mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2025-01-31 20:51:52 +01:00
ab6e036531
Summary: This is related to the changes to the groupstaticsize intrinsic in D61494 which would otherwise make the related tests in these files fail or much less useful. Note that for some reason, SOPK generation is less effective in the amdhsa OS, which is why I chose PAL. I haven't investigated this deeper. Change-Id: I6bb99569338f7a433c28b4c9eb1e3e036b00d166 Reviewers: arsenm Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D63392 llvm-svn: 363600
58 lines
1.8 KiB
LLVM
58 lines
1.8 KiB
LLVM
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
|
|
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
|
|
|
|
; SI-LABEL: {{^}}s_mulk_i32_k0:
|
|
; SI: s_load_dword [[VAL:s[0-9]+]]
|
|
; SI: s_mulk_i32 [[VAL]], 0x41
|
|
; SI: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[VAL]]
|
|
; SI: buffer_store_dword [[VRESULT]]
|
|
; SI: s_endpgm
|
|
define amdgpu_kernel void @s_mulk_i32_k0(i32 addrspace(1)* %out, i32 %b) {
|
|
%mul = mul i32 %b, 65
|
|
store i32 %mul, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}s_mulk_i32_k1:
|
|
; SI: s_mulk_i32 {{s[0-9]+}}, 0x7fff{{$}}
|
|
; SI: s_endpgm
|
|
define amdgpu_kernel void @s_mulk_i32_k1(i32 addrspace(1)* %out, i32 %b) {
|
|
%mul = mul i32 %b, 32767 ; (1 << 15) - 1
|
|
store i32 %mul, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}s_mulk_i32_k2:
|
|
; SI: s_mulk_i32 {{s[0-9]+}}, 0xffef{{$}}
|
|
; SI: s_endpgm
|
|
define amdgpu_kernel void @s_mulk_i32_k2(i32 addrspace(1)* %out, i32 %b) {
|
|
%mul = mul i32 %b, -17
|
|
store i32 %mul, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}no_s_mulk_i32_k0:
|
|
; SI: s_mul_i32 {{s[0-9]+}}, {{s[0-9]+}}, 0x8001{{$}}
|
|
; SI: s_endpgm
|
|
define amdgpu_kernel void @no_s_mulk_i32_k0(i32 addrspace(1)* %out, i32 %b) {
|
|
%mul = mul i32 %b, 32769 ; 1 << 15 + 1
|
|
store i32 %mul, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
@lds = addrspace(3) global [512 x i32] undef, align 4
|
|
|
|
; SI-LABEL: {{^}}commute_s_mulk_i32:
|
|
; SI: s_mulk_i32 s{{[0-9]+}}, 0x800{{$}}
|
|
define amdgpu_kernel void @commute_s_mulk_i32(i32 addrspace(1)* %out, i32 %b) #0 {
|
|
%size = call i32 @llvm.amdgcn.groupstaticsize()
|
|
%add = mul i32 %size, %b
|
|
call void asm sideeffect "; foo $0, $1", "v,s"([512 x i32] addrspace(3)* @lds, i32 %add)
|
|
ret void
|
|
}
|
|
|
|
declare i32 @llvm.amdgcn.groupstaticsize() #1
|
|
|
|
attributes #0 = { nounwind }
|
|
attributes #1 = { nounwind readnone }
|