mirror of
https://github.com/RPCS3/llvm.git
synced 2026-08-28 05:10:05 -04:00
Matt Arsenault
e536396218
------------------------------------------------------------------------ r292982 | arsenm | 2017-01-24 14:02:15 -0800 (Tue, 24 Jan 2017) | 8 lines Enable FeatureFlatForGlobal on Volcanic Islands This switches to the workaround that HSA defaults to for the mesa path. This should be applied to the 4.0 branch. Patch by Vedran Miletić <vedran@miletic.net> ------------------------------------------------------------------------ git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_40@293326 91177308-0d34-0410-b5e6-96231b3b80d8
36 lines
1.5 KiB
LLVM
36 lines
1.5 KiB
LLVM
;RUN: llc < %s -march=amdgcn -mcpu=verde -mattr=+vgpr-spilling -mattr=-promote-alloca -verify-machineinstrs | FileCheck %s -check-prefix=CHECK
|
|
;RUN: llc < %s -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -mattr=+vgpr-spilling -mattr=-promote-alloca -verify-machineinstrs | FileCheck %s -check-prefix=CHECK
|
|
|
|
; Allocate two stack slots of 2052 bytes each requiring a total of 4104 bytes.
|
|
; Extracting the last element of each does not fit into the offset field of
|
|
; MUBUF instructions, so a new base register is needed. This used to not
|
|
; happen, leading to an assertion.
|
|
|
|
; CHECK-LABEL: {{^}}main:
|
|
; CHECK: buffer_store_dword
|
|
; CHECK: buffer_store_dword
|
|
; CHECK: buffer_load_dword
|
|
; CHECK: buffer_load_dword
|
|
define amdgpu_gs float @main(float %v1, float %v2, i32 %idx1, i32 %idx2) {
|
|
main_body:
|
|
%m1 = alloca [513 x float]
|
|
%m2 = alloca [513 x float]
|
|
|
|
%gep1.store = getelementptr [513 x float], [513 x float]* %m1, i32 0, i32 %idx1
|
|
store float %v1, float* %gep1.store
|
|
|
|
%gep2.store = getelementptr [513 x float], [513 x float]* %m2, i32 0, i32 %idx2
|
|
store float %v2, float* %gep2.store
|
|
|
|
; This used to use a base reg equal to 0.
|
|
%gep1.load = getelementptr [513 x float], [513 x float]* %m1, i32 0, i32 0
|
|
%out1 = load float, float* %gep1.load
|
|
|
|
; This used to attempt to re-use the base reg at 0, generating an out-of-bounds instruction offset.
|
|
%gep2.load = getelementptr [513 x float], [513 x float]* %m2, i32 0, i32 512
|
|
%out2 = load float, float* %gep2.load
|
|
|
|
%r = fadd float %out1, %out2
|
|
ret float %r
|
|
}
|