[llvm] [SDAG] Share signed zero handling for maximum and maximumnum (PR #142762)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Jun 4 03:13:16 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Nikita Popov (nikic)
<details>
<summary>Changes</summary>
Use the same code to handle signed zero ordering for maximum and maximumnum legalization.
For maximumnum, this reduces the number of comparisons and fixes legalization for the case where the same-sized integer type is not legal.
---
Patch is 2.57 MiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/142762.diff
7 Files Affected:
- (modified) llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp (+22-23)
- (modified) llvm/test/CodeGen/AMDGPU/fmax3-maximumnum.ll (+312-433)
- (modified) llvm/test/CodeGen/AMDGPU/fmin3-minimumnum.ll (+354-480)
- (modified) llvm/test/CodeGen/AMDGPU/maximumnum.bf16.ll (+7782-9469)
- (modified) llvm/test/CodeGen/AMDGPU/minimumnum.bf16.ll (+8103-9855)
- (modified) llvm/test/CodeGen/Mips/fp-maximumnum-minimumnum.ll (+30-60)
- (modified) llvm/test/CodeGen/X86/fminimumnum-fmaximumnum.ll (+396-203)
``````````diff
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index f34bf0ca7ede0..cefcda79477ee 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -8559,6 +8559,23 @@ SDValue TargetLowering::expandFMINNUM_FMAXNUM(SDNode *Node,
return SDValue();
}
+static SDValue emitSignedZeroOrdering(SelectionDAG &DAG, bool IsMax,
+ SDValue MinMax, SDValue LHS, SDValue RHS,
+ EVT CCVT, SDNodeFlags Flags,
+ const SDLoc &DL) {
+ EVT VT = MinMax.getValueType();
+ SDValue IsZero = DAG.getSetCC(DL, CCVT, MinMax,
+ DAG.getConstantFP(0.0, DL, VT), ISD::SETOEQ);
+ FloatSignAsInt State;
+ DAG.getSignAsIntValue(State, DL, LHS);
+ SDValue IsSpecificZero =
+ DAG.getSetCC(DL, CCVT, State.IntValue,
+ DAG.getConstant(0, DL, State.IntValue.getValueType()),
+ IsMax ? ISD::SETEQ : ISD::SETNE);
+ SDValue Sel = DAG.getSelect(DL, VT, IsSpecificZero, LHS, RHS, Flags);
+ return DAG.getSelect(DL, VT, IsZero, Sel, MinMax, Flags);
+}
+
SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
SelectionDAG &DAG) const {
if (SDValue Expanded = expandVectorNaryOpBySplitting(N, DAG))
@@ -8609,18 +8626,9 @@ SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
// fminimum/fmaximum requires -0.0 less than +0.0
if (!MinMaxMustRespectOrderedZero && !N->getFlags().hasNoSignedZeros() &&
- !DAG.isKnownNeverZeroFloat(RHS) && !DAG.isKnownNeverZeroFloat(LHS)) {
- SDValue IsZero = DAG.getSetCC(DL, CCVT, MinMax,
- DAG.getConstantFP(0.0, DL, VT), ISD::SETOEQ);
- FloatSignAsInt State;
- DAG.getSignAsIntValue(State, DL, LHS);
- SDValue IsSpecificZero =
- DAG.getSetCC(DL, CCVT, State.IntValue,
- DAG.getConstant(0, DL, State.IntValue.getValueType()),
- IsMax ? ISD::SETEQ : ISD::SETNE);
- SDValue Sel = DAG.getSelect(DL, VT, IsSpecificZero, LHS, RHS, Flags);
- MinMax = DAG.getSelect(DL, VT, IsZero, Sel, MinMax, Flags);
- }
+ !DAG.isKnownNeverZeroFloat(RHS) && !DAG.isKnownNeverZeroFloat(LHS))
+ return emitSignedZeroOrdering(DAG, IsMax, MinMax, LHS, RHS, CCVT, Flags,
+ DL);
return MinMax;
}
@@ -8697,17 +8705,8 @@ SDValue TargetLowering::expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
DAG.isKnownNeverZeroFloat(LHS) || DAG.isKnownNeverZeroFloat(RHS)) {
return MinMax;
}
- SDValue TestZero =
- DAG.getTargetConstant(IsMax ? fcPosZero : fcNegZero, DL, MVT::i32);
- SDValue IsZero = DAG.getSetCC(DL, CCVT, MinMax,
- DAG.getConstantFP(0.0, DL, VT), ISD::SETEQ);
- SDValue LCmp = DAG.getSelect(
- DL, VT, DAG.getNode(ISD::IS_FPCLASS, DL, CCVT, LHS, TestZero), LHS,
- MinMax, Flags);
- SDValue RCmp = DAG.getSelect(
- DL, VT, DAG.getNode(ISD::IS_FPCLASS, DL, CCVT, RHS, TestZero), RHS, LCmp,
- Flags);
- return DAG.getSelect(DL, VT, IsZero, RCmp, MinMax, Flags);
+ return emitSignedZeroOrdering(DAG, IsMax, MinMax, LHS, RHS, CCVT, Flags,
+ DL);
}
/// Returns a true value if if this FPClassTest can be performed with an ordered
diff --git a/llvm/test/CodeGen/AMDGPU/fmax3-maximumnum.ll b/llvm/test/CodeGen/AMDGPU/fmax3-maximumnum.ll
index 8c75b5c7c027e..5b7838b3e8237 100644
--- a/llvm/test/CodeGen/AMDGPU/fmax3-maximumnum.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmax3-maximumnum.ll
@@ -1714,30 +1714,26 @@ define bfloat @v_max3_bf16_maximumnum_maximumnum__v_v_v_0(bfloat %a, bfloat %b,
; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v0
; GFX8-NEXT: v_cmp_gt_f32_e32 vcc, v4, v3
; GFX8-NEXT: v_cndmask_b32_e32 v3, v1, v0, vcc
+; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3
; GFX8-NEXT: v_cmp_eq_u16_e32 vcc, 0, v0
-; GFX8-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc
-; GFX8-NEXT: v_cmp_eq_u16_e32 vcc, 0, v1
-; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v3
-; GFX8-NEXT: v_cmp_eq_f32_e32 vcc, 0, v1
+; GFX8-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
+; GFX8-NEXT: v_cmp_eq_f32_e32 vcc, 0, v4
; GFX8-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc
; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v0
; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
-; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v2
; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
-; GFX8-NEXT: v_cndmask_b32_e32 v2, v2, v0, vcc
-; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v0
-; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v2
-; GFX8-NEXT: v_cmp_gt_f32_e32 vcc, v1, v3
+; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
; GFX8-NEXT: v_cndmask_b32_e32 v1, v2, v0, vcc
; GFX8-NEXT: v_cmp_eq_u16_e32 vcc, 0, v0
+; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v0
+; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v1
+; GFX8-NEXT: v_cndmask_b32_e32 v2, v1, v0, vcc
+; GFX8-NEXT: v_cmp_gt_f32_e32 vcc, v3, v4
; GFX8-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
-; GFX8-NEXT: v_cmp_eq_u16_e32 vcc, 0, v2
+; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v0
+; GFX8-NEXT: v_cmp_eq_f32_e32 vcc, 0, v1
; GFX8-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX8-NEXT: v_lshlrev_b32_e32 v2, 16, v1
-; GFX8-NEXT: v_cmp_eq_f32_e32 vcc, 0, v2
-; GFX8-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX900-SDAG-LABEL: v_max3_bf16_maximumnum_maximumnum__v_v_v_0:
@@ -1745,38 +1741,34 @@ define bfloat @v_max3_bf16_maximumnum_maximumnum__v_v_v_0(bfloat %a, bfloat %b,
; GFX900-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v0
; GFX900-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
-; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v4, 16, v1
+; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v1
; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX900-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v4, v4
+; GFX900-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v1, v1, v0, vcc
-; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v0
-; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v4, 16, v1
-; GFX900-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, v3, v4
-; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v3, v1, v0, vcc
; GFX900-SDAG-NEXT: v_cmp_eq_u16_e32 vcc, 0, v0
-; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc
-; GFX900-SDAG-NEXT: v_cmp_eq_u16_e32 vcc, 0, v1
-; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v3
+; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v4, 16, v0
+; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v5, 16, v1
+; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v3, v1, v0, vcc
+; GFX900-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, v4, v5
+; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
+; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v0
; GFX900-SDAG-NEXT: v_cmp_eq_f32_e32 vcc, 0, v1
-; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc
+; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v0
; GFX900-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
-; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v2
; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX900-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
-; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v2, v2, v0, vcc
-; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v0
-; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v2
-; GFX900-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, v1, v3
+; GFX900-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v0, vcc
; GFX900-SDAG-NEXT: v_cmp_eq_u16_e32 vcc, 0, v0
+; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v0
+; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v4, 16, v1
+; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v2, v1, v0, vcc
+; GFX900-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, v3, v4
; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
-; GFX900-SDAG-NEXT: v_cmp_eq_u16_e32 vcc, 0, v2
+; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v0
+; GFX900-SDAG-NEXT: v_cmp_eq_f32_e32 vcc, 0, v1
; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX900-SDAG-NEXT: v_lshlrev_b32_e32 v2, 16, v1
-; GFX900-SDAG-NEXT: v_cmp_eq_f32_e32 vcc, 0, v2
-; GFX900-SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX900-SDAG-NEXT: s_setpc_b64 s[30:31]
;
; GFX950-SDAG-LABEL: v_max3_bf16_maximumnum_maximumnum__v_v_v_0:
@@ -1784,50 +1776,44 @@ define bfloat @v_max3_bf16_maximumnum_maximumnum__v_v_v_0(bfloat %a, bfloat %b,
; GFX950-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v0
; GFX950-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
-; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v4, 16, v1
+; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v1
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX950-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v4, v4
-; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v0
+; GFX950-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
+; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v4, 16, v0
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v1, v1, v0, vcc
-; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v4, 16, v1
-; GFX950-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, v3, v4
-; GFX950-SDAG-NEXT: s_nop 1
-; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v3, v1, v0, vcc
; GFX950-SDAG-NEXT: v_cmp_eq_u16_e32 vcc, 0, v0
+; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v5, 16, v1
+; GFX950-SDAG-NEXT: s_nop 0
+; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v3, v1, v0, vcc
+; GFX950-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, v4, v5
; GFX950-SDAG-NEXT: s_nop 1
-; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc
-; GFX950-SDAG-NEXT: v_cmp_eq_u16_e32 vcc, 0, v1
-; GFX950-SDAG-NEXT: s_nop 1
-; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v3
+; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
+; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v0
; GFX950-SDAG-NEXT: v_cmp_eq_f32_e32 vcc, 0, v1
; GFX950-SDAG-NEXT: s_nop 1
-; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc
+; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v0
; GFX950-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
-; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v2
+; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v2
; GFX950-SDAG-NEXT: s_nop 0
; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX950-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
-; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v0
+; GFX950-SDAG-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
+; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v0
; GFX950-SDAG-NEXT: s_nop 0
-; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v2, v2, v0, vcc
-; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v3, 16, v2
-; GFX950-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, v1, v3
-; GFX950-SDAG-NEXT: s_nop 1
; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v1, v2, v0, vcc
; GFX950-SDAG-NEXT: v_cmp_eq_u16_e32 vcc, 0, v0
+; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v4, 16, v1
+; GFX950-SDAG-NEXT: s_nop 0
+; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v2, v1, v0, vcc
+; GFX950-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, v3, v4
; GFX950-SDAG-NEXT: s_nop 1
; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
-; GFX950-SDAG-NEXT: v_cmp_eq_u16_e32 vcc, 0, v2
+; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v1, 16, v0
+; GFX950-SDAG-NEXT: v_cmp_eq_f32_e32 vcc, 0, v1
; GFX950-SDAG-NEXT: s_nop 1
; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX950-SDAG-NEXT: v_lshlrev_b32_e32 v2, 16, v1
-; GFX950-SDAG-NEXT: v_cmp_eq_f32_e32 vcc, 0, v2
-; GFX950-SDAG-NEXT: s_nop 1
-; GFX950-SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
; GFX950-SDAG-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_max3_bf16_maximumnum_maximumnum__v_v_v_0:
@@ -1845,9 +1831,7 @@ define bfloat @v_max3_bf16_maximumnum_maximumnum__v_v_v_0(bfloat %a, bfloat %b,
; GFX10-NEXT: v_cndmask_b32_e32 v3, v1, v0, vcc_lo
; GFX10-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v0
; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3
-; GFX10-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc_lo
-; GFX10-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v1
-; GFX10-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo
; GFX10-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v4
; GFX10-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc_lo
; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v2
@@ -1862,9 +1846,7 @@ define bfloat @v_max3_bf16_maximumnum_maximumnum__v_v_v_0(bfloat %a, bfloat %b,
; GFX10-NEXT: v_cndmask_b32_e32 v2, v1, v0, vcc_lo
; GFX10-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v0
; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v2
-; GFX10-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
-; GFX10-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v1
-; GFX10-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX10-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo
; GFX10-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v3
; GFX10-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo
; GFX10-NEXT: s_setpc_b64 s[30:31]
@@ -1886,48 +1868,43 @@ define bfloat @v_max3_bf16_maximumnum_maximumnum__v_v_v_0(bfloat %a, bfloat %b,
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v4.l, v1.l
; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3
-; GFX11-SDAG-TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 0, v1.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4
; GFX11-SDAG-TRUE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, v3, v4
; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v3.l, v1.l, v0.l, vcc_lo
; GFX11-SDAG-TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v0.l
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v4.l, v3.l
-; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v0.l, vcc_lo
+; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, vcc_lo
; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v1.l, s0
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-SDAG-TRUE16-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v4
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v0.l, vcc_lo
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l
; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v1
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX11-SDAG-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1
; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2
; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v0.h, vcc_lo
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.l
; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v1.l, v0.h, v0.l, vcc_lo
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v2
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX11-SDAG-TRUE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, v2, v3
; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v2.l, v1.l, v0.l, vcc_lo
; GFX11-SDAG-TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v0.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3)
; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo
-; GFX11-SDAG-TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v1.l
+; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, vcc_lo
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v1.l, vcc_lo
; GFX11-SDAG-TRUE16-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v3
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3)
; GFX11-SDAG-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
@@ -1935,44 +1912,39 @@ define bfloat @v_max3_bf16_maximumnum_maximumnum__v_v_v_0(bfloat %a, bfloat %b,
; GFX11-SDAG-FAKE16: ; %bb.0:
; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v0
-; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX11-SDAG-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3
; GFX11-SDAG-FAKE16-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v1
+; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v0
+; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX11-SDAG-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4
-; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-FAKE16-NEXT: v_cndmask_b32_e32 v1, v1, v0, vcc_lo
-; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v0
; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v1
; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX11-SDAG-FAKE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, v3, v4
; GFX11-SDAG-FAKE16-NEXT: v_cndmask_b32_e32 v3, v1, v0, vcc_lo
; GFX11-SDAG-FAKE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v0
; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v3
-; GFX11-SDAG-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc_lo
-; GFX11-SDAG-FAKE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 0, v1
-; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-FAKE16-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX11-SDAG-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo
+; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-FAKE16-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v4
-; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-FAKE16-NEXT: v_dual_cndmask_b32 v0, v3, v0 :: v_dual_lshlrev_b32 v3, 16, v2
+; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0
-; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4)
; GFX11-SDAG-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1
; GFX11-SDAG-FAKE16-NEXT: v_cndmask...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/142762
More information about the llvm-commits
mailing list