[llvm] [AMDGPU] Fix missing EOF space (PR #165667)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Oct 30 00:52:34 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
@llvm/pr-subscribers-llvm-globalisel
Author: Pankaj Dwivedi (PankajDwivedi-25)
<details>
<summary>Changes</summary>
---
Patch is 125.25 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/165667.diff
19 Files Affected:
- (modified) llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp (+5)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i32.ll (+26-37)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i64.ll (+28-39)
- (added) llvm/test/CodeGen/AMDGPU/amdgpu-miscellaneous-uniform-intrinsic.ll (+173)
- (modified) llvm/test/CodeGen/AMDGPU/amdgpu-simplify-uniform-waterfall.ll (+1)
- (modified) llvm/test/CodeGen/AMDGPU/amdgpu-uniform-intrinsic-combine.ll (+1)
- (modified) llvm/test/CodeGen/AMDGPU/amdgpu-uniform-temporal-divergence.ll (+1)
- (modified) llvm/test/CodeGen/AMDGPU/fix-sgpr-copies-wwm.ll (+4-14)
- (modified) llvm/test/CodeGen/AMDGPU/llc-pipeline-npm.ll (+3-3)
- (modified) llvm/test/CodeGen/AMDGPU/llc-pipeline.ll (+23)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ballot.i32.ll (+29-29)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ballot.i64.ll (+29-29)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.permlane64.ll (+19-56)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.permlane64.ptr.ll (+23-54)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.readfirstlane.ll (+22-26)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.readlane.ll (+23-26)
- (modified) llvm/test/CodeGen/AMDGPU/spill-vgpr-to-agpr-update-regscavenger.ll (+9-14)
- (modified) llvm/test/CodeGen/AMDGPU/splitkit-getsubrangeformask.ll (+98-100)
- (modified) llvm/test/CodeGen/AMDGPU/wqm.ll (+6-12)
``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 75a94ac891819..b28c50e3f5b6d 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -1315,6 +1315,9 @@ void AMDGPUPassConfig::addIRPasses() {
isPassEnabled(EnableImageIntrinsicOptimizer))
addPass(createAMDGPUImageIntrinsicOptimizerPass(&TM));
+ if (EnableUniformIntrinsicCombine)
+ addPass(createAMDGPUUniformIntrinsicCombineLegacyPass());
+
// This can be disabled by passing ::Disable here or on the command line
// with --expand-variadics-override=disable.
addPass(createExpandVariadicsPass(ExpandVariadicsMode::Lowering));
@@ -2066,6 +2069,8 @@ void AMDGPUCodeGenPassBuilder::addIRPasses(AddIRPass &addPass) const {
if (isPassEnabled(EnableImageIntrinsicOptimizer))
addPass(AMDGPUImageIntrinsicOptimizerPass(TM));
+ if (EnableUniformIntrinsicCombine)
+ addPass(AMDGPUUniformIntrinsicCombinePass());
// This can be disabled by passing ::Disable here or on the command line
// with --expand-variadics-override=disable.
addPass(ExpandVariadicsPass(ExpandVariadicsMode::Lowering));
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i32.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i32.ll
index 7714c032d1737..d3e211855d7ed 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i32.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i32.ll
@@ -113,9 +113,9 @@ false:
define amdgpu_cs i32 @branch_uniform_ballot_ne_zero_non_compare(i32 inreg %v) {
; CHECK-LABEL: branch_uniform_ballot_ne_zero_non_compare:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s0, 0, s0
-; CHECK-NEXT: s_cmp_eq_u32 s0, 0
+; CHECK-NEXT: s_xor_b32 s0, s0, 1
+; CHECK-NEXT: s_and_b32 s0, s0, 1
+; CHECK-NEXT: s_cmp_lg_u32 s0, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB8_2
; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
@@ -161,16 +161,17 @@ false:
define amdgpu_cs i32 @branch_uniform_ballot_eq_zero_non_compare(i32 inreg %v) {
; CHECK-LABEL: branch_uniform_ballot_eq_zero_non_compare:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s0, 0, s0
+; CHECK-NEXT: s_xor_b32 s0, s0, 1
+; CHECK-NEXT: s_xor_b32 s0, s0, 1
+; CHECK-NEXT: s_and_b32 s0, s0, 1
; CHECK-NEXT: s_cmp_lg_u32 s0, 0
-; CHECK-NEXT: s_cbranch_scc0 .LBB10_2
-; CHECK-NEXT: ; %bb.1: ; %false
-; CHECK-NEXT: s_mov_b32 s0, 33
-; CHECK-NEXT: s_branch .LBB10_3
-; CHECK-NEXT: .LBB10_2: ; %true
+; CHECK-NEXT: s_cbranch_scc1 .LBB10_2
+; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
; CHECK-NEXT: s_branch .LBB10_3
+; CHECK-NEXT: .LBB10_2: ; %false
+; CHECK-NEXT: s_mov_b32 s0, 33
+; CHECK-NEXT: s_branch .LBB10_3
; CHECK-NEXT: .LBB10_3:
%c = trunc i32 %v to i1
%ballot = call i32 @llvm.amdgcn.ballot.i32(i1 %c)
@@ -208,11 +209,7 @@ false:
define amdgpu_cs i32 @branch_uniform_ballot_ne_zero_compare(i32 inreg %v) {
; CHECK-LABEL: branch_uniform_ballot_ne_zero_compare:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_cmp_lt_u32 s0, 12
-; CHECK-NEXT: s_cselect_b32 s0, 1, 0
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s0, 0, s0
-; CHECK-NEXT: s_cmp_eq_u32 s0, 0
+; CHECK-NEXT: s_cmp_ge_u32 s0, 12
; CHECK-NEXT: s_cbranch_scc1 .LBB12_2
; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
@@ -258,17 +255,13 @@ define amdgpu_cs i32 @branch_uniform_ballot_eq_zero_compare(i32 inreg %v) {
; CHECK-LABEL: branch_uniform_ballot_eq_zero_compare:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_cmp_lt_u32 s0, 12
-; CHECK-NEXT: s_cselect_b32 s0, 1, 0
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s0, 0, s0
-; CHECK-NEXT: s_cmp_lg_u32 s0, 0
-; CHECK-NEXT: s_cbranch_scc0 .LBB14_2
-; CHECK-NEXT: ; %bb.1: ; %false
-; CHECK-NEXT: s_mov_b32 s0, 33
-; CHECK-NEXT: s_branch .LBB14_3
-; CHECK-NEXT: .LBB14_2: ; %true
+; CHECK-NEXT: s_cbranch_scc1 .LBB14_2
+; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
; CHECK-NEXT: s_branch .LBB14_3
+; CHECK-NEXT: .LBB14_2: ; %false
+; CHECK-NEXT: s_mov_b32 s0, 33
+; CHECK-NEXT: s_branch .LBB14_3
; CHECK-NEXT: .LBB14_3:
%c = icmp ult i32 %v, 12
%ballot = call i32 @llvm.amdgcn.ballot.i32(i1 %c)
@@ -310,14 +303,12 @@ false:
define amdgpu_cs i32 @branch_uniform_ballot_ne_zero_and(i32 inreg %v1, i32 inreg %v2) {
; CHECK-LABEL: branch_uniform_ballot_ne_zero_and:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_cmp_lt_u32 s0, 12
+; CHECK-NEXT: s_cmp_ge_u32 s0, 12
; CHECK-NEXT: s_cselect_b32 s0, 1, 0
-; CHECK-NEXT: s_cmp_gt_u32 s1, 34
+; CHECK-NEXT: s_cmp_le_u32 s1, 34
; CHECK-NEXT: s_cselect_b32 s1, 1, 0
-; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s0, 0, s0
-; CHECK-NEXT: s_cmp_eq_u32 s0, 0
+; CHECK-NEXT: s_or_b32 s0, s0, s1
+; CHECK-NEXT: s_cmp_lg_u32 s0, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB16_2
; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
@@ -372,16 +363,14 @@ define amdgpu_cs i32 @branch_uniform_ballot_eq_zero_and(i32 inreg %v1, i32 inreg
; CHECK-NEXT: s_cmp_gt_u32 s1, 34
; CHECK-NEXT: s_cselect_b32 s1, 1, 0
; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s0, 0, s0
; CHECK-NEXT: s_cmp_lg_u32 s0, 0
-; CHECK-NEXT: s_cbranch_scc0 .LBB18_2
-; CHECK-NEXT: ; %bb.1: ; %false
-; CHECK-NEXT: s_mov_b32 s0, 33
-; CHECK-NEXT: s_branch .LBB18_3
-; CHECK-NEXT: .LBB18_2: ; %true
+; CHECK-NEXT: s_cbranch_scc1 .LBB18_2
+; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
; CHECK-NEXT: s_branch .LBB18_3
+; CHECK-NEXT: .LBB18_2: ; %false
+; CHECK-NEXT: s_mov_b32 s0, 33
+; CHECK-NEXT: s_branch .LBB18_3
; CHECK-NEXT: .LBB18_3:
%v1c = icmp ult i32 %v1, 12
%v2c = icmp ugt i32 %v2, 34
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i64.ll
index 7b8166948610b..250fbc7c0f147 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ballot.i64.ll
@@ -116,9 +116,9 @@ false:
define amdgpu_cs i32 @branch_uniform_ballot_ne_zero_non_compare(i32 inreg %v) {
; CHECK-LABEL: branch_uniform_ballot_ne_zero_non_compare:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0
-; CHECK-NEXT: s_cmp_eq_u64 s[0:1], 0
+; CHECK-NEXT: s_xor_b32 s0, s0, 1
+; CHECK-NEXT: s_and_b32 s0, s0, 1
+; CHECK-NEXT: s_cmp_lg_u32 s0, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB8_2
; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
@@ -164,16 +164,17 @@ false:
define amdgpu_cs i32 @branch_uniform_ballot_eq_zero_non_compare(i32 inreg %v) {
; CHECK-LABEL: branch_uniform_ballot_eq_zero_non_compare:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0
-; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0
-; CHECK-NEXT: s_cbranch_scc0 .LBB10_2
-; CHECK-NEXT: ; %bb.1: ; %false
-; CHECK-NEXT: s_mov_b32 s0, 33
-; CHECK-NEXT: s_branch .LBB10_3
-; CHECK-NEXT: .LBB10_2: ; %true
+; CHECK-NEXT: s_xor_b32 s0, s0, 1
+; CHECK-NEXT: s_xor_b32 s0, s0, 1
+; CHECK-NEXT: s_and_b32 s0, s0, 1
+; CHECK-NEXT: s_cmp_lg_u32 s0, 0
+; CHECK-NEXT: s_cbranch_scc1 .LBB10_2
+; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
; CHECK-NEXT: s_branch .LBB10_3
+; CHECK-NEXT: .LBB10_2: ; %false
+; CHECK-NEXT: s_mov_b32 s0, 33
+; CHECK-NEXT: s_branch .LBB10_3
; CHECK-NEXT: .LBB10_3:
%c = trunc i32 %v to i1
%ballot = call i64 @llvm.amdgcn.ballot.i64(i1 %c)
@@ -211,11 +212,7 @@ false:
define amdgpu_cs i32 @branch_uniform_ballot_ne_zero_compare(i32 inreg %v) {
; CHECK-LABEL: branch_uniform_ballot_ne_zero_compare:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_cmp_lt_u32 s0, 12
-; CHECK-NEXT: s_cselect_b32 s0, 1, 0
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0
-; CHECK-NEXT: s_cmp_eq_u64 s[0:1], 0
+; CHECK-NEXT: s_cmp_ge_u32 s0, 12
; CHECK-NEXT: s_cbranch_scc1 .LBB12_2
; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
@@ -261,17 +258,13 @@ define amdgpu_cs i32 @branch_uniform_ballot_eq_zero_compare(i32 inreg %v) {
; CHECK-LABEL: branch_uniform_ballot_eq_zero_compare:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_cmp_lt_u32 s0, 12
-; CHECK-NEXT: s_cselect_b32 s0, 1, 0
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0
-; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0
-; CHECK-NEXT: s_cbranch_scc0 .LBB14_2
-; CHECK-NEXT: ; %bb.1: ; %false
-; CHECK-NEXT: s_mov_b32 s0, 33
-; CHECK-NEXT: s_branch .LBB14_3
-; CHECK-NEXT: .LBB14_2: ; %true
+; CHECK-NEXT: s_cbranch_scc1 .LBB14_2
+; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
; CHECK-NEXT: s_branch .LBB14_3
+; CHECK-NEXT: .LBB14_2: ; %false
+; CHECK-NEXT: s_mov_b32 s0, 33
+; CHECK-NEXT: s_branch .LBB14_3
; CHECK-NEXT: .LBB14_3:
%c = icmp ult i32 %v, 12
%ballot = call i64 @llvm.amdgcn.ballot.i64(i1 %c)
@@ -313,14 +306,12 @@ false:
define amdgpu_cs i32 @branch_uniform_ballot_ne_zero_and(i32 inreg %v1, i32 inreg %v2) {
; CHECK-LABEL: branch_uniform_ballot_ne_zero_and:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_cmp_lt_u32 s0, 12
+; CHECK-NEXT: s_cmp_ge_u32 s0, 12
; CHECK-NEXT: s_cselect_b32 s0, 1, 0
-; CHECK-NEXT: s_cmp_gt_u32 s1, 34
+; CHECK-NEXT: s_cmp_le_u32 s1, 34
; CHECK-NEXT: s_cselect_b32 s1, 1, 0
-; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0
-; CHECK-NEXT: s_cmp_eq_u64 s[0:1], 0
+; CHECK-NEXT: s_or_b32 s0, s0, s1
+; CHECK-NEXT: s_cmp_lg_u32 s0, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB16_2
; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
@@ -375,16 +366,14 @@ define amdgpu_cs i32 @branch_uniform_ballot_eq_zero_and(i32 inreg %v1, i32 inreg
; CHECK-NEXT: s_cmp_gt_u32 s1, 34
; CHECK-NEXT: s_cselect_b32 s1, 1, 0
; CHECK-NEXT: s_and_b32 s0, s0, s1
-; CHECK-NEXT: s_and_b32 s0, 1, s0
-; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, s0
-; CHECK-NEXT: s_cmp_lg_u64 s[0:1], 0
-; CHECK-NEXT: s_cbranch_scc0 .LBB18_2
-; CHECK-NEXT: ; %bb.1: ; %false
-; CHECK-NEXT: s_mov_b32 s0, 33
-; CHECK-NEXT: s_branch .LBB18_3
-; CHECK-NEXT: .LBB18_2: ; %true
+; CHECK-NEXT: s_cmp_lg_u32 s0, 0
+; CHECK-NEXT: s_cbranch_scc1 .LBB18_2
+; CHECK-NEXT: ; %bb.1: ; %true
; CHECK-NEXT: s_mov_b32 s0, 42
; CHECK-NEXT: s_branch .LBB18_3
+; CHECK-NEXT: .LBB18_2: ; %false
+; CHECK-NEXT: s_mov_b32 s0, 33
+; CHECK-NEXT: s_branch .LBB18_3
; CHECK-NEXT: .LBB18_3:
%v1c = icmp ult i32 %v1, 12
%v2c = icmp ugt i32 %v2, 34
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-miscellaneous-uniform-intrinsic.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-miscellaneous-uniform-intrinsic.ll
new file mode 100644
index 0000000000000..34d4c519851d4
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-miscellaneous-uniform-intrinsic.ll
@@ -0,0 +1,173 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -o - %s | FileCheck %s
+define amdgpu_kernel void @readfirstlane_with_readfirstlane(ptr addrspace(1) %out) {
+; CHECK-LABEL: readfirstlane_with_readfirstlane:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
+; CHECK-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 5
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: global_store_b32 v0, v1, s[0:1]
+; CHECK-NEXT: s_endpgm
+ %v1 = call i32 @llvm.amdgcn.readfirstlane(i32 5)
+ %v2 = call i32 @llvm.amdgcn.readfirstlane(i32 %v1)
+ store i32 %v2, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @readfirstlane_with_readlane(ptr addrspace(1) %out) {
+; CHECK-LABEL: readfirstlane_with_readlane:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
+; CHECK-NEXT: v_bfe_u32 v1, v0, 10, 10
+; CHECK-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; CHECK-NEXT: v_readfirstlane_b32 s2, v1
+; CHECK-NEXT: v_readlane_b32 s2, v0, s2
+; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; CHECK-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: global_store_b32 v0, v1, s[0:1]
+; CHECK-NEXT: s_endpgm
+ %tidx = call i32 @llvm.amdgcn.workitem.id.x()
+ %tidy = call i32 @llvm.amdgcn.workitem.id.y()
+ %v1 = call i32 @llvm.amdgcn.readlane(i32 %tidx, i32 %tidy)
+ %v2 = call i32 @llvm.amdgcn.readfirstlane(i32 %v1)
+ store i32 %v2, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @readlane_with_firstlane(ptr addrspace(1) %out) {
+; CHECK-LABEL: readlane_with_firstlane:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
+; CHECK-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; CHECK-NEXT: v_readfirstlane_b32 s2, v0
+; CHECK-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: global_store_b32 v0, v1, s[0:1]
+; CHECK-NEXT: s_endpgm
+ %tidx = call i32 @llvm.amdgcn.workitem.id.x()
+ %v1 = call i32 @llvm.amdgcn.readfirstlane(i32 %tidx)
+ %v2 = call i32 @llvm.amdgcn.readlane(i32 %v1, i32 3)
+ store i32 %v2, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @readlane_readlane(ptr addrspace(1) %out) {
+; CHECK-LABEL: readlane_readlane:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
+; CHECK-NEXT: v_bfe_u32 v1, v0, 10, 10
+; CHECK-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; CHECK-NEXT: v_readfirstlane_b32 s2, v1
+; CHECK-NEXT: v_readlane_b32 s2, v0, s2
+; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; CHECK-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: global_store_b32 v0, v1, s[0:1]
+; CHECK-NEXT: s_endpgm
+ %tidx = call i32 @llvm.amdgcn.workitem.id.x()
+ %tidy = call i32 @llvm.amdgcn.workitem.id.y()
+ %v1 = call i32 @llvm.amdgcn.readlane(i32 %tidx, i32 %tidy)
+ %v2 = call i32 @llvm.amdgcn.readlane(i32 %v1, i32 2)
+ store i32 %v2, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @permlane64_uniform(ptr addrspace(1) %out, i32 %src) {
+; CHECK-LABEL: permlane64_uniform:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_clause 0x1
+; CHECK-NEXT: s_load_b32 s2, s[4:5], 0x8
+; CHECK-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; CHECK-NEXT: global_store_b32 v0, v1, s[0:1]
+; CHECK-NEXT: s_endpgm
+ %v = call i32 @llvm.amdgcn.permlane64(i32 %src)
+ store i32 %v, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @permlane64_nonuniform(i32 addrspace(1)* %out) {
+; CHECK-LABEL: permlane64_nonuniform:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
+; CHECK-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; CHECK-NEXT: v_permlane64_b32 v1, v0
+; CHECK-NEXT: v_lshlrev_b32_e32 v0, 2, v0
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: global_store_b32 v0, v1, s[0:1]
+; CHECK-NEXT: s_endpgm
+ %tid = call i32 @llvm.amdgcn.workitem.id.x()
+ %v = call i32 @llvm.amdgcn.permlane64(i32 %tid)
+ %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
+ store i32 %v, i32 addrspace(1)* %out_ptr
+ ret void
+}
+
+define amdgpu_kernel void @permlane64_nonuniform_expression(i32 addrspace(1)* %out) {
+; CHECK-LABEL: permlane64_nonuniform_expression:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
+; CHECK-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; CHECK-NEXT: v_add_nc_u32_e32 v1, 1, v0
+; CHECK-NEXT: v_lshlrev_b32_e32 v0, 2, v0
+; CHECK-NEXT: v_permlane64_b32 v1, v1
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: global_store_b32 v0, v1, s[0:1]
+; CHECK-NEXT: s_endpgm
+ %tid = call i32 @llvm.amdgcn.workitem.id.x()
+ %tid2 = add i32 %tid, 1
+ %v = call i32 @llvm.amdgcn.permlane64(i32 %tid2)
+ %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
+ store i32 %v, i32 addrspace(1)* %out_ptr
+ ret void
+}
+
+define protected amdgpu_kernel void @trivial_waterfall_eq_zero(ptr addrspace(1) %out) {
+; CHECK-LABEL: trivial_waterfall_eq_zero:
+; CHECK: ; %bb.0: ; %entry
+; CHECK-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
+; CHECK-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 5
+; CHECK-NEXT: s_mov_b32 s2, 0
+; CHECK-NEXT: s_branch .LBB7_2
+; CHECK-NEXT: .LBB7_1: ; %Flow
+; CHECK-NEXT: ; in Loop: Header=BB7_2 Depth=1
+; CHECK-NEXT: s_and_not1_b32 vcc_lo, exec_lo, s2
+; CHECK-NEXT: s_mov_b32 s2, -1
+; CHECK-NEXT: s_cbranch_vccz .LBB7_4
+; CHECK-NEXT: .LBB7_2: ; %while
+; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: s_and_b32 vcc_lo, exec_lo, s2
+; CHECK-NEXT: s_mov_b32 s2, -1
+; CHECK-NEXT: s_cbranch_vccnz .LBB7_1
+; CHECK-NEXT: ; %bb.3: ; %if
+; CHECK-NEXT: ; in Loop: Header=BB7_2 Depth=1
+; CHECK-NEXT: s_mov_b32 s2, 0
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: global_store_b32 v0, v1, s[0:1]
+; CHECK-NEXT: s_branch .LBB7_1
+; CHECK-NEXT: .LBB7_4: ; %exit
+; CHECK-NEXT: s_endpgm
+entry:
+ br label %while
+
+while:
+ %done = phi i1 [ 0, %entry ], [ 1, %if ]
+ %not_done = xor i1 %done, true
+ %ballot = tail call i64 @llvm.amdgcn.ballot.i64(i1 %not_done)
+ %is_done = icmp eq i64 %ballot, 0 ; in this case is_done = !not_done
+ br i1 %is_done, label %exit, label %if
+
+if:
+ store i32 5, ptr addrspace(1) %out
+ br label %while
+
+exit:
+ ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-uniform-waterfall.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-uniform-waterfall.ll
index 33ce278028bba..c962c05d24ad0 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-uniform-waterfall.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-uniform-waterfall.ll
@@ -1,6 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -amdgpu-enable-uniform-intrinsic-combine=0 -O3 -S < %s | FileCheck %s -check-prefix=CURRENT-CHECK
; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=amdgpu-uniform-intrinsic-combine -S < %s | FileCheck %s -check-prefix=PASS-CHECK
+; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -amdgpu-uniform-intrinsic-combine -S < %s | FileCheck %s -check-prefix=PASS-CHECK
; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -O3 -S < %s | FileCheck %s -check-prefix=O3-CHECK
define protected amdgpu_kernel void @trivial_waterfall_eq_zero(ptr addrspace(1) %out) {
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-intrinsic-combine.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-intrinsic-combine.ll
index a3e42e564376c..a7e828c95d69f 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-intrinsic-combine.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-intrinsic-combine.ll
@@ -1,5 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -amdgpu-enable-uniform-intrinsic-combine=0 -O3 -S < %s | FileCheck %s -check-prefix=CURRENT-CHECK
+; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -amdgpu-uniform-intrinsic-combine -S < %s | FileCheck %s -check-prefix=PASS-CHECK
; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=amdgpu-uniform-intrinsic-combine -S < %s | FileCheck %s -check-prefix=PASS-CHECK
; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=amdgpu-uniform-intrinsic-combine,dce -S < %s | FileCheck %s -check-prefix=DCE-C...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/165667
More information about the llvm-commits
mailing list