diff options
Diffstat (limited to 'test/CodeGen/R600/indirect-private-64.ll')
| -rw-r--r-- | test/CodeGen/R600/indirect-private-64.ll | 60 |
1 files changed, 43 insertions, 17 deletions
diff --git a/test/CodeGen/R600/indirect-private-64.ll b/test/CodeGen/R600/indirect-private-64.ll index b127b7e..e0a6ce1 100644 --- a/test/CodeGen/R600/indirect-private-64.ll +++ b/test/CodeGen/R600/indirect-private-64.ll @@ -1,10 +1,16 @@ -; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s +; RUN: llc -march=r600 -mcpu=SI -mattr=-promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=SI-ALLOCA -check-prefix=SI %s +; RUN: llc -march=r600 -mcpu=SI -mattr=+promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=SI-PROMOTE -check-prefix=SI %s + declare void @llvm.AMDGPU.barrier.local() noduplicate nounwind -; SI-LABEL: @private_access_f64_alloca: -; SI: DS_WRITE_B64 -; SI: DS_READ_B64 +; SI-LABEL: {{^}}private_access_f64_alloca: + +; SI-ALLOCA: buffer_store_dwordx2 +; SI-ALLOCA: buffer_load_dwordx2 + +; SI-PROMOTE: ds_write_b64 +; SI-PROMOTE: ds_read_b64 define void @private_access_f64_alloca(double addrspace(1)* noalias %out, double addrspace(1)* noalias %in, i32 %b) nounwind { %val = load double addrspace(1)* %in, align 8 %array = alloca double, i32 16, align 8 @@ -16,11 +22,19 @@ define void @private_access_f64_alloca(double addrspace(1)* noalias %out, double ret void } -; SI-LABEL: @private_access_v2f64_alloca: -; SI: DS_WRITE_B64 -; SI: DS_WRITE_B64 -; SI: DS_READ_B64 -; SI: DS_READ_B64 +; SI-LABEL: {{^}}private_access_v2f64_alloca: + +; SI-ALLOCA: buffer_store_dwordx4 +; SI-ALLOCA: buffer_load_dwordx4 + +; SI-PROMOTE: ds_write_b32 +; SI-PROMOTE: ds_write_b32 +; SI-PROMOTE: ds_write_b32 +; SI-PROMOTE: ds_write_b32 +; SI-PROMOTE: ds_read_b32 +; SI-PROMOTE: ds_read_b32 +; SI-PROMOTE: ds_read_b32 +; SI-PROMOTE: ds_read_b32 define void @private_access_v2f64_alloca(<2 x double> addrspace(1)* noalias %out, <2 x double> addrspace(1)* noalias %in, i32 %b) nounwind { %val = load <2 x double> addrspace(1)* %in, align 16 %array = alloca <2 x double>, i32 16, align 16 @@ -32,9 +46,13 @@ define void @private_access_v2f64_alloca(<2 x double> addrspace(1)* noalias %out ret void } -; SI-LABEL: @private_access_i64_alloca: -; SI: DS_WRITE_B64 -; SI: DS_READ_B64 +; SI-LABEL: {{^}}private_access_i64_alloca: + +; SI-ALLOCA: buffer_store_dwordx2 +; SI-ALLOCA: buffer_load_dwordx2 + +; SI-PROMOTE: ds_write_b64 +; SI-PROMOTE: ds_read_b64 define void @private_access_i64_alloca(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in, i32 %b) nounwind { %val = load i64 addrspace(1)* %in, align 8 %array = alloca i64, i32 16, align 8 @@ -46,11 +64,19 @@ define void @private_access_i64_alloca(i64 addrspace(1)* noalias %out, i64 addrs ret void } -; SI-LABEL: @private_access_v2i64_alloca: -; SI: DS_WRITE_B64 -; SI: DS_WRITE_B64 -; SI: DS_READ_B64 -; SI: DS_READ_B64 +; SI-LABEL: {{^}}private_access_v2i64_alloca: + +; SI-ALLOCA: buffer_store_dwordx4 +; SI-ALLOCA: buffer_load_dwordx4 + +; SI-PROMOTE: ds_write_b32 +; SI-PROMOTE: ds_write_b32 +; SI-PROMOTE: ds_write_b32 +; SI-PROMOTE: ds_write_b32 +; SI-PROMOTE: ds_read_b32 +; SI-PROMOTE: ds_read_b32 +; SI-PROMOTE: ds_read_b32 +; SI-PROMOTE: ds_read_b32 define void @private_access_v2i64_alloca(<2 x i64> addrspace(1)* noalias %out, <2 x i64> addrspace(1)* noalias %in, i32 %b) nounwind { %val = load <2 x i64> addrspace(1)* %in, align 16 %array = alloca <2 x i64>, i32 16, align 16 |