
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test96:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_minmax_u32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1594,14 +1725,14 @@ define i1 @test96(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
}
define i1 @test97(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
-; CHECK-LABEL: test97:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_max3_u32 v0, v0, v2, v3
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test97:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_max3_u32 v0, v0, v2, v3
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1613,14 +1744,14 @@ define i1 @test97(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
}
define i1 @test98(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
-; CHECK-LABEL: test98:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v2, v2, v3
-; CHECK-NEXT: v_minmax_u32 v0, v0, v1, v2
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test98:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v2, v2, v3
+; GCN-NEXT: v_minmax_u32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1632,14 +1763,14 @@ define i1 @test98(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
}
define i1 @test99(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
-; CHECK-LABEL: test99:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v2, v2, v3
-; CHECK-NEXT: v_min3_u32 v0, v0, v1, v2
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test99:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v2, v2, v3
+; GCN-NEXT: v_min3_u32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1651,14 +1782,14 @@ define i1 @test99(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
}
define i1 @test100(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
-; CHECK-LABEL: test100:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v2, v2, v3
-; CHECK-NEXT: v_maxmin_u32 v0, v0, v1, v2
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test100:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v2, v2, v3
+; GCN-NEXT: v_maxmin_u32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v4
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1670,15 +1801,15 @@ define i1 @test100(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %C) {
}
define i1 @test101(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %C) {
-; CHECK-LABEL: test101:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_minmax_u32 v1, v3, v4, v5
-; CHECK-NEXT: v_min3_u32 v0, v0, v2, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v6
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test101:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_minmax_u32 v1, v3, v4, v5
+; GCN-NEXT: v_min3_u32 v0, v0, v2, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v6
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1694,15 +1825,15 @@ define i1 @test101(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %a
}
define i1 @test102(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %C) {
-; CHECK-LABEL: test102:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_min_u32_e32 v1, v2, v3
-; CHECK-NEXT: v_min3_u32 v0, v0, v5, v1
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v6
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test102:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_min_u32_e32 v1, v2, v3
+; GCN-NEXT: v_min3_u32 v0, v0, v5, v1
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v6
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ult i32 %arg3, %C
@@ -1719,17 +1850,17 @@ define i1 @test102(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %a
}
define i1 @test103(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %C) {
-; CHECK-LABEL: test103:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v4, v4, v5
-; CHECK-NEXT: v_max_u32_e32 v2, v2, v3
-; CHECK-NEXT: v_maxmin_u32 v0, v0, v1, v4
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, v2, v6
-; CHECK-NEXT: v_cmp_lt_u32_e64 s0, v0, v6
-; CHECK-NEXT: s_or_b32 s0, s0, vcc_lo
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test103:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v4, v4, v5
+; GCN-NEXT: v_max_u32_e32 v2, v2, v3
+; GCN-NEXT: v_maxmin_u32 v0, v0, v1, v4
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, v2, v6
+; GCN-NEXT: v_cmp_lt_u32_e64 s0, v0, v6
+; GCN-NEXT: s_or_b32 s0, s0, vcc_lo
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ugt i32 %arg3, %C
@@ -1746,23 +1877,23 @@ define i1 @test103(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %a
define i1 @test104(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %arg7, i32 %arg8, i32 %arg9, i32 %arg10, i32 %C) {
-; CHECK-LABEL: test104:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v8, v8, v9
-; CHECK-NEXT: v_max_u32_e32 v2, v2, v3
-; CHECK-NEXT: v_min_u32_e32 v3, v4, v5
-; CHECK-NEXT: v_max_u32_e32 v4, v6, v7
-; CHECK-NEXT: v_min3_u32 v0, v0, v1, v8
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, v2, v10
-; CHECK-NEXT: v_cmp_lt_u32_e64 s0, v3, v10
-; CHECK-NEXT: v_cmp_gt_u32_e64 s1, v4, v10
-; CHECK-NEXT: v_cmp_lt_u32_e64 s2, v0, v10
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: s_or_b32 s1, s2, vcc_lo
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test104:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v8, v8, v9
+; GCN-NEXT: v_max_u32_e32 v2, v2, v3
+; GCN-NEXT: v_min_u32_e32 v3, v4, v5
+; GCN-NEXT: v_max_u32_e32 v4, v6, v7
+; GCN-NEXT: v_min3_u32 v0, v0, v1, v8
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, v2, v10
+; GCN-NEXT: v_cmp_lt_u32_e64 s0, v3, v10
+; GCN-NEXT: v_cmp_gt_u32_e64 s1, v4, v10
+; GCN-NEXT: v_cmp_lt_u32_e64 s2, v0, v10
+; GCN-NEXT: s_or_b32 s0, s0, s1
+; GCN-NEXT: s_or_b32 s1, s2, vcc_lo
+; GCN-NEXT: s_or_b32 s0, s0, s1
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ugt i32 %arg3, %C
@@ -1786,22 +1917,22 @@ define i1 @test104(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %a
}
define i1 @test105(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %arg7, i32 %arg8, i32 %arg9, i32 %arg10, i32 %C) {
-; CHECK-LABEL: test105:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_max_u32_e32 v1, v2, v3
-; CHECK-NEXT: v_max_u32_e32 v2, v4, v5
-; CHECK-NEXT: v_max_u32_e32 v3, v6, v7
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v10
-; CHECK-NEXT: v_cmp_gt_u32_e64 s0, v1, v10
-; CHECK-NEXT: v_cmp_lt_u32_e64 s1, v2, v10
-; CHECK-NEXT: v_cmp_gt_u32_e64 s2, v3, v10
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s1, s2, s1
-; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test105:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_u32_e32 v0, v0, v1
+; GCN-NEXT: v_max_u32_e32 v1, v2, v3
+; GCN-NEXT: v_max_u32_e32 v2, v4, v5
+; GCN-NEXT: v_max_u32_e32 v3, v6, v7
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v10
+; GCN-NEXT: v_cmp_gt_u32_e64 s0, v1, v10
+; GCN-NEXT: v_cmp_lt_u32_e64 s1, v2, v10
+; GCN-NEXT: v_cmp_gt_u32_e64 s2, v3, v10
+; GCN-NEXT: s_and_b32 s0, vcc_lo, s0
+; GCN-NEXT: s_or_b32 s1, s2, s1
+; GCN-NEXT: s_and_b32 s0, s0, s1
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C
%cmp2 = icmp ult i32 %arg2, %C
%cmp3 = icmp ugt i32 %arg3, %C
@@ -1821,24 +1952,24 @@ define i1 @test105(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %a
}
define i1 @test106(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %arg6, i32 %arg7, i32 %arg8, i32 %arg9, i32 %arg10, i32 %arg11, i32 %arg12, i32 %C1, i32 %C2) {
-; CHECK-LABEL: test106:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_min_u32_e32 v6, v6, v7
-; CHECK-NEXT: v_min_u32_e32 v0, v0, v1
-; CHECK-NEXT: v_min_u32_e32 v1, v10, v11
-; CHECK-NEXT: v_min_u32_e32 v2, v2, v3
-; CHECK-NEXT: v_min3_u32 v3, v4, v5, v6
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v12
-; CHECK-NEXT: v_min3_u32 v0, v8, v9, v1
-; CHECK-NEXT: v_cmp_lt_u32_e64 s0, v2, v13
-; CHECK-NEXT: v_cmp_lt_u32_e64 s1, v3, v13
-; CHECK-NEXT: v_cmp_lt_u32_e64 s2, v0, v12
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: s_or_b32 s0, s2, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test106:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_u32_e32 v6, v6, v7
+; GCN-NEXT: v_min_u32_e32 v0, v0, v1
+; GCN-NEXT: v_min_u32_e32 v1, v10, v11
+; GCN-NEXT: v_min_u32_e32 v2, v2, v3
+; GCN-NEXT: v_min3_u32 v3, v4, v5, v6
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v12
+; GCN-NEXT: v_min3_u32 v0, v8, v9, v1
+; GCN-NEXT: v_cmp_lt_u32_e64 s0, v2, v13
+; GCN-NEXT: v_cmp_lt_u32_e64 s1, v3, v13
+; GCN-NEXT: v_cmp_lt_u32_e64 s2, v0, v12
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: s_or_b32 s0, s0, s1
+; GCN-NEXT: s_or_b32 s0, s2, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %C1
%cmp2 = icmp ult i32 %arg2, %C1
%cmp3 = icmp ult i32 %arg3, %C2
@@ -1866,16 +1997,13 @@ define i1 @test106(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, i32 %a
}
define i1 @test107(float %arg1, float %arg2, float %arg3, float %C) {
-; CHECK-LABEL: test107:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v3
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v3
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v3
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test107:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min3_f32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v3
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %C
%cmp2 = fcmp olt float %arg2, %C
%cmp3 = fcmp olt float %arg3, %C
@@ -1885,16 +2013,21 @@ define i1 @test107(float %arg1, float %arg2, float %arg3, float %C) {
}
define i1 @test108(float %arg1, float %arg2, float %arg3, float %C) {
-; CHECK-LABEL: test108:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v3
-; CHECK-NEXT: v_cmp_nge_f32_e64 s0, v1, v3
-; CHECK-NEXT: v_cmp_nge_f32_e64 s1, v2, v3
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test108:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max3_f32 v0, v0, v1, v2
+; GFX11-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v3
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test108:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max3_f32 v0, v0, v1, v2
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v3
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ult float %arg1, %C
%cmp2 = fcmp ult float %arg2, %C
%cmp3 = fcmp ult float %arg3, %C
@@ -1904,18 +2037,27 @@ define i1 @test108(float %arg1, float %arg2, float %arg3, float %C) {
}
define i1 @test109(float %arg1, float %arg2, float %arg3, float %arg4, float %C) {
-; CHECK-LABEL: test109:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v4
-; CHECK-NEXT: v_cmp_gt_f32_e64 s1, v2, v4
-; CHECK-NEXT: v_cmp_gt_f32_e64 s2, v3, v4
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s1, s1, s2
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test109:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v0, v0, v0
+; GFX11-NEXT: v_dual_max_f32 v3, v3, v3 :: v_dual_max_f32 v2, v2, v2
+; GFX11-NEXT: v_dual_min_f32 v0, v0, v1 :: v_dual_max_f32 v1, v2, v3
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v4
+; GFX11-NEXT: v_cmp_gt_f32_e64 s0, v1, v4
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test109:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_dual_min_f32 v0, v0, v1 :: v_dual_max_f32 v1, v2, v3
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v4
+; GFX11NONANS-NEXT: v_cmp_gt_f32_e64 s0, v1, v4
+; GFX11NONANS-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %C
%cmp2 = fcmp olt float %arg2, %C
%cmp3 = fcmp ogt float %arg3, %C
@@ -1927,20 +2069,17 @@ define i1 @test109(float %arg1, float %arg2, float %arg3, float %arg4, float %C)
}
define i1 @test110(float %arg1, float %arg2, float %arg3, float %arg4, float %C1, float %C2, float %C3, float %C4, float %C) #0 {
-; CHECK-LABEL: test110:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
-; CHECK-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v8
-; CHECK-NEXT: v_cmp_gt_f32_e64 s1, v2, v8
-; CHECK-NEXT: v_cmp_gt_f32_e64 s2, v3, v8
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_and_b32 s1, s1, s2
-; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test110:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
+; GCN-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
+; GCN-NEXT: v_dual_max_f32 v0, v0, v1 :: v_dual_min_f32 v1, v2, v3
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
+; GCN-NEXT: v_cmp_gt_f32_e64 s0, v1, v8
+; GCN-NEXT: s_and_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, %C1
%add2 = fadd nnan float %arg2, %C2
%add3 = fadd nnan float %arg3, %C3
@@ -1956,24 +2095,28 @@ define i1 @test110(float %arg1, float %arg2, float %arg3, float %arg4, float %C1
}
define i1 @test111(float %arg1, float %arg2, float %arg3, float %arg4, float %arg5, float %arg6, float %arg7, float %arg8, float %C) {
-; CHECK-LABEL: test111:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s2, v3, v8
-; CHECK-NEXT: s_or_b32 s3, vcc_lo, s0
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v4, v8
-; CHECK-NEXT: s_or_b32 s2, s1, s2
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v5, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v6, v8
-; CHECK-NEXT: s_or_b32 s2, s3, s2
-; CHECK-NEXT: s_or_b32 s2, s2, vcc_lo
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: s_or_b32 s0, s0, s2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test111:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v3, v3, v3 :: v_dual_max_f32 v2, v2, v2
+; GFX11-NEXT: v_dual_min_f32 v2, v2, v3 :: v_dual_max_f32 v3, v4, v4
+; GFX11-NEXT: v_min3_f32 v0, v0, v1, v2
+; GFX11-NEXT: v_min_f32_e32 v0, v0, v3
+; GFX11-NEXT: v_min3_f32 v0, v5, v6, v0
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test111:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v2, v2, v3
+; GFX11NONANS-NEXT: v_min3_f32 v0, v0, v1, v2
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v4
+; GFX11NONANS-NEXT: v_min3_f32 v0, v5, v6, v0
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %C
%cmp2 = fcmp olt float %arg2, %C
%or1 = or i1 %cmp1, %cmp2
@@ -1993,24 +2136,30 @@ define i1 @test111(float %arg1, float %arg2, float %arg3, float %arg4, float %ar
}
define i1 @test112(float %arg1, float %arg2, float %arg3, float %arg4, float %arg5, float %arg6, float %arg7, float %arg8, float %C) {
-; CHECK-LABEL: test112:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s2, v3, v8
-; CHECK-NEXT: s_or_b32 s3, vcc_lo, s0
-; CHECK-NEXT: v_cmp_nge_f32_e32 vcc_lo, v4, v8
-; CHECK-NEXT: s_or_b32 s2, s1, s2
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v5, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v6, v8
-; CHECK-NEXT: s_or_b32 s2, s3, s2
-; CHECK-NEXT: s_or_b32 s2, s2, vcc_lo
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: s_or_b32 s0, s0, s2
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test112:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v3, v3, v3 :: v_dual_max_f32 v2, v2, v2
+; GFX11-NEXT: v_cmp_nge_f32_e32 vcc_lo, v4, v8
+; GFX11-NEXT: v_dual_max_f32 v5, v5, v5 :: v_dual_min_f32 v2, v2, v3
+; GFX11-NEXT: v_max_f32_e32 v3, v6, v6
+; GFX11-NEXT: v_min3_f32 v0, v0, v1, v2
+; GFX11-NEXT: v_min3_f32 v0, v0, v5, v3
+; GFX11-NEXT: v_cmp_lt_f32_e64 s0, v0, v8
+; GFX11-NEXT: s_or_b32 s0, s0, vcc_lo
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test112:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v2, v2, v3
+; GFX11NONANS-NEXT: v_min3_f32 v0, v0, v1, v2
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v4
+; GFX11NONANS-NEXT: v_min3_f32 v0, v5, v6, v0
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %C
%cmp2 = fcmp olt float %arg2, %C
%or1 = or i1 %cmp1, %cmp2
@@ -2030,16 +2179,24 @@ define i1 @test112(float %arg1, float %arg2, float %arg3, float %arg4, float %ar
}
define i1 @test113(float %arg1, float %arg2, float %arg3, float %C) {
-; CHECK-LABEL: test113:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v3
-; CHECK-NEXT: v_cmp_nge_f32_e64 s0, v1, v3
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v3
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test113:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v0, v0, v0
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v2, v3
+; GFX11-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_nge_f32_e64 s0, v0, v3
+; GFX11-NEXT: s_or_b32 s0, s0, vcc_lo
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test113:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_maxmin_f32 v0, v0, v1, v2
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v3
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ult float %arg1, %C
%cmp2 = fcmp ult float %arg2, %C
%cmp3 = fcmp olt float %arg3, %C
@@ -2049,16 +2206,26 @@ define i1 @test113(float %arg1, float %arg2, float %arg3, float %C) {
}
define i1 @test114(float %arg1, float %arg2, float %arg3, float %C) {
-; CHECK-LABEL: test114:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_gt_f32_e32 vcc_lo, v0, v3
-; CHECK-NEXT: v_cmp_gt_f32_e64 s0, v1, v3
-; CHECK-NEXT: v_cmp_nge_f32_e64 s1, v2, v3
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test114:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v0, v0, v0
+; GFX11-NEXT: v_cmp_nge_f32_e32 vcc_lo, v2, v3
+; GFX11-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11-NEXT: v_cmp_gt_f32_e64 s0, v0, v3
+; GFX11-NEXT: s_and_b32 s0, s0, vcc_lo
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test114:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v2, v3
+; GFX11NONANS-NEXT: v_cmp_gt_f32_e64 s0, v0, v3
+; GFX11NONANS-NEXT: s_and_b32 s0, s0, vcc_lo
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ogt float %arg1, %C
%cmp2 = fcmp ogt float %arg2, %C
%cmp3 = fcmp ult float %arg3, %C
@@ -2068,19 +2235,26 @@ define i1 @test114(float %arg1, float %arg2, float %arg3, float %C) {
}
define i1 @test115(float %arg1, float %arg2, float %arg3, float %arg4, float %C) {
-; CHECK-LABEL: test115:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v2, v2, v2 :: v_dual_max_f32 v3, v3, v3
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v4
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v4
-; CHECK-NEXT: v_cmp_nge_f32_e64 s1, v2, v4
-; CHECK-NEXT: v_cmp_nge_f32_e64 s2, v3, v4
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_and_b32 s1, s1, s2
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test115:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v2, v2, v2 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v3, v3, v3
+; GFX11-NEXT: v_dual_min_f32 v0, v0, v1 :: v_dual_max_f32 v1, v2, v3
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v4
+; GFX11-NEXT: v_cmp_nge_f32_e64 s0, v1, v4
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test115:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v2, v2, v3
+; GFX11NONANS-NEXT: v_min3_f32 v0, v0, v1, v2
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v4
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %C
%cmp2 = fcmp olt float %arg2, %C
%var3 = call float @llvm.canonicalize.f32(float %arg3)
@@ -2094,30 +2268,44 @@ define i1 @test115(float %arg1, float %arg2, float %arg3, float %arg4, float %C)
}
define i1 @test116(float %arg1, float %arg2, float %arg3, float %arg4, float %arg5, float %arg6, float %arg7, float %arg8, float %arg9, float %arg10, float %C) {
-; CHECK-LABEL: test116:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v10
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v10
-; CHECK-NEXT: v_cmp_gt_f32_e64 s1, v2, v10
-; CHECK-NEXT: v_cmp_gt_f32_e64 s2, v3, v10
-; CHECK-NEXT: v_cmp_lt_f32_e64 s3, v4, v10
-; CHECK-NEXT: v_cmp_lt_f32_e64 s4, v5, v10
-; CHECK-NEXT: v_cmp_gt_f32_e64 s5, v6, v10
-; CHECK-NEXT: v_cmp_gt_f32_e64 s6, v7, v10
-; CHECK-NEXT: v_cmp_lt_f32_e64 s7, v8, v10
-; CHECK-NEXT: v_cmp_lt_f32_e64 s8, v9, v10
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s1, s1, s2
-; CHECK-NEXT: s_or_b32 s2, s3, s4
-; CHECK-NEXT: s_or_b32 s3, s5, s6
-; CHECK-NEXT: s_or_b32 s4, s7, s8
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: s_or_b32 s1, s2, s3
-; CHECK-NEXT: s_or_b32 s0, s4, s0
-; CHECK-NEXT: s_or_b32 s0, s1, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test116:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v9, v9, v9 :: v_dual_max_f32 v8, v8, v8
+; GFX11-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v0, v0, v0
+; GFX11-NEXT: v_dual_max_f32 v3, v3, v3 :: v_dual_max_f32 v2, v2, v2
+; GFX11-NEXT: v_dual_max_f32 v5, v5, v5 :: v_dual_max_f32 v4, v4, v4
+; GFX11-NEXT: v_dual_max_f32 v7, v7, v7 :: v_dual_max_f32 v6, v6, v6
+; GFX11-NEXT: v_min_f32_e32 v8, v8, v9
+; GFX11-NEXT: v_dual_max_f32 v2, v2, v3 :: v_dual_min_f32 v3, v4, v5
+; GFX11-NEXT: v_max_f32_e32 v4, v6, v7
+; GFX11-NEXT: v_min3_f32 v0, v0, v1, v8
+; GFX11-NEXT: v_cmp_gt_f32_e32 vcc_lo, v2, v10
+; GFX11-NEXT: v_cmp_lt_f32_e64 s0, v3, v10
+; GFX11-NEXT: v_cmp_gt_f32_e64 s1, v4, v10
+; GFX11-NEXT: v_cmp_lt_f32_e64 s2, v0, v10
+; GFX11-NEXT: s_or_b32 s0, s0, s1
+; GFX11-NEXT: s_or_b32 s1, s2, vcc_lo
+; GFX11-NEXT: s_or_b32 s0, s0, s1
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test116:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v8, v8, v9
+; GFX11NONANS-NEXT: v_dual_max_f32 v2, v2, v3 :: v_dual_min_f32 v3, v4, v5
+; GFX11NONANS-NEXT: v_max_f32_e32 v4, v6, v7
+; GFX11NONANS-NEXT: v_min3_f32 v0, v0, v1, v8
+; GFX11NONANS-NEXT: v_cmp_gt_f32_e32 vcc_lo, v2, v10
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e64 s0, v3, v10
+; GFX11NONANS-NEXT: v_cmp_gt_f32_e64 s1, v4, v10
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e64 s2, v0, v10
+; GFX11NONANS-NEXT: s_or_b32 s0, s0, s1
+; GFX11NONANS-NEXT: s_or_b32 s1, s2, vcc_lo
+; GFX11NONANS-NEXT: s_or_b32 s0, s0, s1
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %C
%cmp2 = fcmp olt float %arg2, %C
%cmp3 = fcmp ogt float %arg3, %C
@@ -2141,34 +2329,45 @@ define i1 @test116(float %arg1, float %arg2, float %arg3, float %arg4, float %ar
}
define i1 @test117(float %arg1, float %arg2, float %arg3, float %arg4, float %arg5, float %arg6, float %arg7, float %arg8, float %arg9, float %arg10, float %arg11, float %arg12, float %C1, float %C2) {
-; CHECK-LABEL: test117:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v12
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v12
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v13
-; CHECK-NEXT: v_cmp_lt_f32_e64 s2, v3, v13
-; CHECK-NEXT: v_cmp_lt_f32_e64 s3, v4, v13
-; CHECK-NEXT: v_cmp_lt_f32_e64 s4, v5, v13
-; CHECK-NEXT: v_cmp_lt_f32_e64 s5, v6, v13
-; CHECK-NEXT: v_cmp_lt_f32_e64 s6, v7, v13
-; CHECK-NEXT: v_cmp_lt_f32_e64 s7, v8, v12
-; CHECK-NEXT: v_cmp_lt_f32_e64 s8, v9, v12
-; CHECK-NEXT: v_cmp_lt_f32_e64 s9, v10, v12
-; CHECK-NEXT: v_cmp_lt_f32_e64 s10, v11, v12
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s1, s1, s2
-; CHECK-NEXT: s_or_b32 s2, s3, s4
-; CHECK-NEXT: s_or_b32 s3, s5, s6
-; CHECK-NEXT: s_or_b32 s4, s7, s8
-; CHECK-NEXT: s_or_b32 s5, s9, s10
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: s_or_b32 s1, s2, s3
-; CHECK-NEXT: s_or_b32 s2, s4, s5
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: s_or_b32 s0, s2, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: test117:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v3, v3, v3 :: v_dual_max_f32 v6, v6, v6
+; GFX11-NEXT: v_dual_max_f32 v7, v7, v7 :: v_dual_max_f32 v10, v10, v10
+; GFX11-NEXT: v_dual_max_f32 v1, v1, v1 :: v_dual_max_f32 v0, v0, v0
+; GFX11-NEXT: v_dual_max_f32 v11, v11, v11 :: v_dual_max_f32 v2, v2, v2
+; GFX11-NEXT: v_min_f32_e32 v6, v6, v7
+; GFX11-NEXT: v_dual_min_f32 v0, v0, v1 :: v_dual_min_f32 v1, v10, v11
+; GFX11-NEXT: v_min_f32_e32 v2, v2, v3
+; GFX11-NEXT: v_min3_f32 v3, v4, v5, v6
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v12
+; GFX11-NEXT: v_min3_f32 v0, v8, v9, v1
+; GFX11-NEXT: v_cmp_lt_f32_e64 s0, v2, v13
+; GFX11-NEXT: v_cmp_lt_f32_e64 s1, v3, v13
+; GFX11-NEXT: v_cmp_lt_f32_e64 s2, v0, v12
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: s_or_b32 s0, s0, s1
+; GFX11-NEXT: s_or_b32 s0, s2, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test117:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v6, v6, v7
+; GFX11NONANS-NEXT: v_dual_min_f32 v0, v0, v1 :: v_dual_min_f32 v1, v10, v11
+; GFX11NONANS-NEXT: v_min_f32_e32 v2, v2, v3
+; GFX11NONANS-NEXT: v_min3_f32 v3, v4, v5, v6
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v12
+; GFX11NONANS-NEXT: v_min3_f32 v0, v8, v9, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e64 s0, v2, v13
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e64 s1, v3, v13
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e64 s2, v0, v12
+; GFX11NONANS-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11NONANS-NEXT: s_or_b32 s0, s0, s1
+; GFX11NONANS-NEXT: s_or_b32 s0, s2, s0
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %C1
%cmp2 = fcmp olt float %arg2, %C1
%cmp3 = fcmp olt float %arg3, %C2
@@ -2197,20 +2396,16 @@ define i1 @test117(float %arg1, float %arg2, float %arg3, float %arg4, float %ar
define i1 @test118(float %arg1, float %arg2, float %arg3, float %arg4, float %C1, float %C2, float %C3, float %C4, float %C) #0 {
-; CHECK-LABEL: test118:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
-; CHECK-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s2, v3, v8
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_and_b32 s1, s1, s2
-; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test118:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
+; GCN-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
+; GCN-NEXT: v_min_f32_e32 v0, v0, v1
+; GCN-NEXT: v_max3_f32 v0, v0, v2, v3
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, %C1
%add2 = fadd nnan float %arg2, %C2
%add3 = fadd nnan float %arg3, %C3
@@ -2226,20 +2421,16 @@ define i1 @test118(float %arg1, float %arg2, float %arg3, float %arg4, float %C1
}
define i1 @test119(float %arg1, float %arg2, float %arg3, float %arg4, float %C1, float %C2, float %C3, float %C4, float %C) #0 {
-; CHECK-LABEL: test119:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
-; CHECK-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s2, v3, v8
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s1, s1, s2
-; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test119:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
+; GCN-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
+; GCN-NEXT: v_min_f32_e32 v2, v2, v3
+; GCN-NEXT: v_minmax_f32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, %C1
%add2 = fadd nnan float %arg2, %C2
%add3 = fadd nnan float %arg3, %C3
@@ -2255,20 +2446,16 @@ define i1 @test119(float %arg1, float %arg2, float %arg3, float %arg4, float %C1
}
define i1 @test120(float %arg1, float %arg2, float %arg3, float %arg4, float %C1, float %C2, float %C3, float %C4, float %C) #0 {
-; CHECK-LABEL: test120:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
-; CHECK-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s2, v3, v8
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_and_b32 s1, s1, s2
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test120:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
+; GCN-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
+; GCN-NEXT: v_max_f32_e32 v2, v2, v3
+; GCN-NEXT: v_min3_f32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, %C1
%add2 = fadd nnan float %arg2, %C2
%add3 = fadd nnan float %arg3, %C3
@@ -2284,20 +2471,16 @@ define i1 @test120(float %arg1, float %arg2, float %arg3, float %arg4, float %C1
}
define i1 @test121(float %arg1, float %arg2, float %arg3, float %arg4, float %C1, float %C2, float %C3, float %C4, float %C) #0 {
-; CHECK-LABEL: test121:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
-; CHECK-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s1, v2, v8
-; CHECK-NEXT: v_cmp_lt_f32_e64 s2, v3, v8
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_and_b32 s1, s1, s2
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+; GCN-LABEL: test121:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v3, v3, v7
+; GCN-NEXT: v_dual_add_f32 v0, v0, v4 :: v_dual_add_f32 v1, v1, v5
+; GCN-NEXT: v_max_f32_e32 v2, v2, v3
+; GCN-NEXT: v_maxmin_f32 v0, v0, v1, v2
+; GCN-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v8
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
%add1 = fadd nnan float %arg1, %C1
%add2 = fadd nnan float %arg2, %C2
%add3 = fadd nnan float %arg3, %C3
@@ -2312,141 +2495,173 @@ define i1 @test121(float %arg1, float %arg2, float %arg3, float %arg4, float %C1
ret i1 %and2
}
+define i1 @test122(double %arg1, double %arg2, double %arg3) #1 {
+; GCN-LABEL: test122:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_lt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
+ %cmp1 = fcmp ult double %arg1, %arg3
+ %cmp2 = fcmp ult double %arg2, %arg3
+ %or1 = or i1 %cmp1, %cmp2
+ ret i1 %or1
+}
+
+define i1 @test123(double %arg1, double %arg2, double %arg3) #1 {
+; GCN-LABEL: test123:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GCN-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GCN-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GCN-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GCN-NEXT: s_setpc_b64 s[30:31]
+ %var1 = call double @llvm.canonicalize.f64(double %arg1)
+ %var2 = call double @llvm.canonicalize.f64(double %arg2)
+ %cmp1 = fcmp ogt double %var1, %arg3
+ %cmp2 = fcmp ogt double %var2, %arg3
+ %or1 = and i1 %cmp1, %cmp2
+ ret i1 %or1
+}
+
; The optimization does not apply to the following tests.
-define i1 @test122(i32 %arg1, i64 %arg2) {
-; CHECK-LABEL: test122:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_mov_b64 s[0:1], 0x3e8
-; CHECK-NEXT: v_cmp_gt_i64_e32 vcc_lo, s[0:1], v[1:2]
-; CHECK-NEXT: v_cmp_gt_i32_e64 s0, 0x3e8, v0
-; CHECK-NEXT: s_or_b32 s0, s0, vcc_lo
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test124(i32 %arg1, i64 %arg2) {
+; GCN-LABEL: test124:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_mov_b64 s[0:1], 0x3e8
+; GCN-NEXT: v_cmp_gt_i64_e32 vcc_lo, s[0:1], v[1:2]
+; GCN-NEXT: v_cmp_gt_i32_e64 s0, 0x3e8, v0
+; GCN-NEXT: s_or_b32 s0, s0, vcc_lo
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp slt i32 %arg1, 1000
%cmp2 = icmp slt i64 %arg2, 1000
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test123(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test123:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cmp_eq_u32_e64 s0, 0x3e8, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test125(i32 %arg1, i32 %arg2) {
+; GCN-LABEL: test125:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cmp_eq_u32_e64 s0, 0x3e8, v1
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp eq i32 %arg1, 1000
%cmp2 = icmp eq i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test124(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test124:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0x3e8, v0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s0, 0x3e8, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test126(i32 %arg1, i32 %arg2) {
+; GCN-LABEL: test126:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0x3e8, v0
+; GCN-NEXT: v_cmp_ne_u32_e64 s0, 0x3e8, v1
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ne i32 %arg1, 1000
%cmp2 = icmp ne i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test125(i64 %arg1, i64 %arg2, i64 %arg3) {
-; CHECK-LABEL: test125:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_u64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_lt_u64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test127(i64 %arg1, i64 %arg2, i64 %arg3) {
+; GCN-LABEL: test127:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_lt_u64_e32 vcc_lo, v[0:1], v[4:5]
+; GCN-NEXT: v_cmp_lt_u64_e64 s0, v[2:3], v[4:5]
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i64 %arg1, %arg3
%cmp2 = icmp ult i64 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test126(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test126:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_lt_u32_e64 s0, v2, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test128(i32 %arg1, i32 %arg2, i32 %arg3) {
+; GCN-LABEL: test128:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cmp_lt_u32_e64 s0, v2, v1
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %arg3
%cmp2 = icmp ult i32 %arg3, %arg2
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test127(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test127:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_le_u32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test129(i32 %arg1, i32 %arg2, i32 %arg3) {
+; GCN-LABEL: test129:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cmp_le_u32_e64 s0, v1, v2
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %arg3
%cmp2 = icmp ule i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test128(i32 %arg1, i32 %arg2, i32 %arg3) {
-; CHECK-LABEL: test128:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_le_u32_e32 vcc_lo, v2, v0
-; CHECK-NEXT: v_cmp_gt_u32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test130(i32 %arg1, i32 %arg2, i32 %arg3) {
+; GCN-LABEL: test130:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_le_u32_e32 vcc_lo, v2, v0
+; GCN-NEXT: v_cmp_gt_u32_e64 s0, v1, v2
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ule i32 %arg3, %arg1
%cmp2 = icmp ugt i32 %arg2, %arg3
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test129(i16 %arg1, i32 %arg2) {
-; CHECK-LABEL: test129:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_gt_u16_e32 vcc_lo, 10, v0
-; CHECK-NEXT: v_cmp_gt_u32_e64 s0, 10, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test131(i16 %arg1, i32 %arg2) {
+; GCN-LABEL: test131:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_gt_u16_e32 vcc_lo, 10, v0
+; GCN-NEXT: v_cmp_gt_u32_e64 s0, 10, v1
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i16 %arg1, 10
%cmp2 = icmp ult i32 %arg2, 10
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test130(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4) {
-; CHECK-LABEL: test130:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_lt_u32_e64 s0, v1, v2
-; CHECK-NEXT: v_cmp_lt_u32_e64 s1, v0, v3
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: s_or_b32 s1, s1, vcc_lo
-; CHECK-NEXT: s_or_b32 s0, s0, s1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test132(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4) {
+; GCN-LABEL: test132:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_lt_u32_e32 vcc_lo, v0, v2
+; GCN-NEXT: v_cmp_lt_u32_e64 s0, v1, v2
+; GCN-NEXT: v_cmp_lt_u32_e64 s1, v0, v3
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: s_or_b32 s1, s1, vcc_lo
+; GCN-NEXT: s_or_b32 s0, s0, s1
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, %arg3
%cmp2 = icmp ult i32 %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
@@ -2456,62 +2671,88 @@ define i1 @test130(i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4) {
ret i1 %or3
}
-define i1 @test131(i32 %arg1, i32 %arg2) {
-; CHECK-LABEL: test131:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x64, v0
-; CHECK-NEXT: v_cmp_gt_u32_e64 s0, 0x3e8, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test133(i32 %arg1, i32 %arg2) {
+; GCN-LABEL: test133:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc_lo, 0x64, v0
+; GCN-NEXT: v_cmp_gt_u32_e64 s0, 0x3e8, v1
+; GCN-NEXT: s_or_b32 s0, vcc_lo, s0
+; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%cmp1 = icmp ult i32 %arg1, 100
%cmp2 = icmp ult i32 %arg2, 1000
%or = or i1 %cmp1, %cmp2
ret i1 %or
}
-define i1 @test132(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test132:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_gt_f32_e64 s0, v2, v1
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test134(float %arg1, float %arg2, float %arg3) #0 {
+; GFX11-LABEL: test134:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_gt_f32_e64 s0, v2, v1
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test134:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %arg3
%cmp2 = fcmp ogt float %arg3, %arg2
%and1 = and i1 %cmp1, %cmp2
ret i1 %and1
}
-define i1 @test133(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test133:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nle_f32_e64 s0, v2, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test135(float %arg1, float %arg2, float %arg3) #0 {
+; GFX11-LABEL: test135:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_nle_f32_e64 s0, v2, v1
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test135:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ult float %arg1, %arg3
%cmp2 = fcmp ugt float %arg3, %arg2
%or1 = or i1 %cmp1, %cmp2
ret i1 %or1
}
-define i1 @test134(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test134:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_ge_f64_e64 s0, v[4:5], v[2:3]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test136(double %arg1, double %arg2, double %arg3) {
+; GFX11-LABEL: test136:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_ge_f64_e64 s0, v[4:5], v[2:3]
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test136:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11NONANS-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ole double %var1, %arg3
@@ -2520,16 +2761,24 @@ define i1 @test134(double %arg1, double %arg2, double %arg3) {
ret i1 %and1
}
-define i1 @test135(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test135:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nlt_f32_e64 s0, v2, v1
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test137(float %arg1, float %arg2, float %arg3) {
+; GFX11-LABEL: test137:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_nlt_f32_e64 s0, v2, v1
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test137:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp ule float %var1, %arg3
@@ -2538,136 +2787,208 @@ define i1 @test135(float %arg1, float %arg2, float %arg3) {
ret i1 %or1
}
-define i1 @test136(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test136:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test138(float %arg1, float %arg2, float %arg3) #0 {
+; GFX11-LABEL: test138:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_lt_f32_e64 s0, v1, v2
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test138:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp olt float %arg1, %arg3
%cmp2 = fcmp olt float %arg2, %arg3
%and1 = and i1 %cmp1, %cmp2
ret i1 %and1
}
-define i1 @test137(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test137:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_le_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test139(double %arg1, double %arg2, double %arg3) #0 {
+; GFX11-LABEL: test139:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_le_f64_e64 s0, v[2:3], v[4:5]
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test139:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ole double %arg1, %arg3
%cmp2 = fcmp ole double %arg2, %arg3
%and1 = and i1 %cmp1, %cmp2
ret i1 %and1
}
-define i1 @test138(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test138:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_gt_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test140(double %arg1, double %arg2, double %arg3) #0 {
+; GFX11-LABEL: test140:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_gt_f64_e64 s0, v[2:3], v[4:5]
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test140:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ogt double %arg1, %arg3
%cmp2 = fcmp ogt double %arg2, %arg3
%and1 = and i1 %cmp1, %cmp2
ret i1 %and1
}
-define i1 @test139(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test139:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ge_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test141(float %arg1, float %arg2, float %arg3) #0 {
+; GFX11-LABEL: test141:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_ge_f32_e64 s0, v1, v2
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test141:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp oge float %arg1, %arg3
%cmp2 = fcmp oge float %arg2, %arg3
%and1 = and i1 %cmp1, %cmp2
ret i1 %and1
}
-define i1 @test140(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test140:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nle_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_nle_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test142(double %arg1, double %arg2, double %arg3) #0 {
+; GFX11-LABEL: test142:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_nle_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_nle_f64_e64 s0, v[2:3], v[4:5]
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test142:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ugt double %arg1, %arg3
%cmp2 = fcmp ugt double %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
ret i1 %or1
}
-define i1 @test141(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test141:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nlt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test143(float %arg1, float %arg2, float %arg3) #0 {
+; GFX11-LABEL: test143:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_nlt_f32_e64 s0, v1, v2
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test143:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp uge float %arg1, %arg3
%cmp2 = fcmp uge float %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
ret i1 %or1
}
-define i1 @test142(float %arg1, float %arg2, float %arg3) #0 {
-; CHECK-LABEL: test142:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ngt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test144(float %arg1, float %arg2, float %arg3) #0 {
+; GFX11-LABEL: test144:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_ngt_f32_e64 s0, v1, v2
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test144:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ule float %arg1, %arg3
%cmp2 = fcmp ule float %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
ret i1 %or1
}
-define i1 @test143(double %arg1, double %arg2, double %arg3) #0 {
-; CHECK-LABEL: test143:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cmp_nge_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_nge_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test145(double %arg1, double %arg2, double %arg3) #0 {
+; GFX11-LABEL: test145:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_cmp_nge_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_nge_f64_e64 s0, v[2:3], v[4:5]
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test145:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_lt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%cmp1 = fcmp ult double %arg1, %arg3
%cmp2 = fcmp ult double %arg2, %arg3
%or1 = or i1 %cmp1, %cmp2
ret i1 %or1
}
-define i1 @test144(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test144:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_lt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test146(float %arg1, float %arg2, float %arg3) {
+; GFX11-LABEL: test146:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_lt_f32_e64 s0, v1, v2
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test146:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp olt float %var1, %arg3
@@ -2676,17 +2997,27 @@ define i1 @test144(float %arg1, float %arg2, float %arg3) {
ret i1 %and1
}
-define i1 @test145(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test145:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_le_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test147(double %arg1, double %arg2, double %arg3) {
+; GFX11-LABEL: test147:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_le_f64_e64 s0, v[2:3], v[4:5]
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test147:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11NONANS-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_le_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ole double %var1, %arg3
@@ -2695,17 +3026,27 @@ define i1 @test145(double %arg1, double %arg2, double %arg3) {
ret i1 %and1
}
-define i1 @test146(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test146:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_gt_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test148(double %arg1, double %arg2, double %arg3) {
+; GFX11-LABEL: test148:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_gt_f64_e64 s0, v[2:3], v[4:5]
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test148:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11NONANS-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11NONANS-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ogt double %var1, %arg3
@@ -2714,16 +3055,24 @@ define i1 @test146(double %arg1, double %arg2, double %arg3) {
ret i1 %and1
}
-define i1 @test147(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test147:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ge_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_and_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test149(float %arg1, float %arg2, float %arg3) {
+; GFX11-LABEL: test149:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_ge_f32_e64 s0, v1, v2
+; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test149:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp oge float %var1, %arg3
@@ -2732,17 +3081,27 @@ define i1 @test147(float %arg1, float %arg2, float %arg3) {
ret i1 %and1
}
-define i1 @test148(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test148:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_nle_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_nle_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test150(double %arg1, double %arg2, double %arg3) {
+; GFX11-LABEL: test150:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11-NEXT: v_cmp_nle_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_nle_f64_e64 s0, v[2:3], v[4:5]
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test150:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11NONANS-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_gt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ugt double %var1, %arg3
@@ -2751,16 +3110,24 @@ define i1 @test148(double %arg1, double %arg2, double %arg3) {
ret i1 %or1
}
-define i1 @test149(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test149:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_nlt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test151(float %arg1, float %arg2, float %arg3) {
+; GFX11-LABEL: test151:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_nlt_f32_e64 s0, v1, v2
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test151:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp uge float %var1, %arg3
@@ -2769,16 +3136,24 @@ define i1 @test149(float %arg1, float %arg2, float %arg3) {
ret i1 %or1
}
-define i1 @test150(float %arg1, float %arg2, float %arg3) {
-; CHECK-LABEL: test150:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
-; CHECK-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
-; CHECK-NEXT: v_cmp_ngt_f32_e64 s0, v1, v2
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test152(float %arg1, float %arg2, float %arg3) {
+; GFX11-LABEL: test152:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_max_f32 v0, v0, v0 :: v_dual_max_f32 v1, v1, v1
+; GFX11-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v2
+; GFX11-NEXT: v_cmp_ngt_f32_e64 s0, v1, v2
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test152:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_min_f32_e32 v0, v0, v1
+; GFX11NONANS-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v2
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call float @llvm.canonicalize.f32(float %arg1)
%var2 = call float @llvm.canonicalize.f32(float %arg2)
%cmp1 = fcmp ule float %var1, %arg3
@@ -2787,17 +3162,27 @@ define i1 @test150(float %arg1, float %arg2, float %arg3) {
ret i1 %or1
}
-define i1 @test151(double %arg1, double %arg2, double %arg3) {
-; CHECK-LABEL: test151:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
-; CHECK-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
-; CHECK-NEXT: v_cmp_nge_f64_e32 vcc_lo, v[0:1], v[4:5]
-; CHECK-NEXT: v_cmp_nge_f64_e64 s0, v[2:3], v[4:5]
-; CHECK-NEXT: s_or_b32 s0, vcc_lo, s0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
-; CHECK-NEXT: s_setpc_b64 s[30:31]
+define i1 @test153(double %arg1, double %arg2, double %arg3) {
+; GFX11-LABEL: test153:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11-NEXT: v_cmp_nge_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11-NEXT: v_cmp_nge_f64_e64 s0, v[2:3], v[4:5]
+; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0
+; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11NONANS-LABEL: test153:
+; GFX11NONANS: ; %bb.0:
+; GFX11NONANS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11NONANS-NEXT: v_max_f64 v[0:1], v[0:1], v[0:1]
+; GFX11NONANS-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX11NONANS-NEXT: v_min_f64 v[0:1], v[0:1], v[2:3]
+; GFX11NONANS-NEXT: v_cmp_lt_f64_e32 vcc_lo, v[0:1], v[4:5]
+; GFX11NONANS-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
+; GFX11NONANS-NEXT: s_setpc_b64 s[30:31]
%var1 = call double @llvm.canonicalize.f64(double %arg1)
%var2 = call double @llvm.canonicalize.f64(double %arg2)
%cmp1 = fcmp ult double %var1, %arg3
@@ -2812,3 +3197,4 @@ declare half @llvm.canonicalize.f16(half)
declare <2 x half> @llvm.canonicalize.v2f16(<2 x half>)
attributes #0 = { nounwind "amdgpu-ieee"="false" }
+attributes #1 = { nounwind "unsafe-fp-math"="true" "no-nans-fp-math"="true" }
diff --git a/llvm/test/CodeGen/AMDGPU/fma.f16.ll b/llvm/test/CodeGen/AMDGPU/fma.f16.ll
index a9db224fc843b2..23971f2b681cb0 100644
--- a/llvm/test/CodeGen/AMDGPU/fma.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fma.f16.ll
@@ -92,18 +92,30 @@ define half @test_fmamk(half %x, half %y, half %z) {
; Regression test for a crash caused by D139469.
define i32 @test_D139469_f16(half %arg) {
-; GFX9-LABEL: test_D139469_f16:
-; GFX9: ; %bb.0: ; %bb
-; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_mul_f16_e32 v1, 0x291e, v0
-; GFX9-NEXT: s_movk_i32 s4, 0x291e
-; GFX9-NEXT: v_cmp_gt_f16_e32 vcc, 0, v1
-; GFX9-NEXT: v_mov_b32_e32 v1, 0x211e
-; GFX9-NEXT: v_fma_f16 v0, v0, s4, v1
-; GFX9-NEXT: v_cmp_gt_f16_e64 s[4:5], 0, v0
-; GFX9-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
-; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
-; GFX9-NEXT: s_setpc_b64 s[30:31]
+; GFX9-SDAG-LABEL: test_D139469_f16:
+; GFX9-SDAG: ; %bb.0: ; %bb
+; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-SDAG-NEXT: s_movk_i32 s4, 0x291e
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x211e
+; GFX9-SDAG-NEXT: v_mul_f16_e32 v1, 0x291e, v0
+; GFX9-SDAG-NEXT: v_fma_f16 v0, v0, s4, v2
+; GFX9-SDAG-NEXT: v_min_f16_e32 v0, v1, v0
+; GFX9-SDAG-NEXT: v_cmp_gt_f16_e32 vcc, 0, v0
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
+; GFX9-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-GISEL-LABEL: test_D139469_f16:
+; GFX9-GISEL: ; %bb.0: ; %bb
+; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-GISEL-NEXT: v_mul_f16_e32 v1, 0x291e, v0
+; GFX9-GISEL-NEXT: s_movk_i32 s4, 0x291e
+; GFX9-GISEL-NEXT: v_cmp_gt_f16_e32 vcc, 0, v1
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x211e
+; GFX9-GISEL-NEXT: v_fma_f16 v0, v0, s4, v1
+; GFX9-GISEL-NEXT: v_cmp_gt_f16_e64 s[4:5], 0, v0
+; GFX9-GISEL-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX9-GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
+; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-SDAG-LABEL: test_D139469_f16:
; GFX10-SDAG: ; %bb.0: ; %bb
@@ -111,10 +123,9 @@ define i32 @test_D139469_f16(half %arg) {
; GFX10-SDAG-NEXT: v_mov_b32_e32 v1, 0x211e
; GFX10-SDAG-NEXT: v_mul_f16_e32 v2, 0x291e, v0
; GFX10-SDAG-NEXT: v_fmac_f16_e32 v1, 0x291e, v0
-; GFX10-SDAG-NEXT: v_cmp_gt_f16_e32 vcc_lo, 0, v2
-; GFX10-SDAG-NEXT: v_cmp_gt_f16_e64 s4, 0, v1
-; GFX10-SDAG-NEXT: s_or_b32 s4, vcc_lo, s4
-; GFX10-SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, s4
+; GFX10-SDAG-NEXT: v_min_f16_e32 v0, v2, v1
+; GFX10-SDAG-NEXT: v_cmp_gt_f16_e32 vcc_lo, 0, v0
+; GFX10-SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: test_D139469_f16:
@@ -143,17 +154,14 @@ define <2 x i32> @test_D139469_v2f16(<2 x half> %arg) {
; GFX9-SDAG: ; %bb.0: ; %bb
; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-SDAG-NEXT: s_movk_i32 s4, 0x291e
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0x211e
; GFX9-SDAG-NEXT: v_pk_mul_f16 v1, v0, s4 op_sel_hi:[1,0]
+; GFX9-SDAG-NEXT: v_pk_fma_f16 v0, v0, s4, v2 op_sel_hi:[1,0,0]
+; GFX9-SDAG-NEXT: v_pk_min_f16 v1, v1, v0
; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, 0
; GFX9-SDAG-NEXT: v_cmp_gt_f16_e32 vcc, 0, v1
-; GFX9-SDAG-NEXT: v_cmp_lt_f16_sdwa s[6:7], v1, v2 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x211e
-; GFX9-SDAG-NEXT: v_pk_fma_f16 v0, v0, s4, v1 op_sel_hi:[1,0,0]
-; GFX9-SDAG-NEXT: v_cmp_gt_f16_e64 s[4:5], 0, v0
-; GFX9-SDAG-NEXT: v_cmp_lt_f16_sdwa s[8:9], v0, v2 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-SDAG-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
-; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
-; GFX9-SDAG-NEXT: s_or_b64 s[4:5], s[6:7], s[8:9]
+; GFX9-SDAG-NEXT: v_cmp_lt_f16_sdwa s[4:5], v1, v2 src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX9-SDAG-NEXT: v_cndmask_b32_e64 v1, 0, 1, s[4:5]
; GFX9-SDAG-NEXT: s_setpc_b64 s[30:31]
;
@@ -182,13 +190,10 @@ define <2 x i32> @test_D139469_v2f16(<2 x half> %arg) {
; GFX10-SDAG-NEXT: v_pk_mul_f16 v1, 0x291e, v0 op_sel_hi:[0,1]
; GFX10-SDAG-NEXT: v_pk_fma_f16 v0, 0x291e, v0, s4 op_sel_hi:[0,1,0]
; GFX10-SDAG-NEXT: v_mov_b32_e32 v2, 0
+; GFX10-SDAG-NEXT: v_pk_min_f16 v1, v1, v0
; GFX10-SDAG-NEXT: v_cmp_gt_f16_e32 vcc_lo, 0, v1
-; GFX10-SDAG-NEXT: v_cmp_gt_f16_e64 s4, 0, v0
-; GFX10-SDAG-NEXT: v_cmp_lt_f16_sdwa s5, v1, v2 src0_sel:WORD_1 src1_sel:DWORD
-; GFX10-SDAG-NEXT: v_cmp_lt_f16_sdwa s6, v0, v2 src0_sel:WORD_1 src1_sel:DWORD
-; GFX10-SDAG-NEXT: s_or_b32 s4, vcc_lo, s4
-; GFX10-SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, s4
-; GFX10-SDAG-NEXT: s_or_b32 s4, s5, s6
+; GFX10-SDAG-NEXT: v_cmp_lt_f16_sdwa s4, v1, v2 src0_sel:WORD_1 src1_sel:DWORD
+; GFX10-SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo
; GFX10-SDAG-NEXT: v_cndmask_b32_e64 v1, 0, 1, s4
; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
;
diff --git a/llvm/test/CodeGen/AMDGPU/or.ll b/llvm/test/CodeGen/AMDGPU/or.ll
index 74969a68dd5756..7759120273aa84 100644
--- a/llvm/test/CodeGen/AMDGPU/or.ll
+++ b/llvm/test/CodeGen/AMDGPU/or.ll
@@ -1160,77 +1160,74 @@ define amdgpu_kernel void @or_i1(ptr addrspace(1) %out, ptr addrspace(1) %in0, p
; GFX6-LABEL: or_i1:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd
-; GFX6-NEXT: s_mov_b32 s11, 0xf000
-; GFX6-NEXT: s_mov_b32 s10, -1
-; GFX6-NEXT: s_mov_b32 s2, s10
-; GFX6-NEXT: s_mov_b32 s3, s11
+; GFX6-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0xd
+; GFX6-NEXT: s_mov_b32 s3, 0xf000
+; GFX6-NEXT: s_mov_b32 s2, -1
+; GFX6-NEXT: s_mov_b32 s10, s2
+; GFX6-NEXT: s_mov_b32 s11, s3
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s12, s6
; GFX6-NEXT: s_mov_b32 s13, s7
-; GFX6-NEXT: s_mov_b32 s14, s10
-; GFX6-NEXT: s_mov_b32 s15, s11
-; GFX6-NEXT: buffer_load_dword v0, off, s[0:3], 0
+; GFX6-NEXT: s_mov_b32 s14, s2
+; GFX6-NEXT: s_mov_b32 s15, s3
+; GFX6-NEXT: buffer_load_dword v0, off, s[8:11], 0
; GFX6-NEXT: buffer_load_dword v1, off, s[12:15], 0
-; GFX6-NEXT: s_mov_b32 s8, s4
-; GFX6-NEXT: s_mov_b32 s9, s5
+; GFX6-NEXT: s_mov_b32 s0, s4
+; GFX6-NEXT: s_mov_b32 s1, s5
; GFX6-NEXT: s_waitcnt vmcnt(1)
-; GFX6-NEXT: v_cmp_le_f32_e32 vcc, 0, v0
+; GFX6-NEXT: v_mul_f32_e32 v0, 1.0, v0
; GFX6-NEXT: s_waitcnt vmcnt(0)
-; GFX6-NEXT: v_cmp_le_f32_e64 s[0:1], 0, v1
-; GFX6-NEXT: s_or_b64 s[0:1], s[0:1], vcc
-; GFX6-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1]
-; GFX6-NEXT: buffer_store_dword v0, off, s[8:11], 0
+; GFX6-NEXT: v_mul_f32_e32 v1, 1.0, v1
+; GFX6-NEXT: v_max_f32_e32 v0, v1, v0
+; GFX6-NEXT: v_cmp_le_f32_e32 vcc, 0, v0
+; GFX6-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
+; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: or_i1:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34
-; GFX8-NEXT: s_mov_b32 s11, 0xf000
-; GFX8-NEXT: s_mov_b32 s10, -1
-; GFX8-NEXT: s_mov_b32 s2, s10
-; GFX8-NEXT: s_mov_b32 s3, s11
+; GFX8-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x34
+; GFX8-NEXT: s_mov_b32 s3, 0xf000
+; GFX8-NEXT: s_mov_b32 s2, -1
+; GFX8-NEXT: s_mov_b32 s10, s2
+; GFX8-NEXT: s_mov_b32 s11, s3
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s12, s6
; GFX8-NEXT: s_mov_b32 s13, s7
-; GFX8-NEXT: s_mov_b32 s14, s10
-; GFX8-NEXT: s_mov_b32 s15, s11
-; GFX8-NEXT: buffer_load_dword v0, off, s[0:3], 0
+; GFX8-NEXT: s_mov_b32 s14, s2
+; GFX8-NEXT: s_mov_b32 s15, s3
+; GFX8-NEXT: buffer_load_dword v0, off, s[8:11], 0
; GFX8-NEXT: buffer_load_dword v1, off, s[12:15], 0
-; GFX8-NEXT: s_mov_b32 s8, s4
-; GFX8-NEXT: s_mov_b32 s9, s5
+; GFX8-NEXT: s_mov_b32 s0, s4
+; GFX8-NEXT: s_mov_b32 s1, s5
; GFX8-NEXT: s_waitcnt vmcnt(1)
-; GFX8-NEXT: v_cmp_le_f32_e32 vcc, 0, v0
+; GFX8-NEXT: v_mul_f32_e32 v0, 1.0, v0
; GFX8-NEXT: s_waitcnt vmcnt(0)
-; GFX8-NEXT: v_cmp_le_f32_e64 s[0:1], 0, v1
-; GFX8-NEXT: s_or_b64 s[0:1], s[0:1], vcc
-; GFX8-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1]
-; GFX8-NEXT: buffer_store_dword v0, off, s[8:11], 0
+; GFX8-NEXT: v_mul_f32_e32 v1, 1.0, v1
+; GFX8-NEXT: v_max_f32_e32 v0, v1, v0
+; GFX8-NEXT: v_cmp_le_f32_e32 vcc, 0, v0
+; GFX8-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
+; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX8-NEXT: s_endpgm
;
; EG-LABEL: or_i1:
; EG: ; %bb.0:
-; EG-NEXT: ALU 0, @12, KC0[CB0:0-32], KC1[]
-; EG-NEXT: TEX 0 @8
-; EG-NEXT: ALU 0, @13, KC0[CB0:0-32], KC1[]
-; EG-NEXT: TEX 0 @10
-; EG-NEXT: ALU 5, @14, KC0[CB0:0-32], KC1[]
+; EG-NEXT: ALU 1, @10, KC0[CB0:0-32], KC1[]
+; EG-NEXT: TEX 1 @6
+; EG-NEXT: ALU 4, @12, KC0[CB0:0-32], KC1[]
; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
; EG-NEXT: CF_END
; EG-NEXT: PAD
-; EG-NEXT: Fetch clause starting at 8:
-; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1
-; EG-NEXT: Fetch clause starting at 10:
+; EG-NEXT: Fetch clause starting at 6:
; EG-NEXT: VTX_READ_32 T1.X, T1.X, 0, #1
+; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1
+; EG-NEXT: ALU clause starting at 10:
+; EG-NEXT: MOV T0.X, KC0[2].Z,
+; EG-NEXT: MOV * T1.X, KC0[2].W,
; EG-NEXT: ALU clause starting at 12:
-; EG-NEXT: MOV * T0.X, KC0[2].W,
-; EG-NEXT: ALU clause starting at 13:
-; EG-NEXT: MOV * T1.X, KC0[2].Z,
-; EG-NEXT: ALU clause starting at 14:
-; EG-NEXT: SETGE_DX10 T0.W, T0.X, 0.0,
-; EG-NEXT: SETGE_DX10 * T1.W, T1.X, 0.0,
-; EG-NEXT: OR_INT * T0.W, PS, PV.W,
+; EG-NEXT: MAX_DX10 * T0.W, T0.X, T1.X,
+; EG-NEXT: SETGE_DX10 * T0.W, PV.W, 0.0,
; EG-NEXT: AND_INT T0.X, PV.W, 1,
; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,
; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00)
More information about the llvm-commits
mailing list