[llvm-branch-commits] [llvm] [GlobalISel] Canonicalize more constants (PR #236956)
Alan Li via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Thu Oct 29 16:50:21 PDT 2026
https://github.com/lialan updated https://github.com/llvm/llvm-project/pull/236956
>From 7fb5748de0726bed5a4009f33b7b354e5efa17c3 Mon Sep 17 00:00:00 2001
From: Alan Li <me at alanli.org>
Date: Thu, 29 Oct 2026 12:09:03 -0700
Subject: [PATCH] [GlobalISel] Canonicalize more constants
Move scalar constants behind casts and copies, and vector splats, to
the RHS of commutative operations. Keep two constant operands in their
existing order.
Retain commuted identity coverage for casts and splats that constant
folding cannot simplify, and update affected AArch64 and AMDGPU output.
---
.../lib/CodeGen/GlobalISel/CombinerHelper.cpp | 11 +++-
.../combine-commuted-constant-identity.mir | 44 +++++++++++++++
llvm/test/CodeGen/AArch64/arm64-zip.ll | 4 +-
llvm/test/CodeGen/AArch64/fpclamptosat_vec.ll | 43 +++-----------
.../test/CodeGen/AArch64/fptosi-sat-vector.ll | 6 +-
llvm/test/CodeGen/AArch64/hadd-combine.ll | 12 ++--
.../AArch64/neon-bitwise-instructions.ll | 4 +-
llvm/test/CodeGen/AArch64/reassocmls.ll | 2 +-
.../CodeGen/AArch64/selectcc-to-shiftand.ll | 56 ++++++-------------
llvm/test/CodeGen/AArch64/uadd_sat_vec.ll | 4 +-
llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll | 8 +--
llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll | 8 +--
llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll | 40 ++++++-------
13 files changed, 121 insertions(+), 121 deletions(-)
create mode 100644 llvm/test/CodeGen/AArch64/GlobalISel/combine-commuted-constant-identity.mir
diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
index 451984a814187d..feb67ba8a84fee 100644
--- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
@@ -7096,7 +7096,14 @@ bool CombinerHelper::matchCommuteConstantToRHS(MachineInstr &MI) const {
!mi_match(RHS, MRI, m_MInstr(RHSDef)))
return false;
- if (!getIConstantVRegVal(LHS, MRI)) {
+ // Recognize scalar constants through casts and copies, as well as vector
+ // splats.
+ const auto IsConstant = [&](Register Reg) {
+ return getIConstantVRegValWithLookThrough(Reg, MRI) ||
+ getIConstantSplatVal(Reg, MRI);
+ };
+
+ if (!IsConstant(LHS)) {
// Skip commuting if LHS is not a constant. But, LHS may be a
// G_CONSTANT_FOLD_BARRIER. If so we commute as long as we don't already
// have a constant on the RHS.
@@ -7105,7 +7112,7 @@ bool CombinerHelper::matchCommuteConstantToRHS(MachineInstr &MI) const {
}
// Commute as long as RHS is not a constant or G_CONSTANT_FOLD_BARRIER.
return RHSDef->getOpcode() != TargetOpcode::G_CONSTANT_FOLD_BARRIER &&
- !getIConstantVRegVal(RHS, MRI);
+ !IsConstant(RHS);
}
bool CombinerHelper::matchCommuteFPConstantToRHS(MachineInstr &MI) const {
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/combine-commuted-constant-identity.mir b/llvm/test/CodeGen/AArch64/GlobalISel/combine-commuted-constant-identity.mir
new file mode 100644
index 00000000000000..a8c5abd7b48658
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/combine-commuted-constant-identity.mir
@@ -0,0 +1,44 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=aarch64 -run-pass=aarch64-prelegalizer-combiner \
+# RUN: -aarch64prelegalizercombiner-only-enable-rule=right_identity_one_int,G_MUL_commute_constant_to_rhs,constant_fold_binop \
+# RUN: -verify-machineinstrs %s -o - | FileCheck %s
+
+# Both operands are recognized as constants, so canonicalization does not
+# commute them. Constant folding cannot handle the extension or vector splats.
+# The commuted identity rule must still be available.
+
+---
+name: mul_one_zext_constant
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: mul_one_zext_constant
+ ; CHECK: %five:_(i32) = G_CONSTANT i32 5
+ ; CHECK-NEXT: %wide:_(i64) = G_ZEXT %five(i32)
+ ; CHECK-NEXT: $x0 = COPY %wide(i64)
+ ; CHECK-NEXT: RET_ReallyLR implicit $x0
+ %one:_(i64) = G_CONSTANT i64 1
+ %five:_(i32) = G_CONSTANT i32 5
+ %wide:_(i64) = G_ZEXT %five(i32)
+ %mul:_(i64) = G_MUL %one, %wide
+ $x0 = COPY %mul(i64)
+ RET_ReallyLR implicit $x0
+...
+---
+name: mul_one_splat_constant
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: mul_one_splat_constant
+ ; CHECK: %five:_(i32) = G_CONSTANT i32 5
+ ; CHECK-NEXT: %fives:_(<2 x i32>) = G_BUILD_VECTOR %five(i32), %five(i32)
+ ; CHECK-NEXT: $d0 = COPY %fives(<2 x i32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %one:_(i32) = G_CONSTANT i32 1
+ %five:_(i32) = G_CONSTANT i32 5
+ %ones:_(<2 x i32>) = G_BUILD_VECTOR %one(i32), %one(i32)
+ %fives:_(<2 x i32>) = G_BUILD_VECTOR %five(i32), %five(i32)
+ %mul:_(<2 x i32>) = G_MUL %ones, %fives
+ $d0 = COPY %mul(<2 x i32>)
+ RET_ReallyLR implicit $d0
+...
diff --git a/llvm/test/CodeGen/AArch64/arm64-zip.ll b/llvm/test/CodeGen/AArch64/arm64-zip.ll
index 4726a0842fddb4..a257192b20e23b 100644
--- a/llvm/test/CodeGen/AArch64/arm64-zip.ll
+++ b/llvm/test/CodeGen/AArch64/arm64-zip.ll
@@ -443,7 +443,7 @@ define <4 x i32> @shuffle_zip2(<4 x i32> %arg) {
; CHECK-GI-NEXT: dup.2s v0, v0[0]
; CHECK-GI-NEXT: dup.2s v1, v1[0]
; CHECK-GI-NEXT: mov.d v1[1], v0[0]
-; CHECK-GI-NEXT: and.16b v0, v2, v1
+; CHECK-GI-NEXT: and.16b v0, v1, v2
; CHECK-GI-NEXT: ret
bb:
%inst = icmp ult <4 x i32> zeroinitializer, %arg
@@ -472,7 +472,7 @@ define <4 x i32> @shuffle_zip3(<4 x i32> %arg) {
; CHECK-GI-NEXT: dup.2s v0, v0[0]
; CHECK-GI-NEXT: dup.2s v2, v2[0]
; CHECK-GI-NEXT: mov.d v2[1], v0[0]
-; CHECK-GI-NEXT: and.16b v0, v1, v2
+; CHECK-GI-NEXT: and.16b v0, v2, v1
; CHECK-GI-NEXT: ret
bb:
%inst = icmp slt <4 x i32> zeroinitializer, %arg
diff --git a/llvm/test/CodeGen/AArch64/fpclamptosat_vec.ll b/llvm/test/CodeGen/AArch64/fpclamptosat_vec.ll
index fa464d324672fc..2427a021454481 100644
--- a/llvm/test/CodeGen/AArch64/fpclamptosat_vec.ll
+++ b/llvm/test/CodeGen/AArch64/fpclamptosat_vec.ll
@@ -3448,41 +3448,14 @@ entry:
; i32 non saturate
define <4 x i32> @ustest_f16i32_nsat(<4 x half> %x) {
-; CHECK-CVT-SD-LABEL: ustest_f16i32_nsat:
-; CHECK-CVT-SD: // %bb.0: // %entry
-; CHECK-CVT-SD-NEXT: fcvtl v0.4s, v0.4h
-; CHECK-CVT-SD-NEXT: movi v1.2d, #0000000000000000
-; CHECK-CVT-SD-NEXT: fcvtzs v0.4s, v0.4s
-; CHECK-CVT-SD-NEXT: smin v0.4s, v0.4s, v1.4s
-; CHECK-CVT-SD-NEXT: smax v0.4s, v0.4s, v1.4s
-; CHECK-CVT-SD-NEXT: ret
-;
-; CHECK-FP16-SD-LABEL: ustest_f16i32_nsat:
-; CHECK-FP16-SD: // %bb.0: // %entry
-; CHECK-FP16-SD-NEXT: fcvtl v0.4s, v0.4h
-; CHECK-FP16-SD-NEXT: movi v1.2d, #0000000000000000
-; CHECK-FP16-SD-NEXT: fcvtzs v0.4s, v0.4s
-; CHECK-FP16-SD-NEXT: smin v0.4s, v0.4s, v1.4s
-; CHECK-FP16-SD-NEXT: smax v0.4s, v0.4s, v1.4s
-; CHECK-FP16-SD-NEXT: ret
-;
-; CHECK-CVT-GI-LABEL: ustest_f16i32_nsat:
-; CHECK-CVT-GI: // %bb.0: // %entry
-; CHECK-CVT-GI-NEXT: fcvtl v0.4s, v0.4h
-; CHECK-CVT-GI-NEXT: movi v1.2d, #0000000000000000
-; CHECK-CVT-GI-NEXT: fcvtzs v0.4s, v0.4s
-; CHECK-CVT-GI-NEXT: smin v0.4s, v1.4s, v0.4s
-; CHECK-CVT-GI-NEXT: smax v0.4s, v0.4s, v1.4s
-; CHECK-CVT-GI-NEXT: ret
-;
-; CHECK-FP16-GI-LABEL: ustest_f16i32_nsat:
-; CHECK-FP16-GI: // %bb.0: // %entry
-; CHECK-FP16-GI-NEXT: fcvtl v0.4s, v0.4h
-; CHECK-FP16-GI-NEXT: movi v1.2d, #0000000000000000
-; CHECK-FP16-GI-NEXT: fcvtzs v0.4s, v0.4s
-; CHECK-FP16-GI-NEXT: smin v0.4s, v1.4s, v0.4s
-; CHECK-FP16-GI-NEXT: smax v0.4s, v0.4s, v1.4s
-; CHECK-FP16-GI-NEXT: ret
+; CHECK-LABEL: ustest_f16i32_nsat:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fcvtl v0.4s, v0.4h
+; CHECK-NEXT: movi v1.2d, #0000000000000000
+; CHECK-NEXT: fcvtzs v0.4s, v0.4s
+; CHECK-NEXT: smin v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: smax v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: ret
entry:
%conv = fptosi <4 x half> %x to <4 x i32>
%spec.store.select = call <4 x i32> @llvm.smin.v4i32(<4 x i32> zeroinitializer, <4 x i32> %conv)
diff --git a/llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll b/llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll
index 145aef9123a4e9..020338ce4b64e5 100644
--- a/llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll
+++ b/llvm/test/CodeGen/AArch64/fptosi-sat-vector.ll
@@ -2366,9 +2366,9 @@ define <2 x i1> @test_signed_v2f64_v2i1(<2 x double> %f) {
; CHECK-GI-NEXT: fcvtzs v0.2d, v0.2d
; CHECK-GI-NEXT: cmlt v1.2d, v0.2d, #0
; CHECK-GI-NEXT: and v0.16b, v0.16b, v1.16b
-; CHECK-GI-NEXT: movi v1.2d, #0xffffffffffffffff
-; CHECK-GI-NEXT: cmge v2.2d, v0.2d, #0
-; CHECK-GI-NEXT: bif v0.16b, v1.16b, v2.16b
+; CHECK-GI-NEXT: cmge v1.2d, v0.2d, #0
+; CHECK-GI-NEXT: and v0.16b, v0.16b, v1.16b
+; CHECK-GI-NEXT: orn v0.16b, v0.16b, v1.16b
; CHECK-GI-NEXT: xtn v0.2s, v0.2d
; CHECK-GI-NEXT: ret
%x = call <2 x i1> @llvm.fptosi.sat.v2f64.v2i1(<2 x double> %f)
diff --git a/llvm/test/CodeGen/AArch64/hadd-combine.ll b/llvm/test/CodeGen/AArch64/hadd-combine.ll
index 7738b73a411239..a27dcc143431cc 100644
--- a/llvm/test/CodeGen/AArch64/hadd-combine.ll
+++ b/llvm/test/CodeGen/AArch64/hadd-combine.ll
@@ -542,11 +542,9 @@ define <8 x i16> @rhaddu_const_lhs(<8 x i16> %src1) {
;
; CHECK-GI-LABEL: rhaddu_const_lhs:
; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: movi v1.4s, #1
+; CHECK-GI-NEXT: movi v1.4s, #2
; CHECK-GI-NEXT: uaddw v2.4s, v1.4s, v0.4h
-; CHECK-GI-NEXT: uaddw2 v0.4s, v1.4s, v0.8h
-; CHECK-GI-NEXT: add v2.4s, v2.4s, v1.4s
-; CHECK-GI-NEXT: add v1.4s, v0.4s, v1.4s
+; CHECK-GI-NEXT: uaddw2 v1.4s, v1.4s, v0.8h
; CHECK-GI-NEXT: shrn v0.4h, v2.4s, #1
; CHECK-GI-NEXT: shrn2 v0.8h, v1.4s, #1
; CHECK-GI-NEXT: ret
@@ -787,11 +785,9 @@ define <8 x i16> @rhadds_const_lhs(<8 x i16> %src1) {
;
; CHECK-GI-LABEL: rhadds_const_lhs:
; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: movi v1.4s, #1
+; CHECK-GI-NEXT: movi v1.4s, #2
; CHECK-GI-NEXT: saddw v2.4s, v1.4s, v0.4h
-; CHECK-GI-NEXT: saddw2 v0.4s, v1.4s, v0.8h
-; CHECK-GI-NEXT: add v2.4s, v2.4s, v1.4s
-; CHECK-GI-NEXT: add v1.4s, v0.4s, v1.4s
+; CHECK-GI-NEXT: saddw2 v1.4s, v1.4s, v0.8h
; CHECK-GI-NEXT: shrn v0.4h, v2.4s, #1
; CHECK-GI-NEXT: shrn2 v0.8h, v1.4s, #1
; CHECK-GI-NEXT: ret
diff --git a/llvm/test/CodeGen/AArch64/neon-bitwise-instructions.ll b/llvm/test/CodeGen/AArch64/neon-bitwise-instructions.ll
index 5b4c8dd6f6a946..4006edd9d027cd 100644
--- a/llvm/test/CodeGen/AArch64/neon-bitwise-instructions.ll
+++ b/llvm/test/CodeGen/AArch64/neon-bitwise-instructions.ll
@@ -2753,8 +2753,8 @@ define <8 x i16> @pr149380(<4 x i16> %u1, <1 x i64> %u2, <8 x i16> %vqshlu_n169)
; CHECK-GI-LABEL: pr149380:
; CHECK-GI: // %bb.0: // %entry
; CHECK-GI-NEXT: movi v0.8h, #1
-; CHECK-GI-NEXT: orr v1.16b, v0.16b, v2.16b
-; CHECK-GI-NEXT: sqadd v0.8h, v0.8h, v1.8h
+; CHECK-GI-NEXT: orr v1.16b, v2.16b, v0.16b
+; CHECK-GI-NEXT: sqadd v0.8h, v1.8h, v0.8h
; CHECK-GI-NEXT: ret
entry:
%mul.i = mul <8 x i16> %vqshlu_n169, < i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1 >
diff --git a/llvm/test/CodeGen/AArch64/reassocmls.ll b/llvm/test/CodeGen/AArch64/reassocmls.ll
index 2d9f033990e92b..4e19afb15dc0a7 100644
--- a/llvm/test/CodeGen/AArch64/reassocmls.ll
+++ b/llvm/test/CodeGen/AArch64/reassocmls.ll
@@ -284,7 +284,7 @@ define <8 x i16> @mla_v8i16_C(<8 x i16> %a, <8 x i16> %b, <8 x i16> %c, <8 x i16
; CHECK-GI-NEXT: mul v0.8h, v2.8h, v1.8h
; CHECK-GI-NEXT: movi v1.8h, #10
; CHECK-GI-NEXT: mla v0.8h, v4.8h, v3.8h
-; CHECK-GI-NEXT: add v0.8h, v1.8h, v0.8h
+; CHECK-GI-NEXT: add v0.8h, v0.8h, v1.8h
; CHECK-GI-NEXT: ret
%m1.neg = mul <8 x i16> %c, %b
%m2.neg = mul <8 x i16> %e, %d
diff --git a/llvm/test/CodeGen/AArch64/selectcc-to-shiftand.ll b/llvm/test/CodeGen/AArch64/selectcc-to-shiftand.ll
index 96a7a9d039c21b..bae00754b75f92 100644
--- a/llvm/test/CodeGen/AArch64/selectcc-to-shiftand.ll
+++ b/llvm/test/CodeGen/AArch64/selectcc-to-shiftand.ll
@@ -256,52 +256,34 @@ define <16 x i8> @sel_shift_bool_v16i8(<16 x i1> %t) {
; CHECK-GI-NEXT: shl v0.16b, v0.16b, #7
; CHECK-GI-NEXT: movi v1.16b, #128
; CHECK-GI-NEXT: cmlt v0.16b, v0.16b, #0
-; CHECK-GI-NEXT: and v0.16b, v1.16b, v0.16b
+; CHECK-GI-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-GI-NEXT: ret
%shl = select <16 x i1> %t, <16 x i8> <i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128>, <16 x i8> zeroinitializer
ret <16 x i8> %shl
}
define <8 x i16> @sel_shift_bool_v8i16(<8 x i1> %t) {
-; CHECK-SD-LABEL: sel_shift_bool_v8i16:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: ushll v0.8h, v0.8b, #0
-; CHECK-SD-NEXT: movi v1.8h, #128
-; CHECK-SD-NEXT: shl v0.8h, v0.8h, #15
-; CHECK-SD-NEXT: cmlt v0.8h, v0.8h, #0
-; CHECK-SD-NEXT: and v0.16b, v0.16b, v1.16b
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: sel_shift_bool_v8i16:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: ushll v0.8h, v0.8b, #0
-; CHECK-GI-NEXT: movi v1.8h, #128
-; CHECK-GI-NEXT: shl v0.8h, v0.8h, #15
-; CHECK-GI-NEXT: cmlt v0.8h, v0.8h, #0
-; CHECK-GI-NEXT: and v0.16b, v1.16b, v0.16b
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: sel_shift_bool_v8i16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ushll v0.8h, v0.8b, #0
+; CHECK-NEXT: movi v1.8h, #128
+; CHECK-NEXT: shl v0.8h, v0.8h, #15
+; CHECK-NEXT: cmlt v0.8h, v0.8h, #0
+; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
%shl= select <8 x i1> %t, <8 x i16> <i16 128, i16 128, i16 128, i16 128, i16 128, i16 128, i16 128, i16 128>, <8 x i16> zeroinitializer
ret <8 x i16> %shl
}
define <4 x i32> @sel_shift_bool_v4i32(<4 x i1> %t) {
-; CHECK-SD-LABEL: sel_shift_bool_v4i32:
-; CHECK-SD: // %bb.0:
-; CHECK-SD-NEXT: ushll v0.4s, v0.4h, #0
-; CHECK-SD-NEXT: movi v1.4s, #64
-; CHECK-SD-NEXT: shl v0.4s, v0.4s, #31
-; CHECK-SD-NEXT: cmlt v0.4s, v0.4s, #0
-; CHECK-SD-NEXT: and v0.16b, v0.16b, v1.16b
-; CHECK-SD-NEXT: ret
-;
-; CHECK-GI-LABEL: sel_shift_bool_v4i32:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: ushll v0.4s, v0.4h, #0
-; CHECK-GI-NEXT: movi v1.4s, #64
-; CHECK-GI-NEXT: shl v0.4s, v0.4s, #31
-; CHECK-GI-NEXT: cmlt v0.4s, v0.4s, #0
-; CHECK-GI-NEXT: and v0.16b, v1.16b, v0.16b
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: sel_shift_bool_v4i32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ushll v0.4s, v0.4h, #0
+; CHECK-NEXT: movi v1.4s, #64
+; CHECK-NEXT: shl v0.4s, v0.4s, #31
+; CHECK-NEXT: cmlt v0.4s, v0.4s, #0
+; CHECK-NEXT: and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
%shl = select <4 x i1> %t, <4 x i32> <i32 64, i32 64, i32 64, i32 64>, <4 x i32> zeroinitializer
ret <4 x i32> %shl
}
@@ -324,10 +306,8 @@ define <2 x i64> @sel_shift_bool_v2i64(<2 x i1> %t) {
; CHECK-GI-NEXT: ldr q1, [x8, :lo12:.LCPI16_0]
; CHECK-GI-NEXT: shl v0.2d, v0.2d, #63
; CHECK-GI-NEXT: cmlt v0.2d, v0.2d, #0
-; CHECK-GI-NEXT: and v0.16b, v1.16b, v0.16b
+; CHECK-GI-NEXT: and v0.16b, v0.16b, v1.16b
; CHECK-GI-NEXT: ret
%shl = select <2 x i1> %t, <2 x i64> <i64 65536, i64 65536>, <2 x i64> zeroinitializer
ret <2 x i64> %shl
}
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; CHECK: {{.*}}
diff --git a/llvm/test/CodeGen/AArch64/uadd_sat_vec.ll b/llvm/test/CodeGen/AArch64/uadd_sat_vec.ll
index 6768c733af47f2..e3a9545d91dff5 100644
--- a/llvm/test/CodeGen/AArch64/uadd_sat_vec.ll
+++ b/llvm/test/CodeGen/AArch64/uadd_sat_vec.ll
@@ -322,7 +322,7 @@ define <16 x i4> @v16i4(<16 x i4> %x, <16 x i4> %y) nounwind {
; CHECK-GI-NEXT: and v1.16b, v1.16b, v2.16b
; CHECK-GI-NEXT: cmhi v1.16b, v1.16b, v3.16b
; CHECK-GI-NEXT: eor v3.16b, v1.16b, v2.16b
-; CHECK-GI-NEXT: and v1.16b, v2.16b, v1.16b
+; CHECK-GI-NEXT: and v1.16b, v1.16b, v2.16b
; CHECK-GI-NEXT: and v0.16b, v0.16b, v3.16b
; CHECK-GI-NEXT: orr v0.16b, v1.16b, v0.16b
; CHECK-GI-NEXT: ret
@@ -344,7 +344,7 @@ define <16 x i1> @v16i1(<16 x i1> %x, <16 x i1> %y) nounwind {
; CHECK-GI-NEXT: and v1.16b, v1.16b, v2.16b
; CHECK-GI-NEXT: cmhi v1.16b, v1.16b, v3.16b
; CHECK-GI-NEXT: eor v3.16b, v1.16b, v2.16b
-; CHECK-GI-NEXT: and v1.16b, v2.16b, v1.16b
+; CHECK-GI-NEXT: and v1.16b, v1.16b, v2.16b
; CHECK-GI-NEXT: and v0.16b, v0.16b, v3.16b
; CHECK-GI-NEXT: orr v0.16b, v1.16b, v0.16b
; CHECK-GI-NEXT: ret
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll
index 8f10f74628a50a..0e3e68ae2849ae 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll
@@ -155,15 +155,15 @@ define <2 x i16> @test_max_K0min_K1Val__v2i16(<2 x i16> %a) {
; GFX9-LABEL: test_max_K0min_K1Val__v2i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_pk_min_i16 v0, 17, v0 op_sel_hi:[0,1]
-; GFX9-NEXT: v_pk_max_i16 v0, -12, v0 op_sel_hi:[0,1]
+; GFX9-NEXT: v_pk_min_i16 v0, v0, 17 op_sel_hi:[1,0]
+; GFX9-NEXT: v_pk_max_i16 v0, v0, -12 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: test_max_K0min_K1Val__v2i16:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_pk_min_i16 v0, 17, v0 op_sel_hi:[0,1]
-; GFX10-NEXT: v_pk_max_i16 v0, -12, v0 op_sel_hi:[0,1]
+; GFX10-NEXT: v_pk_min_i16 v0, v0, 17 op_sel_hi:[1,0]
+; GFX10-NEXT: v_pk_max_i16 v0, v0, -12 op_sel_hi:[1,0]
; GFX10-NEXT: s_setpc_b64 s[30:31]
%smin = call <2 x i16> @llvm.smin.v2i16(<2 x i16> <i16 17, i16 17>, <2 x i16> %a)
%smed = call <2 x i16> @llvm.smax.v2i16(<2 x i16> <i16 -12, i16 -12>, <2 x i16> %smin)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll
index d947340503e0d1..683dde0704bacd 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll
@@ -155,15 +155,15 @@ define <2 x i16> @test_max_K0min_K1Val__v2u16(<2 x i16> %a) {
; GFX9-LABEL: test_max_K0min_K1Val__v2u16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_pk_min_u16 v0, 17, v0 op_sel_hi:[0,1]
-; GFX9-NEXT: v_pk_max_u16 v0, 12, v0 op_sel_hi:[0,1]
+; GFX9-NEXT: v_pk_min_u16 v0, v0, 17 op_sel_hi:[1,0]
+; GFX9-NEXT: v_pk_max_u16 v0, v0, 12 op_sel_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: test_max_K0min_K1Val__v2u16:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_pk_min_u16 v0, 17, v0 op_sel_hi:[0,1]
-; GFX10-NEXT: v_pk_max_u16 v0, 12, v0 op_sel_hi:[0,1]
+; GFX10-NEXT: v_pk_min_u16 v0, v0, 17 op_sel_hi:[1,0]
+; GFX10-NEXT: v_pk_max_u16 v0, v0, 12 op_sel_hi:[1,0]
; GFX10-NEXT: s_setpc_b64 s[30:31]
%umin = call <2 x i16> @llvm.umin.v2i16(<2 x i16> <i16 17, i16 17>, <2 x i16> %a)
%umed = call <2 x i16> @llvm.umax.v2i16(<2 x i16> <i16 12, i16 12>, <2 x i16> %umin)
diff --git a/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll b/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll
index 3b14c9540e330d..7a95c19c729ffe 100644
--- a/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll
@@ -1601,8 +1601,8 @@ define i16 @basic_smax_smin_vec_input(<2 x i16> %src) {
; GISEL-GFX9: ; %bb.0:
; GISEL-GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, 0xff00ff
-; GISEL-GFX9-NEXT: v_pk_min_i16 v0, v1, v0
-; GISEL-GFX9-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX9-NEXT: v_pk_min_i16 v0, v0, v1
+; GISEL-GFX9-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v1, 8, v1
; GISEL-GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
@@ -1611,9 +1611,9 @@ define i16 @basic_smax_smin_vec_input(<2 x i16> %src) {
; GISEL-GFX11-TRUE16-LABEL: basic_smax_smin_vec_input:
; GISEL-GFX11-TRUE16: ; %bb.0:
; GISEL-GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-GFX11-TRUE16-NEXT: v_pk_min_i16 v0, 0xff00ff, v0
+; GISEL-GFX11-TRUE16-NEXT: v_pk_min_i16 v0, v0, 0xff00ff
; GISEL-GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX11-TRUE16-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX11-TRUE16-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX11-TRUE16-NEXT: v_and_b16 v0.l, v0.l, 0xff
; GISEL-GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GISEL-GFX11-TRUE16-NEXT: v_lshlrev_b16 v0.h, 8, v0.h
@@ -1623,9 +1623,9 @@ define i16 @basic_smax_smin_vec_input(<2 x i16> %src) {
; GISEL-GFX11-FAKE16-LABEL: basic_smax_smin_vec_input:
; GISEL-GFX11-FAKE16: ; %bb.0:
; GISEL-GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-GFX11-FAKE16-NEXT: v_pk_min_i16 v0, 0xff00ff, v0
+; GISEL-GFX11-FAKE16-NEXT: v_pk_min_i16 v0, v0, 0xff00ff
; GISEL-GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX11-FAKE16-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX11-FAKE16-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GISEL-GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 0xff, v0
; GISEL-GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
@@ -1640,9 +1640,9 @@ define i16 @basic_smax_smin_vec_input(<2 x i16> %src) {
; GISEL-GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GISEL-GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GISEL-GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
-; GISEL-GFX12-TRUE16-NEXT: v_pk_min_i16 v0, 0xff00ff, v0
+; GISEL-GFX12-TRUE16-NEXT: v_pk_min_i16 v0, v0, 0xff00ff
; GISEL-GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX12-TRUE16-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX12-TRUE16-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX12-TRUE16-NEXT: v_and_b16 v0.l, v0.l, 0xff
; GISEL-GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GISEL-GFX12-TRUE16-NEXT: v_lshlrev_b16 v0.h, 8, v0.h
@@ -1656,9 +1656,9 @@ define i16 @basic_smax_smin_vec_input(<2 x i16> %src) {
; GISEL-GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GISEL-GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GISEL-GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
-; GISEL-GFX12-FAKE16-NEXT: v_pk_min_i16 v0, 0xff00ff, v0
+; GISEL-GFX12-FAKE16-NEXT: v_pk_min_i16 v0, v0, 0xff00ff
; GISEL-GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX12-FAKE16-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX12-FAKE16-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GISEL-GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 0xff, v0
; GISEL-GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
@@ -1765,9 +1765,9 @@ define i16 @basic_smax_smin_vec_input_rev(<2 x i16> %src) {
; GISEL-GFX9-LABEL: basic_smax_smin_vec_input_rev:
; GISEL-GFX9: ; %bb.0:
; GISEL-GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-GFX9-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX9-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX9-NEXT: v_mov_b32_e32 v1, 0xff00ff
-; GISEL-GFX9-NEXT: v_pk_min_i16 v0, v1, v0
+; GISEL-GFX9-NEXT: v_pk_min_i16 v0, v0, v1
; GISEL-GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v1, 8, v1
; GISEL-GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
@@ -1776,9 +1776,9 @@ define i16 @basic_smax_smin_vec_input_rev(<2 x i16> %src) {
; GISEL-GFX11-TRUE16-LABEL: basic_smax_smin_vec_input_rev:
; GISEL-GFX11-TRUE16: ; %bb.0:
; GISEL-GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-GFX11-TRUE16-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX11-TRUE16-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX11-TRUE16-NEXT: v_pk_min_i16 v1, 0xff00ff, v0
+; GISEL-GFX11-TRUE16-NEXT: v_pk_min_i16 v1, v0, 0xff00ff
; GISEL-GFX11-TRUE16-NEXT: v_lshlrev_b16 v0.l, 8, v1.h
; GISEL-GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GISEL-GFX11-TRUE16-NEXT: v_or_b16 v0.l, v1.l, v0.l
@@ -1787,9 +1787,9 @@ define i16 @basic_smax_smin_vec_input_rev(<2 x i16> %src) {
; GISEL-GFX11-FAKE16-LABEL: basic_smax_smin_vec_input_rev:
; GISEL-GFX11-FAKE16: ; %bb.0:
; GISEL-GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-GFX11-FAKE16-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX11-FAKE16-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX11-FAKE16-NEXT: v_pk_min_i16 v0, 0xff00ff, v0
+; GISEL-GFX11-FAKE16-NEXT: v_pk_min_i16 v0, v0, 0xff00ff
; GISEL-GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GISEL-GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 0xff, v0
; GISEL-GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
@@ -1804,9 +1804,9 @@ define i16 @basic_smax_smin_vec_input_rev(<2 x i16> %src) {
; GISEL-GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0
; GISEL-GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0
; GISEL-GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0
-; GISEL-GFX12-TRUE16-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX12-TRUE16-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX12-TRUE16-NEXT: v_pk_min_i16 v1, 0xff00ff, v0
+; GISEL-GFX12-TRUE16-NEXT: v_pk_min_i16 v1, v0, 0xff00ff
; GISEL-GFX12-TRUE16-NEXT: v_lshlrev_b16 v0.l, 8, v1.h
; GISEL-GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GISEL-GFX12-TRUE16-NEXT: v_or_b16 v0.l, v1.l, v0.l
@@ -1819,9 +1819,9 @@ define i16 @basic_smax_smin_vec_input_rev(<2 x i16> %src) {
; GISEL-GFX12-FAKE16-NEXT: s_wait_samplecnt 0x0
; GISEL-GFX12-FAKE16-NEXT: s_wait_bvhcnt 0x0
; GISEL-GFX12-FAKE16-NEXT: s_wait_kmcnt 0x0
-; GISEL-GFX12-FAKE16-NEXT: v_pk_max_i16 v0, 0, v0
+; GISEL-GFX12-FAKE16-NEXT: v_pk_max_i16 v0, v0, 0
; GISEL-GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX12-FAKE16-NEXT: v_pk_min_i16 v0, 0xff00ff, v0
+; GISEL-GFX12-FAKE16-NEXT: v_pk_min_i16 v0, v0, 0xff00ff
; GISEL-GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GISEL-GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 0xff, v0
; GISEL-GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
More information about the llvm-branch-commits
mailing list