[llvm] [DAG] visitFREEZE - revisit frozen node after merging with unfrozen uses (PR #188206)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Apr 22 05:31:11 PDT 2026
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-x86
Author: Simon Pilgrim (RKSimon)
<details>
<summary>Changes</summary>
After merging other uses of a N to all use FREEZE(N), make sure we revisit FREEZE(N) to attempt to push the FREEZE through the (now hasOneUse()) operand.
---
Patch is 556.47 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/188206.diff
30 Files Affected:
- (modified) llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (+2)
- (modified) llvm/test/CodeGen/AMDGPU/load-constant-i16.ll (+368-370)
- (modified) llvm/test/CodeGen/AMDGPU/load-constant-i8.ll (+905-907)
- (modified) llvm/test/CodeGen/AMDGPU/load-global-i16.ll (+641-411)
- (modified) llvm/test/CodeGen/AMDGPU/load-global-i8.ll (+803-799)
- (modified) llvm/test/CodeGen/AMDGPU/load-local-i16.ll (+15-15)
- (modified) llvm/test/CodeGen/RISCV/abds.ll (+124-124)
- (modified) llvm/test/CodeGen/RISCV/abdu-neg.ll (+256-252)
- (modified) llvm/test/CodeGen/RISCV/bitint-fp-conv-200.ll (+1-2)
- (modified) llvm/test/CodeGen/RISCV/iabs.ll (+12-12)
- (modified) llvm/test/CodeGen/RISCV/idiv_large.ll (+399-421)
- (modified) llvm/test/CodeGen/RISCV/rvv/setcc-fp-vp.ll (+798-412)
- (modified) llvm/test/CodeGen/RISCV/rvv/vp-splice.ll (+39-39)
- (modified) llvm/test/CodeGen/X86/bmi-select-distrib.ll (+18-18)
- (modified) llvm/test/CodeGen/X86/insertelement-var-index.ll (+24-28)
- (modified) llvm/test/CodeGen/X86/midpoint-int-vec-256.ll (+351-351)
- (modified) llvm/test/CodeGen/X86/midpoint-int-vec-512.ll (+236-236)
- (modified) llvm/test/CodeGen/X86/vector-fshl-256.ll (+16-16)
- (modified) llvm/test/CodeGen/X86/vector-fshl-512.ll (+24-24)
- (modified) llvm/test/CodeGen/X86/vector-fshl-rot-256.ll (+3-3)
- (modified) llvm/test/CodeGen/X86/vector-fshl-rot-512.ll (+27-27)
- (modified) llvm/test/CodeGen/X86/vector-fshr-128.ll (+85-81)
- (modified) llvm/test/CodeGen/X86/vector-fshr-256.ll (+66-66)
- (modified) llvm/test/CodeGen/X86/vector-fshr-rot-128.ll (+9-10)
- (modified) llvm/test/CodeGen/X86/vector-fshr-rot-256.ll (+3-3)
- (modified) llvm/test/CodeGen/X86/vector-rotate-256.ll (+3-3)
- (modified) llvm/test/CodeGen/X86/vector-rotate-512.ll (+27-27)
- (modified) llvm/test/CodeGen/X86/vector-shift-lut.ll (+50-50)
- (modified) llvm/test/CodeGen/X86/vector-shift-shl-256.ll (+38-38)
- (modified) llvm/test/CodeGen/X86/vector-shift-shl-512.ll (+19-19)
``````````diff
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 59ba953d87055..28124ff6dff4d 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -17712,6 +17712,8 @@ SDValue DAGCombiner::visitFREEZE(SDNode *N) {
// creating a cycle in a DAG. Let's undo that by mutating the freeze.
assert(N->getOperand(0) == FrozenN0 && "Expected cycle in DAG");
DAG.UpdateNodeOperands(N, N0);
+ // Revisit the node.
+ AddToWorklist(N);
return FrozenN0;
}
diff --git a/llvm/test/CodeGen/AMDGPU/load-constant-i16.ll b/llvm/test/CodeGen/AMDGPU/load-constant-i16.ll
index 23f30ea11c49d..9d18dd541d15e 100644
--- a/llvm/test/CodeGen/AMDGPU/load-constant-i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/load-constant-i16.ll
@@ -6542,33 +6542,33 @@ define amdgpu_kernel void @constant_sextload_v8i16_to_v8i64(ptr addrspace(1) %ou
; GCN-NOHSA-SI-NEXT: s_mov_b32 s2, -1
; GCN-NOHSA-SI-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NOHSA-SI-NEXT: s_mov_b32 s8, s7
-; GCN-NOHSA-SI-NEXT: s_lshr_b32 s10, s6, 16
-; GCN-NOHSA-SI-NEXT: s_mov_b32 s12, s5
+; GCN-NOHSA-SI-NEXT: s_mov_b32 s10, s5
+; GCN-NOHSA-SI-NEXT: s_lshr_b32 s12, s6, 16
; GCN-NOHSA-SI-NEXT: s_lshr_b32 s14, s4, 16
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[16:17], s[4:5], 0x100000
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[18:19], s[6:7], 0x100000
; GCN-NOHSA-SI-NEXT: s_ashr_i32 s13, s5, 31
; GCN-NOHSA-SI-NEXT: s_ashr_i32 s15, s5, 16
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[4:5], s[12:13], 0x100000
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s12, s7, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s20, s7, 16
+; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[4:5], s[10:11], 0x100000
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s20, s7, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s21, s7, 16
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[6:7], s[8:9], 0x100000
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[8:9], s[14:15], 0x100000
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[10:11], s[10:11], 0x100000
+; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[10:11], s[12:13], 0x100000
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v0, s6
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v1, s7
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s20
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s12
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s21
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s20
; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:48
; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v0, s4
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v1, s5
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s15
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s13
+; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:16
+; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v0, s18
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v1, s19
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s4
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s5
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v4, s15
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v5, s13
-; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[2:5], off, s[0:3], 0 offset:16
-; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v4, s16
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v5, s17
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s10
@@ -6589,8 +6589,8 @@ define amdgpu_kernel void @constant_sextload_v8i16_to_v8i64(ptr addrspace(1) %ou
; GCN-HSA-NEXT: s_load_dwordx4 s[4:7], s[2:3], 0x0
; GCN-HSA-NEXT: s_waitcnt lgkmcnt(0)
; GCN-HSA-NEXT: s_mov_b32 s2, s7
-; GCN-HSA-NEXT: s_lshr_b32 s8, s6, 16
-; GCN-HSA-NEXT: s_mov_b32 s10, s5
+; GCN-HSA-NEXT: s_mov_b32 s8, s5
+; GCN-HSA-NEXT: s_lshr_b32 s10, s6, 16
; GCN-HSA-NEXT: s_lshr_b32 s12, s4, 16
; GCN-HSA-NEXT: s_ashr_i32 s13, s5, 16
; GCN-HSA-NEXT: s_bfe_i64 s[14:15], s[4:5], 0x100000
@@ -6608,25 +6608,25 @@ define amdgpu_kernel void @constant_sextload_v8i16_to_v8i64(ptr addrspace(1) %ou
; GCN-HSA-NEXT: s_addc_u32 s3, s1, 0
; GCN-HSA-NEXT: v_mov_b32_e32 v5, s3
; GCN-HSA-NEXT: v_mov_b32_e32 v4, s2
-; GCN-HSA-NEXT: s_add_u32 s2, s0, 32
+; GCN-HSA-NEXT: s_add_u32 s2, s0, 16
; GCN-HSA-NEXT: v_mov_b32_e32 v2, s7
; GCN-HSA-NEXT: v_mov_b32_e32 v3, s12
; GCN-HSA-NEXT: s_addc_u32 s3, s1, 0
; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
; GCN-HSA-NEXT: v_mov_b32_e32 v5, s3
; GCN-HSA-NEXT: v_mov_b32_e32 v4, s2
-; GCN-HSA-NEXT: s_add_u32 s2, s0, 16
-; GCN-HSA-NEXT: v_mov_b32_e32 v0, s16
-; GCN-HSA-NEXT: v_mov_b32_e32 v1, s17
-; GCN-HSA-NEXT: v_mov_b32_e32 v2, s8
-; GCN-HSA-NEXT: v_mov_b32_e32 v3, s9
+; GCN-HSA-NEXT: s_add_u32 s2, s0, 32
+; GCN-HSA-NEXT: v_mov_b32_e32 v0, s8
+; GCN-HSA-NEXT: v_mov_b32_e32 v1, s9
+; GCN-HSA-NEXT: v_mov_b32_e32 v2, s13
+; GCN-HSA-NEXT: v_mov_b32_e32 v3, s6
; GCN-HSA-NEXT: s_addc_u32 s3, s1, 0
; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
; GCN-HSA-NEXT: v_mov_b32_e32 v5, s3
-; GCN-HSA-NEXT: v_mov_b32_e32 v0, s10
-; GCN-HSA-NEXT: v_mov_b32_e32 v1, s11
-; GCN-HSA-NEXT: v_mov_b32_e32 v2, s13
-; GCN-HSA-NEXT: v_mov_b32_e32 v3, s6
+; GCN-HSA-NEXT: v_mov_b32_e32 v0, s16
+; GCN-HSA-NEXT: v_mov_b32_e32 v1, s17
+; GCN-HSA-NEXT: v_mov_b32_e32 v2, s10
+; GCN-HSA-NEXT: v_mov_b32_e32 v3, s11
; GCN-HSA-NEXT: v_mov_b32_e32 v4, s2
; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
; GCN-HSA-NEXT: v_mov_b32_e32 v5, s1
@@ -7164,12 +7164,12 @@ define amdgpu_kernel void @constant_sextload_v16i16_to_v16i64(ptr addrspace(1) %
; GCN-NOHSA-SI-NEXT: s_mov_b32 s10, -1
; GCN-NOHSA-SI-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NOHSA-SI-NEXT: s_mov_b32 s12, s7
-; GCN-NOHSA-SI-NEXT: s_lshr_b32 s14, s6, 16
-; GCN-NOHSA-SI-NEXT: s_mov_b32 s16, s5
-; GCN-NOHSA-SI-NEXT: s_lshr_b32 s18, s4, 16
-; GCN-NOHSA-SI-NEXT: s_mov_b32 s20, s3
-; GCN-NOHSA-SI-NEXT: s_lshr_b32 s22, s2, 16
-; GCN-NOHSA-SI-NEXT: s_mov_b32 s24, s1
+; GCN-NOHSA-SI-NEXT: s_mov_b32 s14, s5
+; GCN-NOHSA-SI-NEXT: s_mov_b32 s16, s3
+; GCN-NOHSA-SI-NEXT: s_mov_b32 s18, s1
+; GCN-NOHSA-SI-NEXT: s_lshr_b32 s20, s6, 16
+; GCN-NOHSA-SI-NEXT: s_lshr_b32 s22, s4, 16
+; GCN-NOHSA-SI-NEXT: s_lshr_b32 s24, s2, 16
; GCN-NOHSA-SI-NEXT: s_lshr_b32 s26, s0, 16
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[28:29], s[0:1], 0x100000
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[30:31], s[2:3], 0x100000
@@ -7177,60 +7177,60 @@ define amdgpu_kernel void @constant_sextload_v16i16_to_v16i64(ptr addrspace(1) %
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[36:37], s[6:7], 0x100000
; GCN-NOHSA-SI-NEXT: s_ashr_i32 s21, s1, 31
; GCN-NOHSA-SI-NEXT: s_ashr_i32 s23, s1, 16
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[0:1], s[24:25], 0x100000
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s24, s3, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s25, s3, 16
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[2:3], s[20:21], 0x100000
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s20, s5, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s27, s5, 16
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[4:5], s[16:17], 0x100000
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s33, s7, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s38, s7, 16
+; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[0:1], s[18:19], 0x100000
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s25, s3, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s27, s3, 16
+; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[2:3], s[16:17], 0x100000
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s33, s5, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s38, s5, 16
+; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[4:5], s[14:15], 0x100000
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s39, s7, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s40, s7, 16
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[6:7], s[12:13], 0x100000
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[12:13], s[26:27], 0x100000
+; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[14:15], s[24:25], 0x100000
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[16:17], s[22:23], 0x100000
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[18:19], s[18:19], 0x100000
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[14:15], s[14:15], 0x100000
+; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[18:19], s[20:21], 0x100000
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v0, s6
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v1, s7
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s40
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s39
+; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[8:11], 0 offset:112
+; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v0, s4
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v1, s5
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s38
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s33
-; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[8:11], 0 offset:112
+; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[8:11], 0 offset:80
+; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v0, s2
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v1, s3
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s27
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s25
+; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[8:11], 0 offset:48
+; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v0, s0
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v1, s1
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s23
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s21
+; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[8:11], 0 offset:16
; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v0, s36
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v1, s37
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s4
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s5
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v4, s27
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v5, s20
-; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[2:5], off, s[8:11], 0 offset:80
-; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v4, s34
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v5, s35
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v6, s2
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v7, s3
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v8, s25
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v9, s24
-; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[6:9], off, s[8:11], 0 offset:48
-; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v8, s30
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v9, s31
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v10, s0
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v11, s1
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v12, s23
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v13, s21
-; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[10:13], off, s[8:11], 0 offset:16
-; GCN-NOHSA-SI-NEXT: s_waitcnt expcnt(0)
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v12, s28
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v13, s29
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s14
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s15
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v2, s18
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v3, s19
; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[8:11], 0 offset:96
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v6, s18
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v7, s19
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v6, s16
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v7, s17
; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[4:7], off, s[8:11], 0 offset:64
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v10, s16
-; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v11, s17
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v10, s14
+; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v11, s15
; GCN-NOHSA-SI-NEXT: buffer_store_dwordx4 v[8:11], off, s[8:11], 0 offset:32
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v14, s12
; GCN-NOHSA-SI-NEXT: v_mov_b32_e32 v15, s13
@@ -7246,19 +7246,19 @@ define amdgpu_kernel void @constant_sextload_v16i16_to_v16i64(ptr addrspace(1) %
; GCN-HSA-NEXT: s_waitcnt lgkmcnt(0)
; GCN-HSA-NEXT: s_load_dwordx8 s[0:7], s[10:11], 0x0
; GCN-HSA-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-HSA-NEXT: s_mov_b32 s10, s7
-; GCN-HSA-NEXT: s_lshr_b32 s12, s6, 16
+; GCN-HSA-NEXT: s_mov_b32 s12, s7
; GCN-HSA-NEXT: s_mov_b32 s14, s5
-; GCN-HSA-NEXT: s_lshr_b32 s16, s4, 16
-; GCN-HSA-NEXT: s_ashr_i32 s25, s1, 31
+; GCN-HSA-NEXT: s_mov_b32 s16, s3
+; GCN-HSA-NEXT: s_mov_b32 s18, s1
+; GCN-HSA-NEXT: s_ashr_i32 s27, s1, 31
; GCN-HSA-NEXT: s_ashr_i32 s29, s3, 31
; GCN-HSA-NEXT: s_ashr_i32 s30, s3, 16
-; GCN-HSA-NEXT: s_mov_b32 s18, s3
-; GCN-HSA-NEXT: s_lshr_b32 s20, s2, 16
-; GCN-HSA-NEXT: s_mov_b32 s22, s1
-; GCN-HSA-NEXT: s_lshr_b32 s24, s0, 16
-; GCN-HSA-NEXT: s_bfe_i64 s[26:27], s[2:3], 0x100000
-; GCN-HSA-NEXT: s_bfe_i64 s[2:3], s[10:11], 0x100000
+; GCN-HSA-NEXT: s_lshr_b32 s20, s6, 16
+; GCN-HSA-NEXT: s_lshr_b32 s22, s4, 16
+; GCN-HSA-NEXT: s_lshr_b32 s24, s2, 16
+; GCN-HSA-NEXT: s_lshr_b32 s26, s0, 16
+; GCN-HSA-NEXT: s_bfe_i64 s[10:11], s[2:3], 0x100000
+; GCN-HSA-NEXT: s_bfe_i64 s[2:3], s[12:13], 0x100000
; GCN-HSA-NEXT: s_ashr_i32 s28, s1, 16
; GCN-HSA-NEXT: s_ashr_i32 s31, s5, 31
; GCN-HSA-NEXT: s_ashr_i32 s33, s5, 16
@@ -7269,36 +7269,55 @@ define amdgpu_kernel void @constant_sextload_v16i16_to_v16i64(ptr addrspace(1) %
; GCN-HSA-NEXT: s_bfe_i64 s[6:7], s[6:7], 0x100000
; GCN-HSA-NEXT: v_mov_b32_e32 v0, s2
; GCN-HSA-NEXT: v_mov_b32_e32 v1, s3
-; GCN-HSA-NEXT: s_bfe_i64 s[2:3], s[24:25], 0x100000
-; GCN-HSA-NEXT: s_bfe_i64 s[10:11], s[22:23], 0x100000
+; GCN-HSA-NEXT: s_bfe_i64 s[2:3], s[26:27], 0x100000
+; GCN-HSA-NEXT: s_bfe_i64 s[12:13], s[24:25], 0x100000
+; GCN-HSA-NEXT: s_bfe_i64 s[22:23], s[22:23], 0x100000
; GCN-HSA-NEXT: s_bfe_i64 s[20:21], s[20:21], 0x100000
; GCN-HSA-NEXT: s_bfe_i64 s[18:19], s[18:19], 0x100000
; GCN-HSA-NEXT: s_bfe_i64 s[16:17], s[16:17], 0x100000
; GCN-HSA-NEXT: s_bfe_i64 s[14:15], s[14:15], 0x100000
-; GCN-HSA-NEXT: s_bfe_i64 s[12:13], s[12:13], 0x100000
-; GCN-HSA-NEXT: s_add_u32 s22, s8, 0x70
-; GCN-HSA-NEXT: s_addc_u32 s23, s9, 0
-; GCN-HSA-NEXT: v_mov_b32_e32 v4, s6
-; GCN-HSA-NEXT: s_add_u32 s6, s8, 0x60
-; GCN-HSA-NEXT: v_mov_b32_e32 v5, s7
-; GCN-HSA-NEXT: s_addc_u32 s7, s9, 0
-; GCN-HSA-NEXT: v_mov_b32_e32 v8, s22
-; GCN-HSA-NEXT: v_mov_b32_e32 v11, s7
-; GCN-HSA-NEXT: v_mov_b32_e32 v9, s23
-; GCN-HSA-NEXT: v_mov_b32_e32 v10, s6
+; GCN-HSA-NEXT: s_add_u32 s24, s8, 0x70
+; GCN-HSA-NEXT: s_addc_u32 s25, s9, 0
+; GCN-HSA-NEXT: v_mov_b32_e32 v4, s14
+; GCN-HSA-NEXT: s_add_u32 s14, s8, 0x50
+; GCN-HSA-NEXT: v_mov_b32_e32 v8, s24
+; GCN-HSA-NEXT: v_mov_b32_e32 v5, s15
+; GCN-HSA-NEXT: s_addc_u32 s15, s9, 0
+; GCN-HSA-NEXT: v_mov_b32_e32 v10, s14
+; GCN-HSA-NEXT: v_mov_b32_e32 v9, s25
+; GCN-HSA-NEXT: v_mov_b32_e32 v11, s15
; GCN-HSA-NEXT: v_mov_b32_e32 v2, s35
; GCN-HSA-NEXT: v_mov_b32_e32 v3, s34
-; GCN-HSA-NEXT: s_add_u32 s6, s8, 0x50
-; GCN-HSA-NEXT: v_mov_b32_e32 v6, s12
-; GCN-HSA-NEXT: v_mov_b32_e32 v7, s13
+; GCN-HSA-NEXT: s_add_u32 s14, s8, 48
+; GCN-HSA-NEXT: v_mov_b32_e32 v6, s33
+; GCN-HSA-NEXT: v_mov_b32_e32 v7, s31
; GCN-HSA-NEXT: flat_store_dwordx4 v[8:9], v[0:3]
; GCN-HSA-NEXT: flat_store_dwordx4 v[10:11], v[4:7]
+; GCN-HSA-NEXT: s_addc_u32 s15, s9, 0
+; GCN-HSA-NEXT: v_mov_b32_e32 v4, s14
+; GCN-HSA-NEXT: v_mov_b32_e32 v0, s16
+; GCN-HSA-NEXT: v_mov_b32_e32 v1, s17
+; GCN-HSA-NEXT: v_mov_b32_e32 v2, s30
+; GCN-HSA-NEXT: v_mov_b32_e32 v3, s29
+; GCN-HSA-NEXT: v_mov_b32_e32 v5, s15
+; GCN-HSA-NEXT: s_add_u32 s14, s8, 16
+; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
+; GCN-HSA-NEXT: s_addc_u32 s15, s9, 0
+; GCN-HSA-NEXT: v_mov_b32_e32 v4, s14
+; GCN-HSA-NEXT: v_mov_b32_e32 v0, s18
+; GCN-HSA-NEXT: v_mov_b32_e32 v1, s19
+; GCN-HSA-NEXT: v_mov_b32_e32 v2, s28
+; GCN-HSA-NEXT: v_mov_b32_e32 v3, s27
+; GCN-HSA-NEXT: v_mov_b32_e32 v5, s15
+; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
+; GCN-HSA-NEXT: s_nop 0
+; GCN-HSA-NEXT: v_mov_b32_e32 v0, s6
+; GCN-HSA-NEXT: s_add_u32 s6, s8, 0x60
+; GCN-HSA-NEXT: v_mov_b32_e32 v1, s7
; GCN-HSA-NEXT: s_addc_u32 s7, s9, 0
; GCN-HSA-NEXT: v_mov_b32_e32 v4, s6
-; GCN-HSA-NEXT: v_mov_b32_e32 v0, s14
-; GCN-HSA-NEXT: v_mov_b32_e32 v1, s15
-; GCN-HSA-NEXT: v_mov_b32_e32 v2, s33
-; GCN-HSA-NEXT: v_mov_b32_e32 v3, s31
+; GCN-HSA-NEXT: v_mov_b32_e32 v2, s20
+; GCN-HSA-NEXT: v_mov_b32_e32 v3, s21
; GCN-HSA-NEXT: v_mov_b32_e32 v5, s7
; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
; GCN-HSA-NEXT: s_nop 0
@@ -7307,35 +7326,17 @@ define amdgpu_kernel void @constant_sextload_v16i16_to_v16i64(ptr addrspace(1) %
; GCN-HSA-NEXT: v_mov_b32_e32 v1, s5
; GCN-HSA-NEXT: s_addc_u32 s5, s9, 0
; GCN-HSA-NEXT: v_mov_b32_e32 v4, s4
-; GCN-HSA-NEXT: v_mov_b32_e32 v2, s16
-; GCN-HSA-NEXT: v_mov_b32_e32 v3, s17
-; GCN-HSA-NEXT: v_mov_b32_e32 v5, s5
-; GCN-HSA-NEXT: s_add_u32 s4, s8, 48
-; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT: s_addc_u32 s5, s9, 0
-; GCN-HSA-NEXT: v_mov_b32_e32 v4, s4
-; GCN-HSA-NEXT: v_mov_b32_e32 v0, s18
-; GCN-HSA-NEXT: v_mov_b32_e32 v1, s19
-; GCN-HSA-NEXT: v_mov_b32_e32 v2, s30
-; GCN-HSA-NEXT: v_mov_b32_e32 v3, s29
+; GCN-HSA-NEXT: v_mov_b32_e32 v2, s22
+; GCN-HSA-NEXT: v_mov_b32_e32 v3, s23
; GCN-HSA-NEXT: v_mov_b32_e32 v5, s5
; GCN-HSA-NEXT: s_add_u32 s4, s8, 32
; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
; GCN-HSA-NEXT: s_addc_u32 s5, s9, 0
; GCN-HSA-NEXT: v_mov_b32_e32 v4, s4
-; GCN-HSA-NEXT: v_mov_b32_e32 v0, s26
-; GCN-HSA-NEXT: v_mov_b32_e32 v1, s27
-; GCN-HSA-NEXT: v_mov_b32_e32 v2, s20
-; GCN-HSA-NEXT: v_mov_b32_e32 v3, s21
-; GCN-HSA-NEXT: v_mov_b32_e32 v5, s5
-; GCN-HSA-NEXT: s_add_u32 s4, s8, 16
-; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT: s_addc_u32 s5, s9, 0
-; GCN-HSA-NEXT: v_mov_b32_e32 v4, s4
; GCN-HSA-NEXT: v_mov_b32_e32 v0, s10
; GCN-HSA-NEXT: v_mov_b32_e32 v1, s11
-; GCN-HSA-NEXT: v_mov_b32_e32 v2, s28
-; GCN-HSA-NEXT: v_mov_b32_e32 v3, s25
+; GCN-HSA-NEXT: v_mov_b32_e32 v2, s12
+; GCN-HSA-NEXT: v_mov_b32_e32 v3, s13
; GCN-HSA-NEXT: v_mov_b32_e32 v5, s5
; GCN-HSA-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
; GCN-HSA-NEXT: v_mov_b32_e32 v4, s8
@@ -8309,151 +8310,148 @@ define amdgpu_kernel void @constant_sextload_v32i16_to_v32i64(ptr addrspace(1) %
; GCN-NOHSA-SI-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NOHSA-SI-NEXT: s_mov_b32 s18, s15
; GCN-NOHSA-SI-NEXT: s_mov_b32 s20, s13
-; GCN-NOHSA-SI-NEXT: s_mov_b32 s22, s11
-; GCN-NOHSA-SI-NEXT: s_mov_b32 s24, s9
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s23, s1, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s25, s1, 16
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s27, s3, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s29, s3, 16
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s31, s5, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s33, s5, 16
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s35, s7, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s37, s7, 16
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s39, s9, 31
+; GCN-NOHSA-SI-NEXT: s_mov_b32 s24, s11
+; GCN-NOHSA-SI-NEXT: s_mov_b32 s26, s9
+; GCN-NOHSA-SI-NEXT: s_mov_b32 s22, s7
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s33, s1, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s39, s1, 16
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s41, s3, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s43, s3, 16
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s23, s5, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s27, s5, 16
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s29, s7, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s31, s7, 16
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s35, s9, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s37, s9, 16
+; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[46:47], s[26:27], 0x100000
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s55, s11, 31
+; GCN-NOHSA-SI-NEXT: s_ashr_i32 s57, s11, 16
; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[48:49], s[24:25], 0x100000
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[50:51], s[22:23], 0x100000
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[52:53], s[20:21], 0x100000
-; GCN-NOHSA-SI-NEXT: s_bfe_i64 s[54:55], s[18:19], 0x100000
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s43, s9, 16
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s45, s11, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s47, s11, 16
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s41, s13, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s56, s13, 16
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s57, s15, 31
-; GCN-NOHSA-SI-NEXT: s_ashr_i32 s58, s15, 16
-; GCN-NOHSA-SI-NEXT: s_lshr_b32 s38, s14, 16
-; GCN-NOHSA-SI-NEXT: s_lshr_b32 s24, ...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/188206
More information about the llvm-commits
mailing list