File: buffer-schedule.ll

package info (click to toggle)
llvm-toolchain-17 1%3A17.0.6-22
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid, trixie
  • size: 1,799,624 kB
  • sloc: cpp: 6,428,607; ansic: 1,383,196; asm: 793,408; python: 223,504; objc: 75,364; f90: 60,502; lisp: 33,869; pascal: 15,282; sh: 9,684; perl: 7,453; ml: 4,937; awk: 3,523; makefile: 2,889; javascript: 2,149; xml: 888; fortran: 619; cs: 573
file content (53 lines) | stat: -rw-r--r-- 2,314 bytes parent folder | download | duplicates (9)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
; RUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s

; GCN-LABEL: {{^}}test1:
; GCN: buffer_store_dword
; GCN: buffer_load_dword
; GCN: buffer_store_dword
define amdgpu_cs void @test1(<4 x i32> inreg %buf, i32 %off) {
.entry:
  call void @llvm.amdgcn.raw.buffer.store.i32(i32 0, <4 x i32> %buf, i32 8, i32 0, i32 0)
  %val = call i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32> %buf, i32 %off, i32 0, i32 0)
  call void @llvm.amdgcn.raw.buffer.store.i32(i32 %val, <4 x i32> %buf, i32 0, i32 0, i32 0)
  ret void
}

; GCN-LABEL: {{^}}test1_ptrs:
; GCN: buffer_store_dword
; GCN: buffer_load_dword
; GCN: buffer_store_dword
define amdgpu_cs void @test1_ptrs(ptr addrspace(8) inreg %buf, i32 %off) {
.entry:
  call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 0, ptr addrspace(8) %buf, i32 8, i32 0, i32 0)
  %val = call i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) %buf, i32 %off, i32 0, i32 0)
  call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 %val, ptr addrspace(8) %buf, i32 0, i32 0, i32 0)
  ret void
}

;; In the future, the stores should be reorderable because they'd be known to be
;; at distinct offsets.
; GCN-LABEL: {{^}}test1_ptrs_reorderable:
; GCN: buffer_store_dword
; GCN: buffer_load_dword
; GCN: buffer_store_dword
define amdgpu_cs void @test1_ptrs_reorderable(ptr addrspace(8) inreg %buf, i32 %off) {
.entry:
  %shifted.off = shl i32 %off, 4
  call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 0, ptr addrspace(8) %buf, i32 8, i32 0, i32 0)
  %val = call i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) %buf, i32 %shifted.off, i32 0, i32 0)
  call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 %val, ptr addrspace(8) %buf, i32 0, i32 0, i32 0)
  ret void
}

declare i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32>, i32, i32, i32) #2

declare void @llvm.amdgcn.raw.buffer.store.i32(i32, <4 x i32>, i32, i32, i32) #3

declare i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) nocapture, i32, i32, i32) #4

declare void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32, ptr addrspace(8) nocapture, i32, i32, i32) #5

attributes #2 = { nounwind readonly }
attributes #3 = { nounwind writeonly }
attributes #4 = { nounwind memory(argmem: read) }
attributes #5 = { nounwind memory(argmem: write) }