2013-08-27 10:28:26 +00:00
|
|
|
; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG-CHECK %s
|
2014-08-22 18:49:33 +00:00
|
|
|
; RUN: llc < %s -march=r600 -mcpu=verde -verify-machineinstrs | FileCheck --check-prefix=SI-CHECK --check-prefix=SI %s
|
|
|
|
; RUN: llc < %s -march=r600 -mcpu=bonaire -verify-machineinstrs | FileCheck --check-prefix=SI-CHECK --check-prefix=CI %s
|
2013-07-10 16:37:07 +00:00
|
|
|
|
2014-06-06 01:20:28 +00:00
|
|
|
@local_memory_two_objects.local_mem0 = internal unnamed_addr addrspace(3) global [4 x i32] zeroinitializer, align 4
|
|
|
|
@local_memory_two_objects.local_mem1 = internal unnamed_addr addrspace(3) global [4 x i32] zeroinitializer, align 4
|
2013-07-10 16:37:07 +00:00
|
|
|
|
2014-10-01 17:15:17 +00:00
|
|
|
; EG-CHECK: {{^}}local_memory_two_objects:
|
2013-07-10 16:37:07 +00:00
|
|
|
|
|
|
|
; Check that the LDS size emitted correctly
|
2013-08-27 10:28:26 +00:00
|
|
|
; EG-CHECK: .long 166120
|
|
|
|
; EG-CHECK-NEXT: .long 8
|
|
|
|
; SI-CHECK: .long 47180
|
|
|
|
; SI-CHECK-NEXT: .long 32768
|
2013-07-10 16:37:07 +00:00
|
|
|
|
2013-11-15 00:12:45 +00:00
|
|
|
; We would like to check the the lds writes are using different
|
|
|
|
; addresses, but due to variations in the scheduler, we can't do
|
|
|
|
; this consistently on evergreen GPUs.
|
|
|
|
; EG-CHECK: LDS_WRITE
|
|
|
|
; EG-CHECK: LDS_WRITE
|
2014-11-05 14:50:53 +00:00
|
|
|
; SI-CHECK: ds_write_b32 {{v[0-9]*}}, v[[ADDRW:[0-9]*]]
|
|
|
|
; SI-CHECK-NOT: ds_write_b32 {{v[0-9]*}}, v[[ADDRW]]
|
2013-07-10 16:37:07 +00:00
|
|
|
|
|
|
|
; GROUP_BARRIER must be the last instruction in a clause
|
2013-08-27 10:28:26 +00:00
|
|
|
; EG-CHECK: GROUP_BARRIER
|
|
|
|
; EG-CHECK-NEXT: ALU clause
|
2013-07-10 16:37:07 +00:00
|
|
|
|
2014-03-19 22:19:49 +00:00
|
|
|
; Make sure the lds reads are using different addresses, at different
|
|
|
|
; constant offsets.
|
2013-08-27 10:28:26 +00:00
|
|
|
; EG-CHECK: LDS_READ_RET {{[*]*}} OQAP, {{PV|T}}[[ADDRR:[0-9]*\.[XYZW]]]
|
|
|
|
; EG-CHECK-NOT: LDS_READ_RET {{[*]*}} OQAP, T[[ADDRR]]
|
2014-11-05 14:50:53 +00:00
|
|
|
; SI: v_add_i32_e32 [[SIPTR:v[0-9]+]], 16, v{{[0-9]+}}
|
|
|
|
; SI: ds_read_b32 {{v[0-9]+}}, [[SIPTR]] [M0]
|
|
|
|
; CI: ds_read_b32 {{v[0-9]+}}, [[ADDRR:v[0-9]+]] offset:16 [M0]
|
|
|
|
; CI: ds_read_b32 {{v[0-9]+}}, [[ADDRR]] [M0]
|
2013-07-10 16:37:07 +00:00
|
|
|
|
|
|
|
define void @local_memory_two_objects(i32 addrspace(1)* %out) {
|
|
|
|
entry:
|
|
|
|
%x.i = call i32 @llvm.r600.read.tidig.x() #0
|
|
|
|
%arrayidx = getelementptr inbounds [4 x i32] addrspace(3)* @local_memory_two_objects.local_mem0, i32 0, i32 %x.i
|
|
|
|
store i32 %x.i, i32 addrspace(3)* %arrayidx, align 4
|
|
|
|
%mul = shl nsw i32 %x.i, 1
|
|
|
|
%arrayidx1 = getelementptr inbounds [4 x i32] addrspace(3)* @local_memory_two_objects.local_mem1, i32 0, i32 %x.i
|
|
|
|
store i32 %mul, i32 addrspace(3)* %arrayidx1, align 4
|
|
|
|
%sub = sub nsw i32 3, %x.i
|
|
|
|
call void @llvm.AMDGPU.barrier.local()
|
|
|
|
%arrayidx2 = getelementptr inbounds [4 x i32] addrspace(3)* @local_memory_two_objects.local_mem0, i32 0, i32 %sub
|
|
|
|
%0 = load i32 addrspace(3)* %arrayidx2, align 4
|
|
|
|
%arrayidx3 = getelementptr inbounds i32 addrspace(1)* %out, i32 %x.i
|
|
|
|
store i32 %0, i32 addrspace(1)* %arrayidx3, align 4
|
|
|
|
%arrayidx4 = getelementptr inbounds [4 x i32] addrspace(3)* @local_memory_two_objects.local_mem1, i32 0, i32 %sub
|
|
|
|
%1 = load i32 addrspace(3)* %arrayidx4, align 4
|
|
|
|
%add = add nsw i32 %x.i, 4
|
|
|
|
%arrayidx5 = getelementptr inbounds i32 addrspace(1)* %out, i32 %add
|
|
|
|
store i32 %1, i32 addrspace(1)* %arrayidx5, align 4
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
declare i32 @llvm.r600.read.tidig.x() #0
|
|
|
|
declare void @llvm.AMDGPU.barrier.local()
|
|
|
|
|
|
|
|
attributes #0 = { readnone }
|