1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
llvm-mirror/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll
Matt Arsenault bf9d9a1180 AMDGPU: Switch backend default max workgroup size to 1024
Previously this would default to 256, not the maximum supported size
of 1024. Using a maximum lower than the hardware maximum requires
language runtimes to enforce this limit for correctness, which no
language has correctly done. Switch the default to the conservatively
correct maximum, and force frontends to opt-in to the more optimal 256
default maximum.

I don't really understand why the changes in occupancy-levels.ll
increased the computed occupancy, which I expected to decrease. I'm
not sure if these tests should be forcing the old maximum.
2019-11-13 07:11:02 +05:30

68 lines
3.5 KiB
LLVM

; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck %s
; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca -disable-promote-alloca-to-lds< %s | FileCheck -check-prefix=NOLDS %s
; This normally would be fixed by instcombine to be compare to the GEP
; indices
; NOLDS-NOT: addrspace(3)
; CHECK-LABEL: @lds_promoted_alloca_icmp_same_derived_pointer(
; CHECK: [[ARRAYGEP:%[0-9]+]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_same_derived_pointer.alloca, i32 0, i32 %{{[0-9]+}}
; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %a
; CHECK: %ptr1 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %b
; CHECK: %cmp = icmp eq i32 addrspace(3)* %ptr0, %ptr1
define amdgpu_kernel void @lds_promoted_alloca_icmp_same_derived_pointer(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
%alloca = alloca [16 x i32], align 4
%ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
%ptr1 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %b
%cmp = icmp eq i32* %ptr0, %ptr1
%zext = zext i1 %cmp to i32
store volatile i32 %zext, i32 addrspace(1)* %out
ret void
}
; CHECK-LABEL: @lds_promoted_alloca_icmp_null_rhs(
; CHECK: [[ARRAYGEP:%[0-9]+]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_null_rhs.alloca, i32 0, i32 %{{[0-9]+}}
; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %a
; CHECK: %cmp = icmp eq i32 addrspace(3)* %ptr0, null
define amdgpu_kernel void @lds_promoted_alloca_icmp_null_rhs(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
%alloca = alloca [16 x i32], align 4
%ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
%cmp = icmp eq i32* %ptr0, null
%zext = zext i1 %cmp to i32
store volatile i32 %zext, i32 addrspace(1)* %out
ret void
}
; CHECK-LABEL: @lds_promoted_alloca_icmp_null_lhs(
; CHECK: [[ARRAYGEP:%[0-9]+]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_null_lhs.alloca, i32 0, i32 %{{[0-9]+}}
; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %a
; CHECK: %cmp = icmp eq i32 addrspace(3)* null, %ptr0
define amdgpu_kernel void @lds_promoted_alloca_icmp_null_lhs(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
%alloca = alloca [16 x i32], align 4
%ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
%cmp = icmp eq i32* null, %ptr0
%zext = zext i1 %cmp to i32
store volatile i32 %zext, i32 addrspace(1)* %out
ret void
}
; CHECK-LABEL: @lds_promoted_alloca_icmp_unknown_ptr(
; CHECK: %alloca = alloca [16 x i32], align 4
; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
; CHECK: %ptr1 = call i32* @get_unknown_pointer()
; CHECK: %cmp = icmp eq i32* %ptr0, %ptr1
define amdgpu_kernel void @lds_promoted_alloca_icmp_unknown_ptr(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
%alloca = alloca [16 x i32], align 4
%ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a
%ptr1 = call i32* @get_unknown_pointer()
%cmp = icmp eq i32* %ptr0, %ptr1
%zext = zext i1 %cmp to i32
store volatile i32 %zext, i32 addrspace(1)* %out
ret void
}
declare i32* @get_unknown_pointer() #0
attributes #0 = { nounwind "amdgpu-waves-per-eu"="1,1" "amdgpu-flat-work-group-size"="1,256" }