R600/SI: Clean up some of the unused REGISTER_{LOAD,STORE} code

[oota-llvm.git] / test / CodeGen / R600 / indirect-private-64.ll

diff --git a/test/CodeGen/R600/indirect-private-64.ll b/test/CodeGen/R600/indirect-private-64.ll
index a6f1e932dbd444415a1303cc31625e85f3cd4472..5747434935b3ef9e3478fa18f1883a72531389d0 100644 (file)
--- a/test/CodeGen/R600/indirect-private-64.ll
+++ b/test/CodeGen/R600/indirect-private-64.ll
@@ -1,12 +1,17 @@
-; REQUIRES: asserts
-; XFAIL: *
-; 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: @indirect_access_f64_alloca:
-; SI: BUFFER_STORE_DWORD
-define void @f64_alloca(double addrspace(1)* noalias %out, double addrspace(1)* noalias %in, i32 %b) nounwind {
+; 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
   %ptr = getelementptr double* %array, i32 %b
@@ -17,9 +22,20 @@ define void @f64_alloca(double addrspace(1)* noalias %out, double addrspace(1)*
   ret void
 }
 
-; SI-LABEL: @indirect_access_v2f64_alloca:
-; SI: BUFFER_STORE_DWORDX4
-define void @v2f64_alloca(<2 x double> addrspace(1)* noalias %out, <2 x double> addrspace(1)* noalias %in, i32 %b) nounwind {
+; 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
   %ptr = getelementptr <2 x double>* %array, i32 %b
@@ -29,3 +45,45 @@ define void @v2f64_alloca(<2 x double> addrspace(1)* noalias %out, <2 x double>
   store <2 x double> %result, <2 x double> addrspace(1)* %out, align 16
   ret void
 }
+
+; 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
+  %ptr = getelementptr i64* %array, i32 %b
+  store i64 %val, i64* %ptr, align 8
+  call void @llvm.AMDGPU.barrier.local() noduplicate nounwind
+  %result = load i64* %ptr, align 8
+  store i64 %result, i64 addrspace(1)* %out, align 8
+  ret void
+}
+
+; 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
+  %ptr = getelementptr <2 x i64>* %array, i32 %b
+  store <2 x i64> %val, <2 x i64>* %ptr, align 16
+  call void @llvm.AMDGPU.barrier.local() noduplicate nounwind
+  %result = load <2 x i64>* %ptr, align 16
+  store <2 x i64> %result, <2 x i64> addrspace(1)* %out, align 16
+  ret void
+}