[AMDGPU] Unused sdst writing to null #133229
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
|
@llvm/pr-subscribers-llvm-globalisel Author: Ana Mihajlovic (mihajlovicana) ChangesUnused sdst writing to null to avoid a false VALU->SALU dependency stall. This requires using the VOP3 encoding. Patch is 819.85 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/133229.diff 68 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
index f03cde455f295..06d5e48cdab73 100644
--- a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
+++ b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
@@ -973,6 +973,11 @@ bool SIShrinkInstructions::run(MachineFunction &MF) {
continue;
}
+ if (TII->isVOP3(MI.getOpcode()) &&
+ TII->hasVALU32BitEncoding(MI.getOpcode())) {
+ tryReplaceDeadSDST(MI);
+ }
+
if (!TII->hasVALU32BitEncoding(MI.getOpcode())) {
// If there is no chance we will shrink it and use VCC as sdst to get
// a 32 bit form try to replace dead sdst with NULL.
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/addsubu64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/addsubu64.ll
index 6a04dd492fcea..425dd8acd4736 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/addsubu64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/addsubu64.ll
@@ -39,7 +39,8 @@ define amdgpu_ps void @v_add_u64(ptr addrspace(1) %out, i64 %a, i64 %b) {
; GCN-LABEL: v_add_u64:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4
-; GCN-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
+; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GCN-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v5, vcc_lo
; GCN-NEXT: global_store_b64 v[0:1], v[2:3], off
; GCN-NEXT: s_endpgm
entry:
@@ -85,7 +86,8 @@ define amdgpu_ps void @v_sub_u64(ptr addrspace(1) %out, i64 %a, i64 %b) {
; GCN-LABEL: v_sub_u64:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: v_sub_co_u32 v2, vcc_lo, v2, v4
-; GCN-NEXT: v_sub_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
+; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GCN-NEXT: v_sub_co_ci_u32_e64 v3, null, v3, v5, vcc_lo
; GCN-NEXT: global_store_b64 v[0:1], v[2:3], off
; GCN-NEXT: s_endpgm
entry:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_udec_wrap.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_udec_wrap.ll
index dce4048a4b87e..d9be677a0e58d 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_udec_wrap.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_udec_wrap.ll
@@ -1440,16 +1440,16 @@ define amdgpu_kernel void @flat_atomic_dec_ret_i32_offset_addr64(ptr %out, ptr %
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_dec_u32 v3, v[0:1], v3 offset:20 glc
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_gl1_inv
; GFX11-NEXT: buffer_gl0_inv
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_store_b32 v[0:1], v3
; GFX11-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -1539,8 +1539,8 @@ define amdgpu_kernel void @flat_atomic_dec_noret_i32_offset_addr64(ptr %ptr) #1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: v_mov_b32_e32 v2, 42
; GFX11-NEXT: flat_atomic_dec_u32 v[0:1], v2 offset:20
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
@@ -2090,16 +2090,16 @@ define amdgpu_kernel void @flat_atomic_dec_ret_i64_offset_addr64(ptr %out, ptr %
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_dec_u64 v[0:1], v[0:1], v[2:3] offset:40 glc
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_gl1_inv
; GFX11-NEXT: buffer_gl0_inv
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4
-; GFX11-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
+; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: flat_store_b64 v[2:3], v[0:1]
; GFX11-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -2194,8 +2194,8 @@ define amdgpu_kernel void @flat_atomic_dec_noret_i64_offset_addr64(ptr %ptr) #1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_dec_u64 v[0:1], v[2:3] offset:40
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll
index e2d179a77f76c..92a7de9aaefd2 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll
@@ -2824,16 +2824,16 @@ define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset_addr64(ptr %out, ptr %
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_inc_u32 v3, v[0:1], v3 offset:20 glc
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_gl1_inv
; GFX11-NEXT: buffer_gl0_inv
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_store_b32 v[0:1], v3
; GFX11-NEXT: s_endpgm
;
@@ -2846,15 +2846,15 @@ define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset_addr64(ptr %out, ptr %
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_1)
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1)
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: flat_atomic_inc_u32 v3, v[0:1], v3 offset:20 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: global_inv scope:SCOPE_DEV
; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: flat_store_b32 v[0:1], v3
; GFX12-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -2944,8 +2944,8 @@ define amdgpu_kernel void @flat_atomic_inc_noret_i32_offset_addr64(ptr %ptr) #1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: v_mov_b32_e32 v2, 42
; GFX11-NEXT: flat_atomic_inc_u32 v[0:1], v2 offset:20
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
@@ -2963,8 +2963,8 @@ define amdgpu_kernel void @flat_atomic_inc_noret_i32_offset_addr64(ptr %ptr) #1
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: v_mov_b32_e32 v2, 42
; GFX12-NEXT: flat_atomic_inc_u32 v[0:1], v2 offset:20 scope:SCOPE_DEV
; GFX12-NEXT: s_wait_storecnt_dscnt 0x0
@@ -3810,16 +3810,16 @@ define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset_addr64(ptr %out, ptr %
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_inc_u64 v[0:1], v[0:1], v[2:3] offset:40 glc
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_gl1_inv
; GFX11-NEXT: buffer_gl0_inv
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4
-; GFX11-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
+; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: flat_store_b64 v[2:3], v[0:1]
; GFX11-NEXT: s_endpgm
;
@@ -3833,15 +3833,15 @@ define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset_addr64(ptr %out, ptr %
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_1)
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1)
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: flat_atomic_inc_u64 v[0:1], v[0:1], v[2:3] offset:40 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: global_inv scope:SCOPE_DEV
; GFX12-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
; GFX12-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX12-NEXT: flat_store_b64 v[2:3], v[0:1]
; GFX12-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -3936,8 +3936,8 @@ define amdgpu_kernel void @flat_atomic_inc_noret_i64_offset_addr64(ptr %ptr) #1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_inc_u64 v[0:1], v[2:3] offset:40
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
@@ -3955,8 +3955,8 @@ define amdgpu_kernel void @flat_atomic_inc_noret_i64_offset_addr64(ptr %ptr) #1
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: flat_atomic_inc_u64 v[0:1], v[2:3] offset:40 scope:SCOPE_DEV
; GFX12-NEXT: s_wait_storecnt_dscnt 0x0
; GFX12-NEXT: global_inv scope:SCOPE_DEV
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
index a948446aceff1..573017f7a948d 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
@@ -29,10 +29,9 @@ define i32 @v_extract_v64i32_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GFX12-NEXT: v_lshlrev_b32_e32 v2, 2, v2
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: global_load_b32 v0, v[0:1], off
; GFX12-NEXT: s_wait_loadcnt 0x0
-; GFX12-NEXT: s_wait_alu 0xfffd
; GFX12-NEXT: s_setpc_b64 s[30:31]
%vec = load <64 x i32>, ptr addrspace(1) %ptr
%elt = extractelement <64 x i32> %vec, i32 %idx
@@ -63,10 +62,9 @@ define i16 @v_extract_v128i16_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GFX12-NEXT: v_lshlrev_b32_e32 v2, 1, v2
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: global_load_u16 v0, v[0:1], off
; GFX12-NEXT: s_wait_loadcnt 0x0
-; GFX12-NEXT: s_wait_alu 0xfffd
; GFX12-NEXT: s_setpc_b64 s[30:31]
%vec = load <128 x i16>, ptr addrspace(1) %ptr
%elt = extractelement <128 x i16> %vec, i32 %idx
@@ -97,10 +95,9 @@ define i64 @v_extract_v32i64_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GFX12-NEXT: v_lshlrev_b32_e32 v2, 3, v2
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: global_load_b64 v[0:1], v[0:1], off
; GFX12-NEXT: s_wait_loadcnt 0x0
-; GFX12-NEXT: s_wait_alu 0xfffd
; GFX12-NEXT: s_setpc_b64 s[30:31]
%vec = load <32 x i64>, ptr addrspace(1) %ptr
%elt = extractelement <32 x i64> %vec, i32 %idx
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll
index e1ce9ea14a2a9..c424738dceb0e 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll
@@ -126,8 +126,8 @@ define amdgpu_ps i128 @extractelement_vgpr_v4i128_sgpr_idx(ptr addrspace(1) %ptr
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_readfirstlane_b32 s0, v0
@@ -195,7 +195,8 @@ define i128 @extractelement_vgpr_v4i128_vgpr_idx(ptr addrspace(1) %ptr, i32 %idx
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v2, 4, v2
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll
index 021f609053a0f..63c3146ebeedd 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll
@@ -133,8 +133,8 @@ define amdgpu_ps i16 @extractelement_vgpr_v4i16_sgpr_idx(ptr addrspace(1) %ptr,
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
; GFX11-NEXT: global_load_u16 v0, v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_readfirstlane_b32 s0, v0
@@ -199,7 +199,8 @@ define i16 @extractelement_vgpr_v4i16_vgpr_idx(ptr addrspace(1) %ptr, i32 %idx)
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v2, 1, v2
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: global_load_u16 v0, v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
@@ -776,8 +777,8 @@ define amdgpu_ps i16 @extractelement_vgpr_v8i16_sgpr_idx(ptr addrspace(1) %ptr,
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
; GFX11-NEXT: global_load_u16 v0, v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_readfirstlane_b32 s0, v0
@@ -842,7 +843,8 @@ define i16 @extractelement_vgpr_v8i16_vgpr_idx(ptr addrspace(1) %ptr, i32 %idx)
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v2, 1, v2
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: global_load_u16 v0, v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i8.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i8.ll
index c2394ec461490..e6a02c6ae6919 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i8.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i8.ll
@@ -131,9 +131,9 @@ define amdgpu_ps i8 @extractelement_vgpr_v4i8_sgpr_idx(ptr addrspace(1) %ptr, i3
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_ashr_i32 s1, s0, 31
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
-; GFX11-NEXT: s_...
[truncated]
|
Member
|
@llvm/pr-subscribers-backend-amdgpu Author: Ana Mihajlovic (mihajlovicana) ChangesUnused sdst writing to null to avoid a false VALU->SALU dependency stall. This requires using the VOP3 encoding. Patch is 819.85 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/133229.diff 68 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
index f03cde455f295..06d5e48cdab73 100644
--- a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
+++ b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
@@ -973,6 +973,11 @@ bool SIShrinkInstructions::run(MachineFunction &MF) {
continue;
}
+ if (TII->isVOP3(MI.getOpcode()) &&
+ TII->hasVALU32BitEncoding(MI.getOpcode())) {
+ tryReplaceDeadSDST(MI);
+ }
+
if (!TII->hasVALU32BitEncoding(MI.getOpcode())) {
// If there is no chance we will shrink it and use VCC as sdst to get
// a 32 bit form try to replace dead sdst with NULL.
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/addsubu64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/addsubu64.ll
index 6a04dd492fcea..425dd8acd4736 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/addsubu64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/addsubu64.ll
@@ -39,7 +39,8 @@ define amdgpu_ps void @v_add_u64(ptr addrspace(1) %out, i64 %a, i64 %b) {
; GCN-LABEL: v_add_u64:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4
-; GCN-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
+; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GCN-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v5, vcc_lo
; GCN-NEXT: global_store_b64 v[0:1], v[2:3], off
; GCN-NEXT: s_endpgm
entry:
@@ -85,7 +86,8 @@ define amdgpu_ps void @v_sub_u64(ptr addrspace(1) %out, i64 %a, i64 %b) {
; GCN-LABEL: v_sub_u64:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: v_sub_co_u32 v2, vcc_lo, v2, v4
-; GCN-NEXT: v_sub_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
+; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GCN-NEXT: v_sub_co_ci_u32_e64 v3, null, v3, v5, vcc_lo
; GCN-NEXT: global_store_b64 v[0:1], v[2:3], off
; GCN-NEXT: s_endpgm
entry:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_udec_wrap.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_udec_wrap.ll
index dce4048a4b87e..d9be677a0e58d 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_udec_wrap.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_udec_wrap.ll
@@ -1440,16 +1440,16 @@ define amdgpu_kernel void @flat_atomic_dec_ret_i32_offset_addr64(ptr %out, ptr %
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_dec_u32 v3, v[0:1], v3 offset:20 glc
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_gl1_inv
; GFX11-NEXT: buffer_gl0_inv
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_store_b32 v[0:1], v3
; GFX11-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -1539,8 +1539,8 @@ define amdgpu_kernel void @flat_atomic_dec_noret_i32_offset_addr64(ptr %ptr) #1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: v_mov_b32_e32 v2, 42
; GFX11-NEXT: flat_atomic_dec_u32 v[0:1], v2 offset:20
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
@@ -2090,16 +2090,16 @@ define amdgpu_kernel void @flat_atomic_dec_ret_i64_offset_addr64(ptr %out, ptr %
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_dec_u64 v[0:1], v[0:1], v[2:3] offset:40 glc
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_gl1_inv
; GFX11-NEXT: buffer_gl0_inv
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4
-; GFX11-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
+; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: flat_store_b64 v[2:3], v[0:1]
; GFX11-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -2194,8 +2194,8 @@ define amdgpu_kernel void @flat_atomic_dec_noret_i64_offset_addr64(ptr %ptr) #1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_dec_u64 v[0:1], v[2:3] offset:40
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll
index e2d179a77f76c..92a7de9aaefd2 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll
@@ -2824,16 +2824,16 @@ define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset_addr64(ptr %out, ptr %
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_inc_u32 v3, v[0:1], v3 offset:20 glc
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_gl1_inv
; GFX11-NEXT: buffer_gl0_inv
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_store_b32 v[0:1], v3
; GFX11-NEXT: s_endpgm
;
@@ -2846,15 +2846,15 @@ define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset_addr64(ptr %out, ptr %
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_1)
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1)
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: flat_atomic_inc_u32 v3, v[0:1], v3 offset:20 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: global_inv scope:SCOPE_DEV
; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: flat_store_b32 v[0:1], v3
; GFX12-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -2944,8 +2944,8 @@ define amdgpu_kernel void @flat_atomic_inc_noret_i32_offset_addr64(ptr %ptr) #1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: v_mov_b32_e32 v2, 42
; GFX11-NEXT: flat_atomic_inc_u32 v[0:1], v2 offset:20
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
@@ -2963,8 +2963,8 @@ define amdgpu_kernel void @flat_atomic_inc_noret_i32_offset_addr64(ptr %ptr) #1
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: v_mov_b32_e32 v2, 42
; GFX12-NEXT: flat_atomic_inc_u32 v[0:1], v2 offset:20 scope:SCOPE_DEV
; GFX12-NEXT: s_wait_storecnt_dscnt 0x0
@@ -3810,16 +3810,16 @@ define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset_addr64(ptr %out, ptr %
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_inc_u64 v[0:1], v[0:1], v[2:3] offset:40 glc
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_gl1_inv
; GFX11-NEXT: buffer_gl0_inv
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4
-; GFX11-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
+; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX11-NEXT: flat_store_b64 v[2:3], v[0:1]
; GFX11-NEXT: s_endpgm
;
@@ -3833,15 +3833,15 @@ define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset_addr64(ptr %out, ptr %
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_1)
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1)
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: flat_atomic_inc_u64 v[0:1], v[0:1], v[2:3] offset:40 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: global_inv scope:SCOPE_DEV
; GFX12-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
; GFX12-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, vcc_lo
; GFX12-NEXT: flat_store_b64 v[2:3], v[0:1]
; GFX12-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x()
@@ -3936,8 +3936,8 @@ define amdgpu_kernel void @flat_atomic_inc_noret_i64_offset_addr64(ptr %ptr) #1
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: flat_atomic_inc_u64 v[0:1], v[2:3] offset:40
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
@@ -3955,8 +3955,8 @@ define amdgpu_kernel void @flat_atomic_inc_noret_i64_offset_addr64(ptr %ptr) #1
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v4
-; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: flat_atomic_inc_u64 v[0:1], v[2:3] offset:40 scope:SCOPE_DEV
; GFX12-NEXT: s_wait_storecnt_dscnt 0x0
; GFX12-NEXT: global_inv scope:SCOPE_DEV
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
index a948446aceff1..573017f7a948d 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
@@ -29,10 +29,9 @@ define i32 @v_extract_v64i32_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GFX12-NEXT: v_lshlrev_b32_e32 v2, 2, v2
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: global_load_b32 v0, v[0:1], off
; GFX12-NEXT: s_wait_loadcnt 0x0
-; GFX12-NEXT: s_wait_alu 0xfffd
; GFX12-NEXT: s_setpc_b64 s[30:31]
%vec = load <64 x i32>, ptr addrspace(1) %ptr
%elt = extractelement <64 x i32> %vec, i32 %idx
@@ -63,10 +62,9 @@ define i16 @v_extract_v128i16_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GFX12-NEXT: v_lshlrev_b32_e32 v2, 1, v2
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: global_load_u16 v0, v[0:1], off
; GFX12-NEXT: s_wait_loadcnt 0x0
-; GFX12-NEXT: s_wait_alu 0xfffd
; GFX12-NEXT: s_setpc_b64 s[30:31]
%vec = load <128 x i16>, ptr addrspace(1) %ptr
%elt = extractelement <128 x i16> %vec, i32 %idx
@@ -97,10 +95,9 @@ define i64 @v_extract_v32i64_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GFX12-NEXT: v_lshlrev_b32_e32 v2, 3, v2
; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
; GFX12-NEXT: s_wait_alu 0xfffd
-; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX12-NEXT: global_load_b64 v[0:1], v[0:1], off
; GFX12-NEXT: s_wait_loadcnt 0x0
-; GFX12-NEXT: s_wait_alu 0xfffd
; GFX12-NEXT: s_setpc_b64 s[30:31]
%vec = load <32 x i64>, ptr addrspace(1) %ptr
%elt = extractelement <32 x i64> %vec, i32 %idx
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll
index e1ce9ea14a2a9..c424738dceb0e 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll
@@ -126,8 +126,8 @@ define amdgpu_ps i128 @extractelement_vgpr_v4i128_sgpr_idx(ptr addrspace(1) %ptr
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_readfirstlane_b32 s0, v0
@@ -195,7 +195,8 @@ define i128 @extractelement_vgpr_v4i128_vgpr_idx(ptr addrspace(1) %ptr, i32 %idx
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v2, 4, v2
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll
index 021f609053a0f..63c3146ebeedd 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll
@@ -133,8 +133,8 @@ define amdgpu_ps i16 @extractelement_vgpr_v4i16_sgpr_idx(ptr addrspace(1) %ptr,
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
; GFX11-NEXT: global_load_u16 v0, v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_readfirstlane_b32 s0, v0
@@ -199,7 +199,8 @@ define i16 @extractelement_vgpr_v4i16_vgpr_idx(ptr addrspace(1) %ptr, i32 %idx)
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v2, 1, v2
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: global_load_u16 v0, v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
@@ -776,8 +777,8 @@ define amdgpu_ps i16 @extractelement_vgpr_v8i16_sgpr_idx(ptr addrspace(1) %ptr,
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
; GFX11-NEXT: global_load_u16 v0, v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_readfirstlane_b32 s0, v0
@@ -842,7 +843,8 @@ define i16 @extractelement_vgpr_v8i16_vgpr_idx(ptr addrspace(1) %ptr, i32 %idx)
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_lshlrev_b32_e32 v2, 1, v2
; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
-; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, 0, v1, vcc_lo
; GFX11-NEXT: global_load_u16 v0, v[0:1], off
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i8.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i8.ll
index c2394ec461490..e6a02c6ae6919 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i8.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i8.ll
@@ -131,9 +131,9 @@ define amdgpu_ps i8 @extractelement_vgpr_v4i8_sgpr_idx(ptr addrspace(1) %ptr, i3
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_ashr_i32 s1, s0, 31
; GFX11-NEXT: v_dual_mov_b32 v3, s1 :: v_dual_mov_b32 v2, s0
-; GFX11-NEXT: s_...
[truncated]
|
arsenm
reviewed
Mar 27, 2025
| } | ||
|
|
||
| if (TII->isVOP3(MI.getOpcode()) && | ||
| TII->hasVALU32BitEncoding(MI.getOpcode())) { |
Contributor
There was a problem hiding this comment.
Merge this with the existing hasVALU32BitEncoding check below
arsenm
approved these changes
Mar 28, 2025
| ; GCN: ; %bb.0: ; %entry | ||
| ; GCN-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4 | ||
| ; GCN-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo | ||
| ; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1) |
Contributor
There was a problem hiding this comment.
Adds an extra instruction, is this really better?
Author
There was a problem hiding this comment.
Delay alu pass skips implicit def ops (suppose that should be done since it's marked with TODO).
VOP2:
renamable $vgpr13 = V_ADDC_U32_e32 0, killed $vgpr13, implicit-def dead $vcc_lo, implicit killed $vcc_lo, implicit $exec
VOP3:
renamable $vgpr13, dead $sgpr_null = V_ADDC_U32_e64 0, killed $vgpr13, killed $vcc_lo, 0, implicit $exec
In VOP2, vcc is implicit so delay alu will not consider it, but as I understood the correct behaviour would be for delay_alu to be in both?
Contributor
There was a problem hiding this comment.
There are two things going on here:
- In general
AMDGPUInsertDelayAlushould look at implicit ops, like you said. - But in this specific case of carry out -> carry in, the hardware has a fast forward that means there is no delay, so no
s_delay_aluis needed.
jayfoad
reviewed
Apr 3, 2025
Comment on lines
+976
to
+977
| // If there is no chance we will shrink it and use VCC as sdst to get | ||
| // a 32 bit form try to replace dead sdst with NULL. |
Contributor
There was a problem hiding this comment.
Comment needs updating. We are now trying to replace dead sdst with null even if there is a chance that we could shrink to a 32-bit encoding.
Comment on lines
+980
to
+982
| if (!TII->hasVALU32BitEncoding(MI.getOpcode())) { | ||
| continue; | ||
| } |
Contributor
There was a problem hiding this comment.
I really don't see the point of adding this check here. What does it buy us? If you want to save some execution time it would be better for tryReplaceDeadSDST to return a bool saying whether it succeeded. If it did, you can probably continue here because there is no chance of shrinking it.
Also nit: don't need the braces.
Author
There was a problem hiding this comment.
Skipping this check will lead to a change in a lot of tests and illegal virtual register errors in some of them
| ; GCN: ; %bb.0: ; %entry | ||
| ; GCN-NEXT: v_add_co_u32 v2, vcc_lo, v2, v4 | ||
| ; GCN-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo | ||
| ; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1) |
Contributor
There was a problem hiding this comment.
There are two things going on here:
- In general
AMDGPUInsertDelayAlushould look at implicit ops, like you said. - But in this specific case of carry out -> carry in, the hardware has a fast forward that means there is no delay, so no
s_delay_aluis needed.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Unused sdst writing to null to avoid a false VALU->SALU dependency stall. This requires using the VOP3 encoding.