[llvm] 8e97378 - [NFC][AMDGPU] don't test asyncmark for non-async variants on pre-gfx12 (#190774)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Apr 8 01:57:49 PDT 2026
Author: Sameer Sahasrabuddhe
Date: 2026-04-08T14:27:44+05:30
New Revision: 8e9737804061a2c10db90a69362ee26e3ee4a7bf
URL: https://github.com/llvm/llvm-project/commit/8e9737804061a2c10db90a69362ee26e3ee4a7bf
DIFF: https://github.com/llvm/llvm-project/commit/8e9737804061a2c10db90a69362ee26e3ee4a7bf.diff
LOG: [NFC][AMDGPU] don't test asyncmark for non-async variants on pre-gfx12 (#190774)
The test for asyncmark on pref-gfx12 targets, checks the effect of async
marks on both async and non-async instructions. This is achieved by a
sed invocation in the RUN line to remove the .async from intrinsic name.
Such a test serves no purpose other than a minor flex. The two sets of
CHECK lines using the WITH and WITHOUT prefix are not easy to compare by
a simple eyeballing. Eliminating this hack makes it easier to add GISEL
variants later. With and without GISEL, the combinations will increase
to four instead of the current two.
Assisted-By: Claude Opus 4.6
Added:
Modified:
llvm/test/CodeGen/AMDGPU/asyncmark-pregfx12.ll
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/AMDGPU/asyncmark-pregfx12.ll b/llvm/test/CodeGen/AMDGPU/asyncmark-pregfx12.ll
index 41292bf4aa829..b8ec819f8e310 100644
--- a/llvm/test/CodeGen/AMDGPU/asyncmark-pregfx12.ll
+++ b/llvm/test/CodeGen/AMDGPU/asyncmark-pregfx12.ll
@@ -1,6 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
-; RUN: sed 's/.ASYNC/.async/' %s | llc -mtriple=amdgcn -mcpu=gfx900 -o - | FileCheck %s -check-prefixes=WITHASYNC
-; RUN: sed 's/.ASYNC//' %s | llc -mtriple=amdgcn -mcpu=gfx900 -o - | FileCheck %s -check-prefixes=WITHOUT
+; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck %s -check-prefixes=WITHASYNC
; Demonstrate that wait.asyncmark is a code motion barrier for loads from LDS.
; This is the simplest demo possible. We don't actually use async ops, but just
@@ -20,17 +19,6 @@ define void @code_barrier(ptr addrspace(1) %foo, ptr addrspace(3) %lds, ptr addr
; WITHASYNC-NEXT: s_waitcnt lgkmcnt(0)
; WITHASYNC-NEXT: s_setpc_b64 s[30:31]
;
-; WITHOUT-LABEL: code_barrier:
-; WITHOUT: ; %bb.0:
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: ds_read_b32 v0, v2
-; WITHOUT-NEXT: ; wait_asyncmark(0)
-; WITHOUT-NEXT: ds_read_b32 v1, v2 offset:4
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(0)
-; WITHOUT-NEXT: v_add_u32_e32 v0, v0, v1
-; WITHOUT-NEXT: ds_write_b32 v3, v0
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(0)
-; WITHOUT-NEXT: s_setpc_b64 s[30:31]
%lds_gep1 = getelementptr i32, ptr addrspace(3) %lds, i32 1
%val1 = load i32, ptr addrspace(3) %lds
call void @llvm.amdgcn.wait.asyncmark(i16 0)
@@ -79,49 +67,18 @@ define void @interleaved_global_and_dma(ptr addrspace(1) %foo, ptr addrspace(3)
; WITHASYNC-NEXT: s_waitcnt vmcnt(0)
; WITHASYNC-NEXT: s_setpc_b64 s[30:31]
;
-; WITHOUT-LABEL: interleaved_global_and_dma:
-; WITHOUT: ; %bb.0: ; %entry
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: v_readfirstlane_b32 s4, v2
-; WITHOUT-NEXT: global_load_dword v7, v[3:4], off
-; WITHOUT-NEXT: global_load_dword v8, v[0:1], off
-; WITHOUT-NEXT: s_mov_b32 m0, s4
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: s_nop 0
-; WITHOUT-NEXT: global_load_dword v[3:4], off lds
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: global_load_dword v0, v[0:1], off
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: s_nop 0
-; WITHOUT-NEXT: global_load_dword v[3:4], off lds
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: global_load_dword v1, v[3:4], off
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: ; wait_asyncmark(1)
-; WITHOUT-NEXT: s_waitcnt vmcnt(1)
-; WITHOUT-NEXT: ds_read_b32 v3, v2
-; WITHOUT-NEXT: ; wait_asyncmark(0)
-; WITHOUT-NEXT: ds_read_b32 v2, v2
-; WITHOUT-NEXT: v_add_u32_e32 v4, v8, v7
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(1)
-; WITHOUT-NEXT: v_add3_u32 v0, v4, v3, v0
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: v_add3_u32 v0, v0, v1, v2
-; WITHOUT-NEXT: global_store_dword v[5:6], v0, off
-; WITHOUT-NEXT: s_waitcnt vmcnt(0)
-; WITHOUT-NEXT: s_setpc_b64 s[30:31]
entry:
; First batch: global load, global load, async global-to-LDS
%bar_v11 = load i32, ptr addrspace(1) %bar
%foo_v1 = load i32, ptr addrspace(1) %foo
call void @llvm.amdgcn.wave.barrier()
- call void @llvm.amdgcn.global.load.ASYNC.lds(ptr addrspace(1) %bar, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
+ call void @llvm.amdgcn.global.load.async.lds(ptr addrspace(1) %bar, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
call void @llvm.amdgcn.asyncmark()
; Second batch: global load, async global-to-LDS, global load
%foo_v2 = load i32, ptr addrspace(1) %foo
call void @llvm.amdgcn.wave.barrier()
- call void @llvm.amdgcn.global.load.ASYNC.lds(ptr addrspace(1) %bar, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
+ call void @llvm.amdgcn.global.load.async.lds(ptr addrspace(1) %bar, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
call void @llvm.amdgcn.wave.barrier()
%bar_v12 = load i32, ptr addrspace(1) %bar
call void @llvm.amdgcn.asyncmark()
@@ -180,49 +137,18 @@ define void @interleaved_buffer_and_dma(ptr addrspace(8) inreg %buf, ptr addrspa
; WITHASYNC-NEXT: s_waitcnt vmcnt(0)
; WITHASYNC-NEXT: s_setpc_b64 s[30:31]
;
-; WITHOUT-LABEL: interleaved_buffer_and_dma:
-; WITHOUT: ; %bb.0: ; %entry
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: s_mov_b32 m0, s20
-; WITHOUT-NEXT: global_load_dword v6, v[2:3], off
-; WITHOUT-NEXT: global_load_dword v7, v[0:1], off
-; WITHOUT-NEXT: v_mov_b32_e32 v8, 0x54
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: buffer_load_dword v8, s[16:19], 0 offen lds
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: global_load_dword v0, v[0:1], off
-; WITHOUT-NEXT: v_mov_b32_e32 v1, 0x58
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: buffer_load_dword v1, s[16:19], 0 offen lds
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: global_load_dword v1, v[2:3], off
-; WITHOUT-NEXT: v_mov_b32_e32 v2, s20
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: ; wait_asyncmark(1)
-; WITHOUT-NEXT: s_waitcnt vmcnt(1)
-; WITHOUT-NEXT: ds_read_b32 v3, v2
-; WITHOUT-NEXT: ; wait_asyncmark(0)
-; WITHOUT-NEXT: ds_read_b32 v2, v2
-; WITHOUT-NEXT: v_add_u32_e32 v6, v7, v6
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(1)
-; WITHOUT-NEXT: v_add3_u32 v0, v6, v3, v0
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: v_add3_u32 v0, v0, v1, v2
-; WITHOUT-NEXT: global_store_dword v[4:5], v0, off
-; WITHOUT-NEXT: s_waitcnt vmcnt(0)
-; WITHOUT-NEXT: s_setpc_b64 s[30:31]
entry:
; First batch: global load, global load, async global-to-LDS.
%bar_v11 = load i32, ptr addrspace(1) %bar
%foo_v1 = load i32, ptr addrspace(1) %foo
call void @llvm.amdgcn.wave.barrier()
- call void @llvm.amdgcn.raw.ptr.buffer.load.ASYNC.lds(ptr addrspace(8) %buf, ptr addrspace(3) %lds, i32 4, i32 84, i32 0, i32 0, i32 0)
+ call void @llvm.amdgcn.raw.ptr.buffer.load.async.lds(ptr addrspace(8) %buf, ptr addrspace(3) %lds, i32 4, i32 84, i32 0, i32 0, i32 0)
call void @llvm.amdgcn.asyncmark()
; Second batch: global load, async global-to-LDS, global load.
%foo_v2 = load i32, ptr addrspace(1) %foo
call void @llvm.amdgcn.wave.barrier()
- call void @llvm.amdgcn.raw.ptr.buffer.load.ASYNC.lds(ptr addrspace(8) %buf, ptr addrspace(3) %lds, i32 4, i32 88, i32 0, i32 0, i32 0)
+ call void @llvm.amdgcn.raw.ptr.buffer.load.async.lds(ptr addrspace(8) %buf, ptr addrspace(3) %lds, i32 4, i32 88, i32 0, i32 0, i32 0)
call void @llvm.amdgcn.wave.barrier()
%bar_v12 = load i32, ptr addrspace(1) %bar
call void @llvm.amdgcn.asyncmark()
@@ -281,51 +207,19 @@ define void @fence_with_asyncmark(ptr addrspace(8) inreg %buf, ptr addrspace(1)
; WITHASYNC-NEXT: s_waitcnt vmcnt(0)
; WITHASYNC-NEXT: s_setpc_b64 s[30:31]
;
-; WITHOUT-LABEL: fence_with_asyncmark:
-; WITHOUT: ; %bb.0: ; %entry
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: s_mov_b32 m0, s20
-; WITHOUT-NEXT: global_load_dword v6, v[2:3], off
-; WITHOUT-NEXT: global_load_dword v7, v[0:1], off
-; WITHOUT-NEXT: v_mov_b32_e32 v8, 0x54
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: buffer_load_dword v8, s[16:19], 0 offen lds
-; WITHOUT-NEXT: s_waitcnt vmcnt(0)
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: global_load_dword v0, v[0:1], off
-; WITHOUT-NEXT: v_mov_b32_e32 v1, 0x58
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: buffer_load_dword v1, s[16:19], 0 offen lds
-; WITHOUT-NEXT: ; wave barrier
-; WITHOUT-NEXT: global_load_dword v1, v[2:3], off
-; WITHOUT-NEXT: v_mov_b32_e32 v2, s20
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: ; wait_asyncmark(1)
-; WITHOUT-NEXT: s_waitcnt vmcnt(1)
-; WITHOUT-NEXT: ds_read_b32 v3, v2
-; WITHOUT-NEXT: ; wait_asyncmark(0)
-; WITHOUT-NEXT: ds_read_b32 v2, v2
-; WITHOUT-NEXT: v_add_u32_e32 v6, v7, v6
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(1)
-; WITHOUT-NEXT: v_add3_u32 v0, v6, v3, v0
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: v_add3_u32 v0, v0, v1, v2
-; WITHOUT-NEXT: global_store_dword v[4:5], v0, off
-; WITHOUT-NEXT: s_waitcnt vmcnt(0)
-; WITHOUT-NEXT: s_setpc_b64 s[30:31]
entry:
; First batch: global load, global load, async global-to-LDS.
%bar_v11 = load i32, ptr addrspace(1) %bar
%foo_v1 = load i32, ptr addrspace(1) %foo
call void @llvm.amdgcn.wave.barrier()
- call void @llvm.amdgcn.raw.ptr.buffer.load.ASYNC.lds(ptr addrspace(8) %buf, ptr addrspace(3) %lds, i32 4, i32 84, i32 0, i32 0, i32 0)
+ call void @llvm.amdgcn.raw.ptr.buffer.load.async.lds(ptr addrspace(8) %buf, ptr addrspace(3) %lds, i32 4, i32 84, i32 0, i32 0, i32 0)
fence release
call void @llvm.amdgcn.asyncmark()
; Second batch: global load, async global-to-LDS, global load.
%foo_v2 = load i32, ptr addrspace(1) %foo
call void @llvm.amdgcn.wave.barrier()
- call void @llvm.amdgcn.raw.ptr.buffer.load.ASYNC.lds(ptr addrspace(8) %buf, ptr addrspace(3) %lds, i32 4, i32 88, i32 0, i32 0, i32 0)
+ call void @llvm.amdgcn.raw.ptr.buffer.load.async.lds(ptr addrspace(8) %buf, ptr addrspace(3) %lds, i32 4, i32 88, i32 0, i32 0, i32 0)
call void @llvm.amdgcn.wave.barrier()
%bar_v12 = load i32, ptr addrspace(1) %bar
call void @llvm.amdgcn.asyncmark()
@@ -394,51 +288,13 @@ define void @test_pipelined_loop(ptr addrspace(1) %foo, ptr addrspace(3) %lds, p
; WITHASYNC-NEXT: s_waitcnt vmcnt(0)
; WITHASYNC-NEXT: s_setpc_b64 s[30:31]
;
-; WITHOUT-LABEL: test_pipelined_loop:
-; WITHOUT: ; %bb.0: ; %prolog
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: v_readfirstlane_b32 s4, v2
-; WITHOUT-NEXT: s_mov_b32 m0, s4
-; WITHOUT-NEXT: v_mov_b32_e32 v5, 0
-; WITHOUT-NEXT: global_load_dword v[0:1], off lds
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: global_load_dword v[0:1], off lds
-; WITHOUT-NEXT: s_mov_b32 s6, 2
-; WITHOUT-NEXT: s_mov_b64 s[4:5], 0
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: .LBB4_1: ; %loop_body
-; WITHOUT-NEXT: ; =>This Inner Loop Header: Depth=1
-; WITHOUT-NEXT: v_readfirstlane_b32 s7, v2
-; WITHOUT-NEXT: s_mov_b32 m0, s7
-; WITHOUT-NEXT: s_add_i32 s6, s6, 1
-; WITHOUT-NEXT: global_load_dword v[0:1], off lds
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: ; wait_asyncmark(2)
-; WITHOUT-NEXT: s_waitcnt vmcnt(0)
-; WITHOUT-NEXT: ds_read_b32 v6, v2
-; WITHOUT-NEXT: v_cmp_ge_i32_e32 vcc, s6, v7
-; WITHOUT-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(0)
-; WITHOUT-NEXT: v_add_u32_e32 v5, v5, v6
-; WITHOUT-NEXT: s_andn2_b64 exec, exec, s[4:5]
-; WITHOUT-NEXT: s_cbranch_execnz .LBB4_1
-; WITHOUT-NEXT: ; %bb.2: ; %epilog
-; WITHOUT-NEXT: s_or_b64 exec, exec, s[4:5]
-; WITHOUT-NEXT: ; wait_asyncmark(1)
-; WITHOUT-NEXT: ds_read_b32 v0, v2
-; WITHOUT-NEXT: ; wait_asyncmark(0)
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(0)
-; WITHOUT-NEXT: v_add_u32_e32 v0, v5, v0
-; WITHOUT-NEXT: global_store_dword v[3:4], v0, off
-; WITHOUT-NEXT: s_waitcnt vmcnt(0)
-; WITHOUT-NEXT: s_setpc_b64 s[30:31]
prolog:
; Load first iteration
- call void @llvm.amdgcn.global.load.ASYNC.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
+ call void @llvm.amdgcn.global.load.async.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
call void @llvm.amdgcn.asyncmark()
; Load second iteration
- call void @llvm.amdgcn.global.load.ASYNC.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
+ call void @llvm.amdgcn.global.load.async.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
call void @llvm.amdgcn.asyncmark()
br label %loop_body
@@ -448,7 +304,7 @@ loop_body:
%sum = phi i32 [ 0, %prolog ], [ %sum_i, %loop_body ]
; Load next iteration
- call void @llvm.amdgcn.global.load.ASYNC.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
+ call void @llvm.amdgcn.global.load.async.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
call void @llvm.amdgcn.asyncmark()
; Wait for iteration i-2 and process
@@ -540,76 +396,17 @@ define void @test_pipelined_loop_with_global(ptr addrspace(1) %foo, ptr addrspac
; WITHASYNC-NEXT: s_waitcnt vmcnt(0)
; WITHASYNC-NEXT: s_setpc_b64 s[30:31]
;
-; WITHOUT-LABEL: test_pipelined_loop_with_global:
-; WITHOUT: ; %bb.0: ; %prolog
-; WITHOUT-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WITHOUT-NEXT: v_readfirstlane_b32 s4, v2
-; WITHOUT-NEXT: s_mov_b32 m0, s4
-; WITHOUT-NEXT: global_load_dword v10, v[0:1], off
-; WITHOUT-NEXT: global_load_dword v14, v[3:4], off
-; WITHOUT-NEXT: s_mov_b32 s6, 2
-; WITHOUT-NEXT: global_load_dword v[0:1], off lds
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: global_load_dword v8, v[0:1], off
-; WITHOUT-NEXT: global_load_dword v9, v[3:4], off
-; WITHOUT-NEXT: s_mov_b64 s[4:5], 0
-; WITHOUT-NEXT: global_load_dword v[0:1], off lds
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: s_waitcnt vmcnt(2)
-; WITHOUT-NEXT: v_mov_b32_e32 v13, v8
-; WITHOUT-NEXT: s_waitcnt vmcnt(1)
-; WITHOUT-NEXT: v_mov_b32_e32 v15, v9
-; WITHOUT-NEXT: .LBB5_1: ; %loop_body
-; WITHOUT-NEXT: ; =>This Inner Loop Header: Depth=1
-; WITHOUT-NEXT: v_readfirstlane_b32 s7, v2
-; WITHOUT-NEXT: s_waitcnt vmcnt(1)
-; WITHOUT-NEXT: v_mov_b32_e32 v12, v15
-; WITHOUT-NEXT: v_mov_b32_e32 v11, v13
-; WITHOUT-NEXT: global_load_dword v13, v[0:1], off
-; WITHOUT-NEXT: global_load_dword v15, v[3:4], off
-; WITHOUT-NEXT: s_mov_b32 m0, s7
-; WITHOUT-NEXT: s_add_i32 s6, s6, 1
-; WITHOUT-NEXT: global_load_dword v[0:1], off lds
-; WITHOUT-NEXT: v_cmp_ge_i32_e32 vcc, s6, v7
-; WITHOUT-NEXT: v_mov_b32_e32 v16, v14
-; WITHOUT-NEXT: v_mov_b32_e32 v17, v10
-; WITHOUT-NEXT: v_mov_b32_e32 v10, v8
-; WITHOUT-NEXT: v_mov_b32_e32 v14, v9
-; WITHOUT-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
-; WITHOUT-NEXT: ; asyncmark
-; WITHOUT-NEXT: ; wait_asyncmark(2)
-; WITHOUT-NEXT: s_andn2_b64 exec, exec, s[4:5]
-; WITHOUT-NEXT: s_cbranch_execnz .LBB5_1
-; WITHOUT-NEXT: ; %bb.2: ; %epilog
-; WITHOUT-NEXT: s_or_b64 exec, exec, s[4:5]
-; WITHOUT-NEXT: s_waitcnt vmcnt(0)
-; WITHOUT-NEXT: ds_read_b32 v0, v2
-; WITHOUT-NEXT: ; wait_asyncmark(1)
-; WITHOUT-NEXT: ds_read_b32 v1, v2
-; WITHOUT-NEXT: ; wait_asyncmark(0)
-; WITHOUT-NEXT: ds_read_b32 v2, v2
-; WITHOUT-NEXT: v_add_u32_e32 v3, v17, v16
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(2)
-; WITHOUT-NEXT: v_add3_u32 v0, v3, v0, v12
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(1)
-; WITHOUT-NEXT: v_add3_u32 v0, v11, v0, v1
-; WITHOUT-NEXT: v_add_u32_e32 v1, v13, v15
-; WITHOUT-NEXT: s_waitcnt lgkmcnt(0)
-; WITHOUT-NEXT: v_add3_u32 v0, v1, v2, v0
-; WITHOUT-NEXT: global_store_dword v[5:6], v0, off
-; WITHOUT-NEXT: s_waitcnt vmcnt(0)
-; WITHOUT-NEXT: s_setpc_b64 s[30:31]
prolog:
; Load first iteration
%v0 = load i32, ptr addrspace(1) %foo
%g0 = load i32, ptr addrspace(1) %bar
- call void @llvm.amdgcn.global.load.ASYNC.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
+ call void @llvm.amdgcn.global.load.async.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
call void @llvm.amdgcn.asyncmark()
; Load second iteration
%v1 = load i32, ptr addrspace(1) %foo
%g1 = load i32, ptr addrspace(1) %bar
- call void @llvm.amdgcn.global.load.ASYNC.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
+ call void @llvm.amdgcn.global.load.async.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
call void @llvm.amdgcn.asyncmark()
br label %loop_body
@@ -632,7 +429,7 @@ loop_body:
; Load next iteration
%cur_v = load i32, ptr addrspace(1) %foo
%cur_g = load i32, ptr addrspace(1) %bar
- call void @llvm.amdgcn.global.load.ASYNC.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
+ call void @llvm.amdgcn.global.load.async.lds(ptr addrspace(1) %foo, ptr addrspace(3) %lds, i32 4, i32 0, i32 0)
call void @llvm.amdgcn.asyncmark()
; Wait for iteration i-2 and process
More information about the llvm-commits
mailing list