mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-26 12:43:36 +01:00
dd9ab77318
Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
29 lines
1.1 KiB
LLVM
29 lines
1.1 KiB
LLVM
; RUN: llc < %s -march=r600 -mcpu=redwood -verify-machineinstrs | FileCheck %s
|
|
|
|
; The test is for a bug in R600EmitClauseMarkers.cpp where this pass
|
|
; was searching for a use of the OQAP register in order to determine
|
|
; if an LDS instruction could fit in the current clause, but never finding
|
|
; one. This created an infinite loop and hung the compiler.
|
|
;
|
|
; The LDS instruction should not have been defining OQAP in the first place,
|
|
; because the LDS instructions are pseudo instructions and the OQAP
|
|
; reads and writes are bundled together in the same instruction.
|
|
|
|
; CHECK: {{^}}lds_crash:
|
|
define amdgpu_kernel void @lds_crash(i32 addrspace(1)* %out, i32 addrspace(3)* %in, i32 %a, i32 %b, i32 %c) {
|
|
entry:
|
|
%0 = load i32, i32 addrspace(3)* %in
|
|
; This block needs to be > 115 ISA instructions to hit the bug,
|
|
; so we'll use udiv instructions.
|
|
%div0 = udiv i32 %0, %b
|
|
%div1 = udiv i32 %div0, %a
|
|
%div2 = udiv i32 %div1, 11
|
|
%div3 = udiv i32 %div2, %a
|
|
%div4 = udiv i32 %div3, %b
|
|
%div5 = udiv i32 %div4, %c
|
|
%div6 = udiv i32 %div5, %div0
|
|
%div7 = udiv i32 %div6, %div1
|
|
store i32 %div7, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|