reference, declaration → definition
definition → references, declarations, derived classes, virtual overrides
reference to multiple definitions → definitions
unreferenced

    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
   54
   55
   56
   57
   58
   59
   60
   61
   62
   63
   64
   65

; RUN: opt -S -mtriple=amdgcn-- -data-layout=A5 -amdgpu-promote-alloca -sroa -instcombine < %s | FileCheck -check-prefix=OPT %s

; Show that what the alloca promotion pass will do for non-atomic load/store.

; OPT-LABEL: @vector_alloca_not_atomic(
;
; OPT: extractelement <3 x i32> <i32 0, i32 1, i32 2>, i64 %index
define amdgpu_kernel void @vector_alloca_not_atomic(i32 addrspace(1)* %out, i64 %index) {
entry:
  %alloca = alloca [3 x i32], addrspace(5)
  %a0 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 0
  %a1 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 1
  %a2 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 2
  store i32 0, i32 addrspace(5)* %a0
  store i32 1, i32 addrspace(5)* %a1
  store i32 2, i32 addrspace(5)* %a2
  %tmp = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i64 0, i64 %index
  %data = load i32, i32 addrspace(5)* %tmp
  store i32 %data, i32 addrspace(1)* %out
  ret void
}

; OPT-LABEL: @vector_alloca_atomic_read(
;
; OPT: alloca [3 x i32]
; OPT: store i32 0, i32 addrspace(5)*
; OPT: store i32 1, i32 addrspace(5)*
; OPT: store i32 2, i32 addrspace(5)*
; OPT: load atomic i32, i32 addrspace(5)*
define amdgpu_kernel void @vector_alloca_atomic_read(i32 addrspace(1)* %out, i64 %index) {
entry:
  %alloca = alloca [3 x i32], addrspace(5)
  %a0 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 0
  %a1 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 1
  %a2 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 2
  store i32 0, i32 addrspace(5)* %a0
  store i32 1, i32 addrspace(5)* %a1
  store i32 2, i32 addrspace(5)* %a2
  %tmp = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i64 0, i64 %index
  %data = load atomic i32, i32 addrspace(5)* %tmp acquire, align 4
  store i32 %data, i32 addrspace(1)* %out
  ret void
}

; OPT-LABEL: @vector_alloca_atomic_write(
;
; OPT: alloca [3 x i32]
; OPT: store atomic i32 0, i32 addrspace(5)
; OPT: store atomic i32 1, i32 addrspace(5)
; OPT: store atomic i32 2, i32 addrspace(5)
; OPT: load i32, i32 addrspace(5)*
define amdgpu_kernel void @vector_alloca_atomic_write(i32 addrspace(1)* %out, i64 %index) {
entry:
  %alloca = alloca [3 x i32], addrspace(5)
  %a0 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 0
  %a1 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 1
  %a2 = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i32 0, i32 2
  store atomic i32 0, i32 addrspace(5)* %a0 release, align 4
  store atomic i32 1, i32 addrspace(5)* %a1 release, align 4
  store atomic i32 2, i32 addrspace(5)* %a2  release, align 4
  %tmp = getelementptr [3 x i32], [3 x i32] addrspace(5)* %alloca, i64 0, i64 %index
  %data = load i32, i32 addrspace(5)* %tmp
  store i32 %data, i32 addrspace(1)* %out
  ret void
}