[llvm] [LegalizeTypes][DAG] Use SHL(X,1) instead of ADD(X,X) for variable vector indices for extraction/insertion legalization (PR #188277)

via llvm-commits llvm-commits at lists.llvm.org
Wed Apr 22 08:17:29 PDT 2026


llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-powerpc

@llvm/pr-subscribers-backend-risc-v

Author: Simon Pilgrim (RKSimon)

<details>
<summary>Changes</summary>

Avoid ADD(X,X) as it doesn't correctly handle undef elements and helps avoid some FREEZE() fold headaches

Resurrects #<!-- -->86857

---

Patch is 37.84 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/188277.diff


16 Files Affected:

- (modified) llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp (+6-6) 
- (modified) llvm/test/CodeGen/AArch64/insertextract.ll (+8-8) 
- (modified) llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll (+71-169) 
- (modified) llvm/test/CodeGen/ARM/combine-vmovdrr.ll (+9-8) 
- (modified) llvm/test/CodeGen/ARM/pr122670-regression.ll (+2-2) 
- (modified) llvm/test/CodeGen/LoongArch/lasx/ir-instruction/extractelement.ll (+1-1) 
- (modified) llvm/test/CodeGen/LoongArch/lasx/ir-instruction/insertelement.ll (+1-1) 
- (modified) llvm/test/CodeGen/LoongArch/lsx/ir-instruction/extractelement.ll (+1-1) 
- (modified) llvm/test/CodeGen/LoongArch/lsx/ir-instruction/insertelement.ll (+1-1) 
- (modified) llvm/test/CodeGen/Mips/msa/basic_operations.ll (+8-8) 
- (modified) llvm/test/CodeGen/PowerPC/aix-vec_insert_elt.ll (+6-6) 
- (modified) llvm/test/CodeGen/PowerPC/aix32-p8-scalar_vector_conversions.ll (+14-12) 
- (modified) llvm/test/CodeGen/PowerPC/vec_insert_elt.ll (+3-3) 
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-extract.ll (+1-1) 
- (modified) llvm/test/CodeGen/X86/insertelement-var-index.ll (+20-24) 
- (modified) llvm/test/CodeGen/X86/vector-extend-inreg.ll (+8-10) 


``````````diff
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
index 26036207c4b82..0cff883f83186 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
@@ -231,8 +231,8 @@ void DAGTypeLegalizer::ExpandRes_EXTRACT_VECTOR_ELT(SDNode *N, SDValue &Lo,
 
   // Extract the elements at 2 * Idx and 2 * Idx + 1 from the new vector.
   SDValue Idx = N->getOperand(1);
-
-  Idx = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx, Idx);
+  Idx = DAG.getNode(ISD::SHL, dl, Idx.getValueType(), Idx,
+                    DAG.getShiftAmountConstant(1, Idx.getValueType(), dl));
   Lo = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, NewVT, NewVec, Idx);
 
   Idx = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx,
@@ -446,12 +446,12 @@ SDValue DAGTypeLegalizer::ExpandOp_INSERT_VECTOR_ELT(SDNode *N) {
     std::swap(Lo, Hi);
 
   SDValue Idx = N->getOperand(2);
-  Idx = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx, Idx);
+  Idx = DAG.getNode(ISD::SHL, dl, Idx.getValueType(), Idx,
+                    DAG.getShiftAmountConstant(1, Idx.getValueType(), dl));
   NewVec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, NewVecVT, NewVec, Lo, Idx);
-  Idx = DAG.getNode(ISD::ADD, dl,
-                    Idx.getValueType(), Idx,
+  Idx = DAG.getNode(ISD::ADD, dl, Idx.getValueType(), Idx,
                     DAG.getConstant(1, dl, Idx.getValueType()));
-  NewVec =  DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, NewVecVT, NewVec, Hi, Idx);
+  NewVec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, NewVecVT, NewVec, Hi, Idx);
 
   // Convert the new vector to the old vector type.
   return DAG.getNode(ISD::BITCAST, dl, VecVT, NewVec);
diff --git a/llvm/test/CodeGen/AArch64/insertextract.ll b/llvm/test/CodeGen/AArch64/insertextract.ll
index 641701beeda43..facee0829611f 100644
--- a/llvm/test/CodeGen/AArch64/insertextract.ll
+++ b/llvm/test/CodeGen/AArch64/insertextract.ll
@@ -2856,23 +2856,23 @@ define i128 @extract_v2i128_c(<2 x i128> %a, i32 %c) {
 ; CHECK-SD-NEXT:    sub sp, sp, #64
 ; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
 ; CHECK-SD-NEXT:    adds x9, x0, x0
-; CHECK-SD-NEXT:    mov w8, w4
+; CHECK-SD-NEXT:    mov w8, #1 // =0x1
+; CHECK-SD-NEXT:    // kill: def $w4 killed $w4 def $x4
 ; CHECK-SD-NEXT:    adc x10, x1, x1
 ; CHECK-SD-NEXT:    adds x11, x2, x2
 ; CHECK-SD-NEXT:    fmov d1, x9
 ; CHECK-SD-NEXT:    fmov d0, x11
-; CHECK-SD-NEXT:    adc x12, x3, x3
-; CHECK-SD-NEXT:    add x8, x8, x8
-; CHECK-SD-NEXT:    and x9, x8, #0x2
-; CHECK-SD-NEXT:    orr w8, w8, #0x1
-; CHECK-SD-NEXT:    mov x11, sp
+; CHECK-SD-NEXT:    adc x11, x3, x3
+; CHECK-SD-NEXT:    orr w8, w8, w4, lsl #1
+; CHECK-SD-NEXT:    ubfiz x9, x4, #4, #1
 ; CHECK-SD-NEXT:    mov v1.d[1], x10
 ; CHECK-SD-NEXT:    add x10, sp, #32
 ; CHECK-SD-NEXT:    and x8, x8, #0x3
-; CHECK-SD-NEXT:    mov v0.d[1], x12
+; CHECK-SD-NEXT:    mov v0.d[1], x11
+; CHECK-SD-NEXT:    mov x11, sp
 ; CHECK-SD-NEXT:    stp q1, q0, [sp]
 ; CHECK-SD-NEXT:    stp q1, q0, [sp, #32]
-; CHECK-SD-NEXT:    ldr x0, [x10, x9, lsl #3]
+; CHECK-SD-NEXT:    ldr x0, [x10, x9]
 ; CHECK-SD-NEXT:    ldr x1, [x11, x8, lsl #3]
 ; CHECK-SD-NEXT:    add sp, sp, #64
 ; CHECK-SD-NEXT:    ret
diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll
index 084b7a2d59b2f..ab6597363825e 100644
--- a/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll
+++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll
@@ -6,18 +6,15 @@
 define amdgpu_kernel void @test_bitcast_llc_v128i8_v16i8(ptr addrspace(1) %out, i32 %idx) {
 ; GFX9-LABEL: test_bitcast_llc_v128i8_v16i8:
 ; GFX9:       ; %bb.0: ; %entry
+; GFX9-NEXT:    s_lshl_b32 s0, s0, 8
 ; GFX9-NEXT:    s_load_dwordx2 s[34:35], s[4:5], 0x0
 ; GFX9-NEXT:    s_load_dword s33, s[4:5], 0x8
-; GFX9-NEXT:    s_lshl_b32 s0, s0, 8
 ; GFX9-NEXT:    s_and_b32 s1, s0, 0xff
 ; GFX9-NEXT:    s_or_b32 s0, s1, s0
 ; GFX9-NEXT:    s_and_b32 s1, s0, 0xffff
 ; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
 ; GFX9-NEXT:    s_or_b32 s0, s1, s0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_i32 s33, s33, s33
 ; GFX9-NEXT:    s_mov_b32 s1, s0
-; GFX9-NEXT:    s_lshl_b32 s33, s33, 1
 ; GFX9-NEXT:    s_mov_b32 s2, s0
 ; GFX9-NEXT:    s_mov_b32 s3, s0
 ; GFX9-NEXT:    s_mov_b32 s4, s0
@@ -48,7 +45,8 @@ define amdgpu_kernel void @test_bitcast_llc_v128i8_v16i8(ptr addrspace(1) %out,
 ; GFX9-NEXT:    s_mov_b32 s29, s0
 ; GFX9-NEXT:    s_mov_b32 s30, s0
 ; GFX9-NEXT:    s_mov_b32 s31, s0
-; GFX9-NEXT:    s_add_i32 s36, s33, 3
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshl_b32 s33, s33, 2
 ; GFX9-NEXT:    v_mov_b64_e32 v[0:1], s[0:1]
 ; GFX9-NEXT:    v_mov_b64_e32 v[2:3], s[2:3]
 ; GFX9-NEXT:    v_mov_b64_e32 v[4:5], s[4:5]
@@ -65,12 +63,9 @@ define amdgpu_kernel void @test_bitcast_llc_v128i8_v16i8(ptr addrspace(1) %out,
 ; GFX9-NEXT:    v_mov_b64_e32 v[26:27], s[26:27]
 ; GFX9-NEXT:    v_mov_b64_e32 v[28:29], s[28:29]
 ; GFX9-NEXT:    v_mov_b64_e32 v[30:31], s[30:31]
-; GFX9-NEXT:    s_set_gpr_idx_on s36, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v35, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_add_i32 s0, s33, 2
-; GFX9-NEXT:    s_set_gpr_idx_on s0, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v34, v0
+; GFX9-NEXT:    s_set_gpr_idx_on s33, gpr_idx(SRC0)
+; GFX9-NEXT:    v_mov_b32_e32 v35, v3
+; GFX9-NEXT:    v_mov_b32_e32 v34, v2
 ; GFX9-NEXT:    s_set_gpr_idx_off
 ; GFX9-NEXT:    v_mov_b32_e32 v36, 0
 ; GFX9-NEXT:    s_set_gpr_idx_on s33, gpr_idx(SRC0)
@@ -125,10 +120,11 @@ define amdgpu_kernel void @test_bitcast_llc_v128i8_v16i8(ptr addrspace(1) %out,
 ; GFX11-NEXT:    s_mov_b32 s29, s0
 ; GFX11-NEXT:    s_mov_b32 s30, s0
 ; GFX11-NEXT:    s_mov_b32 s31, s0
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_add_i32 s33, s33, s33
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    s_lshl_b32 m0, s33, 2
+; GFX11-NEXT:    v_dual_mov_b32 v30, s30 :: v_dual_mov_b32 v31, s31
 ; GFX11-NEXT:    v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s5
 ; GFX11-NEXT:    v_dual_mov_b32 v6, s6 :: v_dual_mov_b32 v7, s7
 ; GFX11-NEXT:    v_dual_mov_b32 v8, s8 :: v_dual_mov_b32 v9, s9
@@ -142,14 +138,8 @@ define amdgpu_kernel void @test_bitcast_llc_v128i8_v16i8(ptr addrspace(1) %out,
 ; GFX11-NEXT:    v_dual_mov_b32 v24, s24 :: v_dual_mov_b32 v25, s25
 ; GFX11-NEXT:    v_dual_mov_b32 v26, s26 :: v_dual_mov_b32 v27, s27
 ; GFX11-NEXT:    v_dual_mov_b32 v28, s28 :: v_dual_mov_b32 v29, s29
-; GFX11-NEXT:    v_dual_mov_b32 v30, s30 :: v_dual_mov_b32 v31, s31
-; GFX11-NEXT:    s_lshl_b32 s0, s33, 1
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_add_i32 m0, s0, 3
-; GFX11-NEXT:    v_movrels_b32_e32 v34, v0
-; GFX11-NEXT:    s_add_i32 m0, s0, 2
-; GFX11-NEXT:    v_movrels_b32_e32 v33, v0
-; GFX11-NEXT:    s_mov_b32 m0, s0
+; GFX11-NEXT:    v_movrels_b32_e32 v34, v3
+; GFX11-NEXT:    v_movrels_b32_e32 v33, v2
 ; GFX11-NEXT:    v_movrels_b32_e32 v32, v1
 ; GFX11-NEXT:    v_movrels_b32_e32 v31, v0
 ; GFX11-NEXT:    global_store_b128 v35, v[31:34], s[34:35]
@@ -198,10 +188,11 @@ define amdgpu_kernel void @test_bitcast_llc_v128i8_v16i8(ptr addrspace(1) %out,
 ; GFX12-NEXT:    s_mov_b32 s29, s0
 ; GFX12-NEXT:    s_mov_b32 s30, s0
 ; GFX12-NEXT:    s_mov_b32 s31, s0
-; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    s_add_co_i32 s33, s38, s38
 ; GFX12-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX12-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    s_lshl_b32 m0, s38, 2
+; GFX12-NEXT:    v_dual_mov_b32 v30, s30 :: v_dual_mov_b32 v31, s31
 ; GFX12-NEXT:    v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s5
 ; GFX12-NEXT:    v_dual_mov_b32 v6, s6 :: v_dual_mov_b32 v7, s7
 ; GFX12-NEXT:    v_dual_mov_b32 v8, s8 :: v_dual_mov_b32 v9, s9
@@ -215,14 +206,8 @@ define amdgpu_kernel void @test_bitcast_llc_v128i8_v16i8(ptr addrspace(1) %out,
 ; GFX12-NEXT:    v_dual_mov_b32 v24, s24 :: v_dual_mov_b32 v25, s25
 ; GFX12-NEXT:    v_dual_mov_b32 v26, s26 :: v_dual_mov_b32 v27, s27
 ; GFX12-NEXT:    v_dual_mov_b32 v28, s28 :: v_dual_mov_b32 v29, s29
-; GFX12-NEXT:    v_dual_mov_b32 v30, s30 :: v_dual_mov_b32 v31, s31
-; GFX12-NEXT:    s_lshl_b32 s0, s33, 1
-; GFX12-NEXT:    s_wait_alu depctr_sa_sdst(0)
-; GFX12-NEXT:    s_add_co_i32 m0, s0, 3
-; GFX12-NEXT:    v_movrels_b32_e32 v34, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s0, 2
-; GFX12-NEXT:    v_movrels_b32_e32 v33, v0
-; GFX12-NEXT:    s_mov_b32 m0, s0
+; GFX12-NEXT:    v_movrels_b32_e32 v34, v3
+; GFX12-NEXT:    v_movrels_b32_e32 v33, v2
 ; GFX12-NEXT:    v_movrels_b32_e32 v32, v1
 ; GFX12-NEXT:    v_movrels_b32_e32 v31, v0
 ; GFX12-NEXT:    global_store_b128 v35, v[31:34], s[36:37]
@@ -243,20 +228,13 @@ define amdgpu_kernel void @test_bitcast_llc_v64i16_v8i16(ptr addrspace(1) %out,
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_i32 s2, s2, s2
-; GFX9-NEXT:    s_lshl_b32 s2, s2, 1
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 2
 ; GFX9-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
+; GFX9-NEXT:    v_mov_b32_e32 v3, v3
+; GFX9-NEXT:    v_mov_b32_e32 v2, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v1
-; GFX9-NEXT:    s_add_i32 s3, s2, 3
 ; GFX9-NEXT:    v_mov_b32_e32 v0, v0
 ; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_set_gpr_idx_on s3, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_add_i32 s2, s2, 2
-; GFX9-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -265,38 +243,26 @@ define amdgpu_kernel void @test_bitcast_llc_v64i16_v8i16(ptr addrspace(1) %out,
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX11-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_add_i32 s2, s2, s2
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_lshl_b32 s2, s2, 1
-; GFX11-NEXT:    s_mov_b32 m0, s2
+; GFX11-NEXT:    s_lshl_b32 m0, s2, 2
+; GFX11-NEXT:    v_movrels_b32_e32 v3, v3
+; GFX11-NEXT:    v_movrels_b32_e32 v2, v2
 ; GFX11-NEXT:    v_movrels_b32_e32 v1, v1
 ; GFX11-NEXT:    v_movrels_b32_e32 v0, v0
-; GFX11-NEXT:    s_add_i32 m0, s2, 3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_movrels_b32_e32 v3, v0
-; GFX11-NEXT:    s_add_i32 m0, s2, 2
-; GFX11-NEXT:    v_mov_b32_e32 v4, 0
-; GFX11-NEXT:    v_movrels_b32_e32 v2, v0
 ; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: test_bitcast_llc_v64i16_v8i16:
 ; GFX12:       ; %bb.0: ; %entry
 ; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x0
+; GFX12-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    s_add_co_i32 s2, s2, s2
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    s_lshl_b32 s2, s2, 1
-; GFX12-NEXT:    s_mov_b32 m0, s2
+; GFX12-NEXT:    s_lshl_b32 m0, s2, 2
+; GFX12-NEXT:    v_movrels_b32_e32 v3, v3
+; GFX12-NEXT:    v_movrels_b32_e32 v2, v2
 ; GFX12-NEXT:    v_movrels_b32_e32 v1, v1
 ; GFX12-NEXT:    v_movrels_b32_e32 v0, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s2, 3
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_movrels_b32_e32 v3, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s2, 2
-; GFX12-NEXT:    v_mov_b32_e32 v4, 0
-; GFX12-NEXT:    v_movrels_b32_e32 v2, v0
 ; GFX12-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
 ; GFX12-NEXT:    s_endpgm
 entry:
@@ -315,20 +281,13 @@ define amdgpu_kernel void @test_bitcast_llc_v32i32_v4i32(ptr addrspace(1) %out,
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_i32 s2, s2, s2
-; GFX9-NEXT:    s_lshl_b32 s2, s2, 1
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 2
 ; GFX9-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
+; GFX9-NEXT:    v_mov_b32_e32 v3, v3
+; GFX9-NEXT:    v_mov_b32_e32 v2, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v1
-; GFX9-NEXT:    s_add_i32 s3, s2, 3
 ; GFX9-NEXT:    v_mov_b32_e32 v0, v0
 ; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_set_gpr_idx_on s3, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_add_i32 s2, s2, 2
-; GFX9-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -337,38 +296,26 @@ define amdgpu_kernel void @test_bitcast_llc_v32i32_v4i32(ptr addrspace(1) %out,
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX11-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_add_i32 s2, s2, s2
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_lshl_b32 s2, s2, 1
-; GFX11-NEXT:    s_mov_b32 m0, s2
+; GFX11-NEXT:    s_lshl_b32 m0, s2, 2
+; GFX11-NEXT:    v_movrels_b32_e32 v3, v3
+; GFX11-NEXT:    v_movrels_b32_e32 v2, v2
 ; GFX11-NEXT:    v_movrels_b32_e32 v1, v1
 ; GFX11-NEXT:    v_movrels_b32_e32 v0, v0
-; GFX11-NEXT:    s_add_i32 m0, s2, 3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_movrels_b32_e32 v3, v0
-; GFX11-NEXT:    s_add_i32 m0, s2, 2
-; GFX11-NEXT:    v_mov_b32_e32 v4, 0
-; GFX11-NEXT:    v_movrels_b32_e32 v2, v0
 ; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: test_bitcast_llc_v32i32_v4i32:
 ; GFX12:       ; %bb.0: ; %entry
 ; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x0
+; GFX12-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    s_add_co_i32 s2, s2, s2
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    s_lshl_b32 s2, s2, 1
-; GFX12-NEXT:    s_mov_b32 m0, s2
+; GFX12-NEXT:    s_lshl_b32 m0, s2, 2
+; GFX12-NEXT:    v_movrels_b32_e32 v3, v3
+; GFX12-NEXT:    v_movrels_b32_e32 v2, v2
 ; GFX12-NEXT:    v_movrels_b32_e32 v1, v1
 ; GFX12-NEXT:    v_movrels_b32_e32 v0, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s2, 3
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_movrels_b32_e32 v3, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s2, 2
-; GFX12-NEXT:    v_mov_b32_e32 v4, 0
-; GFX12-NEXT:    v_movrels_b32_e32 v2, v0
 ; GFX12-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
 ; GFX12-NEXT:    s_endpgm
 entry:
@@ -387,104 +334,59 @@ define amdgpu_kernel void @test_bitcast_llc_v16i64_v4i256(ptr addrspace(1) %out,
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_add_i32 s2, s2, s2
-; GFX9-NEXT:    s_add_i32 s3, s2, 1
-; GFX9-NEXT:    s_add_i32 s3, s3, s3
-; GFX9-NEXT:    s_lshl_b32 s3, s3, 1
-; GFX9-NEXT:    s_set_gpr_idx_on s3, gpr_idx(SRC0)
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX9-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
+; GFX9-NEXT:    v_mov_b32_e32 v3, v3
+; GFX9-NEXT:    v_mov_b32_e32 v2, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v1
-; GFX9-NEXT:    s_add_i32 s4, s3, 3
 ; GFX9-NEXT:    v_mov_b32_e32 v0, v0
+; GFX9-NEXT:    v_mov_b32_e32 v7, v7
+; GFX9-NEXT:    v_mov_b32_e32 v6, v6
+; GFX9-NEXT:    v_mov_b32_e32 v5, v5
+; GFX9-NEXT:    v_mov_b32_e32 v4, v4
 ; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_add_i32 s5, s3, 2
-; GFX9-NEXT:    s_set_gpr_idx_on s4, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_add_i32 s2, s2, s2
-; GFX9-NEXT:    s_set_gpr_idx_on s5, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_lshl_b32 s2, s2, 1
-; GFX9-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v5, v1
-; GFX9-NEXT:    v_mov_b32_e32 v4, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_add_i32 s3, s2, 3
-; GFX9-NEXT:    s_set_gpr_idx_on s3, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v7, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    s_add_i32 s2, s2, 2
-; GFX9-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
-; GFX9-NEXT:    v_mov_b32_e32 v6, v0
-; GFX9-NEXT:    s_set_gpr_idx_off
-; GFX9-NEXT:    global_store_dwordx4 v8, v[4:7], s[0:1]
-; GFX9-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1] offset:16
+; GFX9-NEXT:    global_store_dwordx4 v8, v[4:7], s[0:1] offset:16
+; GFX9-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: test_bitcast_llc_v16i64_v4i256:
 ; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_load_b32 s0, s[4:5], 0x8
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_add_i32 s2, s0, s0
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_add_i32 s0, s2, 1
-; GFX11-NEXT:    s_add_i32 s2, s2, s2
-; GFX11-NEXT:    s_add_i32 s0, s0, s0
-; GFX11-NEXT:    s_lshl_b32 s2, s2, 1
-; GFX11-NEXT:    s_lshl_b32 s3, s0, 1
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
-; GFX11-NEXT:    s_mov_b32 m0, s3
-; GFX11-NEXT:    v_movrels_b32_e32 v1, v1
-; GFX11-NEXT:    v_movrels_b32_e32 v0, v0
-; GFX11-NEXT:    s_add_i32 m0, s3, 3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_movrels_b32_e32 v3, v0
-; GFX11-NEXT:    s_add_i32 m0, s3, 2
-; GFX11-NEXT:    v_movrels_b32_e32 v2, v0
-; GFX11-NEXT:    s_mov_b32 m0, s2
-; GFX11-NEXT:    v_movrels_b32_e32 v5, v1
-; GFX11-NEXT:    v_movrels_b32_e32 v4, v0
-; GFX11-NEXT:    s_add_i32 m0, s2, 3
-; GFX11-NEXT:    v_movrels_b32_e32 v7, v0
-; GFX11-NEXT:    s_add_i32 m0, s2, 2
 ; GFX11-NEXT:    v_mov_b32_e32 v8, 0
-; GFX11-NEXT:    v_movrels_b32_e32 v6, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    s_lshl_b32 m0, s2, 3
+; GFX11-NEXT:    v_movrels_b32_e32 v3, v3
+; GFX11-NEXT:    v_movrels_b32_e32 v2, v2
+; GFX11-NEXT:    v_movrels_b32_e32 v1, v1
+; GFX11-NEXT:    v_movrels_b32_e32 v7, v7
+; GFX11-NEXT:    v_movrels_b32_e32 v6, v6
+; GFX11-NEXT:    v_movrels_b32_e32 v5, v5
+; GFX11-NEXT:    v_movrels_b32_e32 v4, v4
+; GFX11-NEXT:    v_movrels_b32_e32 v0, v0
 ; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[0:1]
-; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[0:1] offset:16
+; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[0:1] offset:16
+; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[0:1]
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: test_bitcast_llc_v16i64_v4i256:
 ; GFX12:       ; %bb.0: ; %entry
 ; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x0
+; GFX12-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    s_add_co_i32 s2, s2, s2
-; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
-; GFX12-NEXT:    s_add_co_i32 s3, s2, 1
-; GFX12-NEXT:    s_add_co_i32 s2, s2, s2
-; GFX12-NEXT:    s_add_co_i32 s3, s3, s3
-; GFX12-NEXT:    s_lshl_b32 s2, s2, 1
-; GFX12-NEXT:    s_lshl_b32 s3, s3, 1
-; GFX12-NEXT:    s_mov_b32 m0, s3
+; GFX12-NEXT:    s_lshl_b32 m0, s2, 3
+; GFX12-NEXT:    v_movrels_b32_e32 v3, v3
+; GFX12-NEXT:    v_movrels_b32_e32 v2, v2
 ; GFX12-NEXT:    v_movrels_b32_e32 v1, v1
+; GFX12-NEXT:    v_movrels_b32_e32 v7, v7
+; GFX12-NEXT:    v_movrels_b32_e32 v6, v6
+; GFX12-NEXT:    v_movrels_b32_e32 v5, v5
+; GFX12-NEXT:    v_movrels_b32_e32 v4, v4
 ; GFX12-NEXT:    v_movrels_b32_e32 v0, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s3, 3
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_movrels_b32_e32 v3, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s3, 2
-; GFX12-NEXT:    v_movrels_b32_e32 v2, v0
-; GFX12-NEXT:    s_mov_b32 m0, s2
-; GFX12-NEXT:    v_movrels_b32_e32 v5, v1
-; GFX12-NEXT:    v_movrels_b32_e32 v4, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s2, 3
-; GFX12-NEXT:    v_movrels_b32_e32 v7, v0
-; GFX12-NEXT:    s_add_co_i32 m0, s2, 2
-; GFX12-NEXT:    v_mov_b32_e32 v8, 0
-; GFX...
[truncated]

``````````

</details>


https://github.com/llvm/llvm-project/pull/188277


More information about the llvm-commits mailing list