[llvm] [AMDGPU] Remove s_delay_alu for VALU->SGPR->SALU (PR #127212)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Feb 14 05:54:56 PST 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: None (mihajlovicana)
<details>
<summary>Changes</summary>
---
Patch is 196.78 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/127212.diff
51 Files Affected:
- (modified) llvm/lib/Target/AMDGPU/AMDGPUInsertDelayAlu.cpp (+4-1)
- (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll (-12)
- (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll (+4-16)
- (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll (+21-49)
- (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_raw_buffer.ll (-8)
- (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_struct_buffer.ll (-8)
- (modified) llvm/test/CodeGen/AMDGPU/bf16.ll (+3-3)
- (modified) llvm/test/CodeGen/AMDGPU/branch-relaxation.ll (-1)
- (modified) llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fadd.ll (+11-21)
- (modified) llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll (+11-23)
- (modified) llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmin.ll (+11-23)
- (modified) llvm/test/CodeGen/AMDGPU/carryout-selection.ll (+5-8)
- (modified) llvm/test/CodeGen/AMDGPU/combine-add-zext-xor.ll (+8-8)
- (modified) llvm/test/CodeGen/AMDGPU/dynamic_stackalloc.ll (+2-37)
- (modified) llvm/test/CodeGen/AMDGPU/expand-scalar-carry-out-select-user.ll (+1-1)
- (modified) llvm/test/CodeGen/AMDGPU/fcopysign.f16.ll (+3-4)
- (modified) llvm/test/CodeGen/AMDGPU/flat_atomics_i64.ll (+8-8)
- (modified) llvm/test/CodeGen/AMDGPU/fma.f16.ll (+4-7)
- (modified) llvm/test/CodeGen/AMDGPU/fneg-combines.f16.ll (+12-15)
- (modified) llvm/test/CodeGen/AMDGPU/fp-classify.ll (+4-4)
- (modified) llvm/test/CodeGen/AMDGPU/fptrunc.ll (+3-4)
- (modified) llvm/test/CodeGen/AMDGPU/idiv-licm.ll (+5-11)
- (modified) llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll (+6-6)
- (modified) llvm/test/CodeGen/AMDGPU/insert_waitcnt_for_precise_memory.ll (+4-6)
- (modified) llvm/test/CodeGen/AMDGPU/issue92561-restore-undef-scc-verifier-error.ll (+4-8)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.exp.row.ll (+2-2)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.raw.ptr.buffer.atomic.fadd.v2bf16.ll (-1)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.reduce.umax.ll (+4-8)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.reduce.umin.ll (+4-8)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.s.sleep.var.ll (-1)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.s.ttracedata.ll (-1)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.buffer.load.format.v3f16.ll (+2-2)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.ptr.buffer.atomic.fadd.v2bf16.ll (-2)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.ptr.buffer.atomic.fadd_nortn.ll (-2)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.ptr.buffer.atomic.fadd_rtn.ll (-2)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32.ll (+2-5)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.ptr.buffer.atomic.fmin.f32.ll (+2-5)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.ptr.buffer.load.format.v3f16.ll (+1-1)
- (modified) llvm/test/CodeGen/AMDGPU/load-constant-i1.ll (+5-7)
- (modified) llvm/test/CodeGen/AMDGPU/min.ll (-4)
- (modified) llvm/test/CodeGen/AMDGPU/mubuf-legalize-operands-non-ptr-intrinsics.ll (+10-10)
- (modified) llvm/test/CodeGen/AMDGPU/mubuf-legalize-operands.ll (+10-10)
- (modified) llvm/test/CodeGen/AMDGPU/no-dup-inst-prefetch.ll (+1-2)
- (modified) llvm/test/CodeGen/AMDGPU/pseudo-scalar-transcendental.ll (+1-1)
- (modified) llvm/test/CodeGen/AMDGPU/saddo.ll (+4-5)
- (modified) llvm/test/CodeGen/AMDGPU/sitofp.f16.ll (+2-4)
- (modified) llvm/test/CodeGen/AMDGPU/skip-if-dead.ll (-2)
- (modified) llvm/test/CodeGen/AMDGPU/uitofp.f16.ll (+2-4)
- (modified) llvm/test/CodeGen/AMDGPU/v_cmp_gfx11.ll (+2-3)
- (modified) llvm/test/CodeGen/AMDGPU/v_cndmask.ll (+2-4)
- (modified) llvm/test/CodeGen/AMDGPU/vgpr-descriptor-waterfall-loop-idom-update.ll (+1-1)
``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInsertDelayAlu.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInsertDelayAlu.cpp
index 3f2bb5df8836b..7eb608fc93e63 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInsertDelayAlu.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInsertDelayAlu.cpp
@@ -371,7 +371,10 @@ class AMDGPUInsertDelayAlu : public MachineFunctionPass {
for (MCRegUnit Unit : TRI->regunits(Op.getReg())) {
auto It = State.find(Unit);
if (It != State.end()) {
- Delay.merge(It->second);
+ if (!(SII->isSALU(MI.getOpcode())) ||
+ !AMDGPU::isSGPR(Op.getReg(), TRI) ||
+ It->second.VALUCycles == 0)
+ Delay.merge(It->second);
State.erase(Unit);
}
}
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll
index cd405fabf002d..4b68f8a4bd194 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll
@@ -777,7 +777,6 @@ define amdgpu_kernel void @add_i32_varying_vdata(ptr addrspace(1) %out, ptr addr
; GFX11W64-NEXT: s_lshl_b64 s[6:7], 1, s3
; GFX11W64-NEXT: v_writelane_b32 v0, s2, s3
; GFX11W64-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[6:7]
-; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT: s_add_i32 s2, s2, s8
; GFX11W64-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX11W64-NEXT: s_cbranch_scc1 .LBB2_1
@@ -822,7 +821,6 @@ define amdgpu_kernel void @add_i32_varying_vdata(ptr addrspace(1) %out, ptr addr
; GFX11W32-NEXT: s_lshl_b32 s6, 1, s2
; GFX11W32-NEXT: v_writelane_b32 v0, s0, s2
; GFX11W32-NEXT: s_and_not1_b32 s1, s1, s6
-; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT: s_add_i32 s0, s0, s3
; GFX11W32-NEXT: s_cmp_lg_u32 s1, 0
; GFX11W32-NEXT: s_cbranch_scc1 .LBB2_1
@@ -864,7 +862,6 @@ define amdgpu_kernel void @add_i32_varying_vdata(ptr addrspace(1) %out, ptr addr
; GFX12W64-NEXT: s_lshl_b64 s[6:7], 1, s3
; GFX12W64-NEXT: v_writelane_b32 v0, s2, s3
; GFX12W64-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[6:7]
-; GFX12W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12W64-NEXT: s_add_co_i32 s2, s2, s8
; GFX12W64-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX12W64-NEXT: s_cbranch_scc1 .LBB2_1
@@ -910,7 +907,6 @@ define amdgpu_kernel void @add_i32_varying_vdata(ptr addrspace(1) %out, ptr addr
; GFX12W32-NEXT: s_lshl_b32 s6, 1, s2
; GFX12W32-NEXT: v_writelane_b32 v0, s0, s2
; GFX12W32-NEXT: s_and_not1_b32 s1, s1, s6
-; GFX12W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12W32-NEXT: s_add_co_i32 s0, s0, s3
; GFX12W32-NEXT: s_wait_alu 0xfffe
; GFX12W32-NEXT: s_cmp_lg_u32 s1, 0
@@ -1178,7 +1174,6 @@ define amdgpu_kernel void @struct_add_i32_varying_vdata(ptr addrspace(1) %out, p
; GFX11W64-NEXT: s_lshl_b64 s[6:7], 1, s3
; GFX11W64-NEXT: v_writelane_b32 v0, s2, s3
; GFX11W64-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[6:7]
-; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT: s_add_i32 s2, s2, s8
; GFX11W64-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX11W64-NEXT: s_cbranch_scc1 .LBB3_1
@@ -1226,7 +1221,6 @@ define amdgpu_kernel void @struct_add_i32_varying_vdata(ptr addrspace(1) %out, p
; GFX11W32-NEXT: s_lshl_b32 s6, 1, s2
; GFX11W32-NEXT: v_writelane_b32 v0, s0, s2
; GFX11W32-NEXT: s_and_not1_b32 s1, s1, s6
-; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT: s_add_i32 s0, s0, s3
; GFX11W32-NEXT: s_cmp_lg_u32 s1, 0
; GFX11W32-NEXT: s_cbranch_scc1 .LBB3_1
@@ -1270,7 +1264,6 @@ define amdgpu_kernel void @struct_add_i32_varying_vdata(ptr addrspace(1) %out, p
; GFX12W64-NEXT: s_lshl_b64 s[6:7], 1, s3
; GFX12W64-NEXT: v_writelane_b32 v0, s2, s3
; GFX12W64-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[6:7]
-; GFX12W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12W64-NEXT: s_add_co_i32 s2, s2, s8
; GFX12W64-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX12W64-NEXT: s_cbranch_scc1 .LBB3_1
@@ -1319,7 +1312,6 @@ define amdgpu_kernel void @struct_add_i32_varying_vdata(ptr addrspace(1) %out, p
; GFX12W32-NEXT: s_lshl_b32 s6, 1, s2
; GFX12W32-NEXT: v_writelane_b32 v0, s0, s2
; GFX12W32-NEXT: s_and_not1_b32 s1, s1, s6
-; GFX12W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12W32-NEXT: s_add_co_i32 s0, s0, s3
; GFX12W32-NEXT: s_wait_alu 0xfffe
; GFX12W32-NEXT: s_cmp_lg_u32 s1, 0
@@ -2246,7 +2238,6 @@ define amdgpu_kernel void @sub_i32_varying_vdata(ptr addrspace(1) %out, ptr addr
; GFX11W64-NEXT: s_lshl_b64 s[6:7], 1, s3
; GFX11W64-NEXT: v_writelane_b32 v0, s2, s3
; GFX11W64-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[6:7]
-; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT: s_add_i32 s2, s2, s8
; GFX11W64-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX11W64-NEXT: s_cbranch_scc1 .LBB7_1
@@ -2291,7 +2282,6 @@ define amdgpu_kernel void @sub_i32_varying_vdata(ptr addrspace(1) %out, ptr addr
; GFX11W32-NEXT: s_lshl_b32 s6, 1, s2
; GFX11W32-NEXT: v_writelane_b32 v0, s0, s2
; GFX11W32-NEXT: s_and_not1_b32 s1, s1, s6
-; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT: s_add_i32 s0, s0, s3
; GFX11W32-NEXT: s_cmp_lg_u32 s1, 0
; GFX11W32-NEXT: s_cbranch_scc1 .LBB7_1
@@ -2334,7 +2324,6 @@ define amdgpu_kernel void @sub_i32_varying_vdata(ptr addrspace(1) %out, ptr addr
; GFX12W64-NEXT: s_lshl_b64 s[6:7], 1, s3
; GFX12W64-NEXT: v_writelane_b32 v0, s2, s3
; GFX12W64-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[6:7]
-; GFX12W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12W64-NEXT: s_add_co_i32 s2, s2, s8
; GFX12W64-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX12W64-NEXT: s_cbranch_scc1 .LBB7_1
@@ -2380,7 +2369,6 @@ define amdgpu_kernel void @sub_i32_varying_vdata(ptr addrspace(1) %out, ptr addr
; GFX12W32-NEXT: s_lshl_b32 s6, 1, s2
; GFX12W32-NEXT: v_writelane_b32 v0, s0, s2
; GFX12W32-NEXT: s_and_not1_b32 s1, s1, s6
-; GFX12W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12W32-NEXT: s_add_co_i32 s0, s0, s3
; GFX12W32-NEXT: s_wait_alu 0xfffe
; GFX12W32-NEXT: s_cmp_lg_u32 s1, 0
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
index 8bb8ecb079a34..2bcce6c04c0bb 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
@@ -899,7 +899,6 @@ define amdgpu_kernel void @add_i32_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[2:3], 1, s7
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s7
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[2:3]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1164_ITERATIVE-NEXT: s_add_i32 s6, s6, s8
; GFX1164_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX1164_ITERATIVE-NEXT: s_cbranch_scc1 .LBB2_1
@@ -950,7 +949,6 @@ define amdgpu_kernel void @add_i32_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1132_ITERATIVE-NEXT: s_lshl_b32 s3, 1, s1
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s1
; GFX1132_ITERATIVE-NEXT: s_and_not1_b32 s0, s0, s3
-; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1132_ITERATIVE-NEXT: s_add_i32 s6, s6, s2
; GFX1132_ITERATIVE-NEXT: s_cmp_lg_u32 s0, 0
; GFX1132_ITERATIVE-NEXT: s_cbranch_scc1 .LBB2_1
@@ -999,7 +997,6 @@ define amdgpu_kernel void @add_i32_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1264_ITERATIVE-NEXT: s_lshl_b64 s[2:3], 1, s7
; GFX1264_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s7
; GFX1264_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[2:3]
-; GFX1264_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1264_ITERATIVE-NEXT: s_add_co_i32 s6, s6, s8
; GFX1264_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX1264_ITERATIVE-NEXT: s_cbranch_scc1 .LBB2_1
@@ -1049,7 +1046,6 @@ define amdgpu_kernel void @add_i32_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1232_ITERATIVE-NEXT: s_lshl_b32 s3, 1, s1
; GFX1232_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s1
; GFX1232_ITERATIVE-NEXT: s_and_not1_b32 s0, s0, s3
-; GFX1232_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1232_ITERATIVE-NEXT: s_add_co_i32 s6, s6, s2
; GFX1232_ITERATIVE-NEXT: s_wait_alu 0xfffe
; GFX1232_ITERATIVE-NEXT: s_cmp_lg_u32 s0, 0
@@ -2576,17 +2572,16 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1164_ITERATIVE-NEXT: .LBB5_1: ; %ComputeLoop
; GFX1164_ITERATIVE-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1164_ITERATIVE-NEXT: s_ctz_i32_b64 s2, s[0:1]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_4)
+; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s3, v2, s2
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s8, v3, s2
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s2
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v1, s7, s2
; GFX1164_ITERATIVE-NEXT: s_add_u32 s6, s6, s3
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_addc_u32 s7, s7, s8
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[2:3], 1, s2
+; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[2:3]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX1164_ITERATIVE-NEXT: s_cbranch_scc1 .LBB5_1
; GFX1164_ITERATIVE-NEXT: ; %bb.2: ; %ComputeEnd
@@ -2639,7 +2634,6 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1132_ITERATIVE-NEXT: v_readlane_b32 s3, v3, s1
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s1
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v1, s7, s1
-; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX1132_ITERATIVE-NEXT: s_add_u32 s6, s6, s2
; GFX1132_ITERATIVE-NEXT: s_addc_u32 s7, s7, s3
; GFX1132_ITERATIVE-NEXT: s_lshl_b32 s1, 1, s1
@@ -4454,7 +4448,6 @@ define amdgpu_kernel void @sub_i32_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[2:3], 1, s7
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s7
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[2:3]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1164_ITERATIVE-NEXT: s_add_i32 s6, s6, s8
; GFX1164_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX1164_ITERATIVE-NEXT: s_cbranch_scc1 .LBB8_1
@@ -4505,7 +4498,6 @@ define amdgpu_kernel void @sub_i32_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1132_ITERATIVE-NEXT: s_lshl_b32 s3, 1, s1
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s1
; GFX1132_ITERATIVE-NEXT: s_and_not1_b32 s0, s0, s3
-; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1132_ITERATIVE-NEXT: s_add_i32 s6, s6, s2
; GFX1132_ITERATIVE-NEXT: s_cmp_lg_u32 s0, 0
; GFX1132_ITERATIVE-NEXT: s_cbranch_scc1 .LBB8_1
@@ -4554,7 +4546,6 @@ define amdgpu_kernel void @sub_i32_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1264_ITERATIVE-NEXT: s_lshl_b64 s[2:3], 1, s7
; GFX1264_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s7
; GFX1264_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[2:3]
-; GFX1264_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1264_ITERATIVE-NEXT: s_add_co_i32 s6, s6, s8
; GFX1264_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX1264_ITERATIVE-NEXT: s_cbranch_scc1 .LBB8_1
@@ -4604,7 +4595,6 @@ define amdgpu_kernel void @sub_i32_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1232_ITERATIVE-NEXT: s_lshl_b32 s3, 1, s1
; GFX1232_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s1
; GFX1232_ITERATIVE-NEXT: s_and_not1_b32 s0, s0, s3
-; GFX1232_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1232_ITERATIVE-NEXT: s_add_co_i32 s6, s6, s2
; GFX1232_ITERATIVE-NEXT: s_wait_alu 0xfffe
; GFX1232_ITERATIVE-NEXT: s_cmp_lg_u32 s0, 0
@@ -6164,17 +6154,16 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1164_ITERATIVE-NEXT: .LBB11_1: ; %ComputeLoop
; GFX1164_ITERATIVE-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1164_ITERATIVE-NEXT: s_ctz_i32_b64 s2, s[0:1]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_4)
+; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s3, v2, s2
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s8, v3, s2
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s2
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v1, s7, s2
; GFX1164_ITERATIVE-NEXT: s_add_u32 s6, s6, s3
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_addc_u32 s7, s7, s8
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[2:3], 1, s2
+; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[2:3]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX1164_ITERATIVE-NEXT: s_cbranch_scc1 .LBB11_1
; GFX1164_ITERATIVE-NEXT: ; %bb.2: ; %ComputeEnd
@@ -6227,7 +6216,6 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out, ptr addrspace(
; GFX1132_ITERATIVE-NEXT: v_readlane_b32 s3, v3, s1
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v0, s6, s1
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v1, s7, s1
-; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX1132_ITERATIVE-NEXT: s_add_u32 s6, s6, s2
; GFX1132_ITERATIVE-NEXT: s_addc_u32 s7, s7, s3
; GFX1132_ITERATIVE-NEXT: s_lshl_b32 s1, 1, s1
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll
index 3c0646c46efd0..eb5353e928682 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll
@@ -669,7 +669,6 @@ define amdgpu_kernel void @add_i32_varying(ptr addrspace(1) %out) {
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[6:7], 1, s3
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v0, s2, s3
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[6:7]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1164_ITERATIVE-NEXT: s_add_i32 s2, s2, s8
; GFX1164_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX1164_ITERATIVE-NEXT: s_cbranch_scc1 .LBB2_1
@@ -715,7 +714,6 @@ define amdgpu_kernel void @add_i32_varying(ptr addrspace(1) %out) {
; GFX1132_ITERATIVE-NEXT: s_lshl_b32 s6, 1, s2
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v0, s0, s2
; GFX1132_ITERATIVE-NEXT: s_and_not1_b32 s1, s1, s6
-; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1132_ITERATIVE-NEXT: s_add_i32 s0, s0, s3
; GFX1132_ITERATIVE-NEXT: s_cmp_lg_u32 s1, 0
; GFX1132_ITERATIVE-NEXT: s_cbranch_scc1 .LBB2_1
@@ -1215,7 +1213,7 @@ define amdgpu_kernel void @add_i32_varying_nouse() {
; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s6, v0, s3
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[4:5], 1, s3
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[4:5]
; GFX1164_ITERATIVE-NEXT: s_add_i32 s2, s2, s6
; GFX1164_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
@@ -1248,7 +1246,7 @@ define amdgpu_kernel void @add_i32_varying_nouse() {
; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX1132_ITERATIVE-NEXT: v_readlane_b32 s3, v0, s2
; GFX1132_ITERATIVE-NEXT: s_lshl_b32 s2, 1, s2
-; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1132_ITERATIVE-NEXT: s_and_not1_b32 s1, s1, s2
; GFX1132_ITERATIVE-NEXT: s_add_i32 s0, s0, s3
; GFX1132_ITERATIVE-NEXT: s_cmp_lg_u32 s1, 0
@@ -2217,17 +2215,16 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out) {
; GFX1164_ITERATIVE-NEXT: .LBB6_1: ; %ComputeLoop
; GFX1164_ITERATIVE-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1164_ITERATIVE-NEXT: s_ctz_i32_b64 s6, s[2:3]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_4)
+; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s7, v2, s6
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s8, v3, s6
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v0, s0, s6
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v1, s1, s6
; GFX1164_ITERATIVE-NEXT: s_add_u32 s0, s0, s7
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_addc_u32 s1, s1, s8
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[6:7], 1, s6
+; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[2:3], s[2:3], s[6:7]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_cmp_lg_u64 s[2:3], 0
; GFX1164_ITERATIVE-NEXT: s_cbranch_scc1 .LBB6_1
; GFX1164_ITERATIVE-NEXT: ; %bb.2: ; %ComputeEnd
@@ -2275,7 +2272,6 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out) {
; GFX1132_ITERATIVE-NEXT: v_readlane_b32 s7, v3, s3
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v0, s0, s3
; GFX1132_ITERATIVE-NEXT: v_writelane_b32 v1, s1, s3
-; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX1132_ITERATIVE-NEXT: s_add_u32 s0, s0, s6
; GFX1132_ITERATIVE-NEXT: s_addc_u32 s1, s1, s7
; GFX1132_ITERATIVE-NEXT: s_lshl_b32 s3, 1, s3
@@ -3019,11 +3015,10 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
; GFX1164_ITERATIVE-NEXT: .LBB7_1: ; %ComputeLoop
; GFX1164_ITERATIVE-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1164_ITERATIVE-NEXT: s_ctz_i32_b64 s4, s[2:3]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s5, v0, s4
; GFX1164_ITERATIVE-NEXT: v_readlane_b32 s6, v1, s4
; GFX1164_ITERATIVE-NEXT: s_add_u32 s0, s0, s5
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX1164_ITERATIVE-NEXT: s_addc_u32 s1, s1, s6
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[4:5], 1, s4
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[2:3], s[2:3], s[4:5]
@@ -3059,7 +3054,6 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX1132_ITERATIVE-NEXT: v_readlane_b32 s4, v0, s3
; GFX1132_ITERATIVE-NEXT: v_readlane_b32 s5, v1, s3
-; GFX1132_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1132_ITERATIVE-NEXT: s_add_u32 s0, s0, s4
; GFX1132_ITERATIVE-NEXT: s_addc_u32 s1, s1, s5
; GFX1132_ITERATIVE-NEXT: s_lshl_b32 s3, 1, s3
@@ -4091,7 +4085,6 @@ define amdgpu_kernel void @sub_i32_varying(ptr addrspace(1) %out) {
; GFX1164_ITERATIVE-NEXT: s_lshl_b64 s[6:7], 1, s3
; GFX1164_ITERATIVE-NEXT: v_writelane_b32 v0, s2, s3
; GFX1164_ITERATIVE-NEXT: s_and_not1_b64 s[0:1], s[0:1], s[6:7]
-; GFX1164_ITERATIVE-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1164_ITERATIVE-NEXT: s_add_i32 s2, s2, s8
; GFX1164_ITERATIVE-NEXT: s_cmp_lg_u64 s[0:1], 0
; GFX1164_ITERATIVE-NEXT: s_cbranch_scc1 .LBB10_1
@@ -4137,7 +4130,6 @@ define amdg...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/127212
More information about the llvm-commits
mailing list