Diana Picus
bc6955f18c
At the moment, the emergency spill slot is a fixed object for entry functions and chain functions, and a regular stack object otherwise. This patch adopts the latter behaviour for entry/chain functions too. It seems this was always the intention [1] and it will also save us a bit of stack space in cases where the first stack object has a large alignment. [1] https://github.com/llvm/llvm-project/commit/34c8b835b16fb3879f1b9770e91df21883356bb6
45 lines
1.2 KiB
LLVM
45 lines
1.2 KiB
LLVM
; RUN: llc -O0 -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck %s
|
|
|
|
; Test that the alignment of kernel arguments does not impact the
|
|
; alignment of the stack
|
|
|
|
; CHECK-LABEL: {{^}}no_args:
|
|
; CHECK: ScratchSize: 8{{$}}
|
|
define amdgpu_kernel void @no_args() {
|
|
%alloca = alloca i8, addrspace(5)
|
|
store volatile i8 0, ptr addrspace(5) %alloca
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: {{^}}force_align32:
|
|
; CHECK: ScratchSize: 8{{$}}
|
|
define amdgpu_kernel void @force_align32(<8 x i32>) {
|
|
%alloca = alloca i8, addrspace(5)
|
|
store volatile i8 0, ptr addrspace(5) %alloca
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: {{^}}force_align64:
|
|
; CHECK: ScratchSize: 8{{$}}
|
|
define amdgpu_kernel void @force_align64(<16 x i32>) {
|
|
%alloca = alloca i8, addrspace(5)
|
|
store volatile i8 0, ptr addrspace(5) %alloca
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: {{^}}force_align128:
|
|
; CHECK: ScratchSize: 8{{$}}
|
|
define amdgpu_kernel void @force_align128(<32 x i32>) {
|
|
%alloca = alloca i8, addrspace(5)
|
|
store volatile i8 0, ptr addrspace(5) %alloca
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: {{^}}force_align256:
|
|
; CHECK: ScratchSize: 8{{$}}
|
|
define amdgpu_kernel void @force_align256(<64 x i32>) {
|
|
%alloca = alloca i8, addrspace(5)
|
|
store volatile i8 0, ptr addrspace(5) %alloca
|
|
ret void
|
|
}
|