1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00
llvm-mirror/test/CodeGen/AMDGPU/v1024.ll
Stanislav Mekhanoshin f7a5ad2308 [AMDGPU] Change register type for v32 vectors
When it is AReg_1024 this results in unnecessary copying into
AGPRs of a 32 element vectors even though they are not intended
for an mfma instruction.

Differential Revision: https://reviews.llvm.org/D64815

llvm-svn: 366252
2019-07-16 20:06:00 +00:00

30 lines
1.2 KiB
LLVM

; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
; Check that we do not use AGPRs for v32i32 type
; GCN-LABEL: {{^}}test_v1024:
; GCN-NOT: v_accvgpr
; GCN-COUNT-32: v_mov_b32_e32
; GCN-NOT: v_accvgpr
define amdgpu_kernel void @test_v1024() {
entry:
%alloca = alloca <32 x i32>, align 16, addrspace(5)
%cast = bitcast <32 x i32> addrspace(5)* %alloca to i8 addrspace(5)*
br i1 undef, label %if.then.i.i, label %if.else.i
if.then.i.i: ; preds = %entry
call void @llvm.memcpy.p5i8.p5i8.i64(i8 addrspace(5)* align 16 %cast, i8 addrspace(5)* align 4 undef, i64 128, i1 false)
br label %if.then.i62.i
if.else.i: ; preds = %entry
br label %if.then.i62.i
if.then.i62.i: ; preds = %if.else.i, %if.then.i.i
call void @llvm.memcpy.p1i8.p5i8.i64(i8 addrspace(1)* align 4 undef, i8 addrspace(5)* align 16 %cast, i64 128, i1 false)
ret void
}
declare void @llvm.memcpy.p5i8.p5i8.i64(i8 addrspace(5)* nocapture writeonly, i8 addrspace(5)* nocapture readonly, i64, i1 immarg)
declare void @llvm.memcpy.p1i8.p5i8.i64(i8 addrspace(1)* nocapture writeonly, i8 addrspace(5)* nocapture readonly, i64, i1 immarg)