R600/SI: Avoid generating REGISTER_LOAD instructions.

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

diff --git a/test/CodeGen/R600/private-memory.ll b/test/CodeGen/R600/private-memory.ll
index 22bd3589571166ec4b2f25a144e506198c2b2776..124d9fa645058c8eb35a673fd92dc4f251f020a9 100644 (file)
--- a/test/CodeGen/R600/private-memory.ll
+++ b/test/CodeGen/R600/private-memory.ll
@@ -1,19 +1,23 @@
-; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s --check-prefix=R600-CHECK --check-prefix=FUNC
-; RUN: llc -verify-machineinstrs -march=r600 -mcpu=SI < %s | FileCheck %s --check-prefix=SI-CHECK --check-prefix=FUNC
+; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC
+; RUN: llc -mattr=+promote-alloca -verify-machineinstrs -march=r600 -mcpu=SI < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC
+; RUN: llc -mattr=-promote-alloca -verify-machineinstrs -march=r600 -mcpu=SI < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
 
 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
 
 ; FUNC-LABEL: @mova_same_clause
 
-; R600-CHECK: LDS_WRITE
-; R600-CHECK: LDS_WRITE
-; R600-CHECK: LDS_READ
-; R600-CHECK: LDS_READ
+; R600: LDS_WRITE
+; R600: LDS_WRITE
+; R600: LDS_READ
+; R600: LDS_READ
 
-; SI-CHECK: DS_WRITE_B32
-; SI-CHECK: DS_WRITE_B32
-; SI-CHECK: DS_READ_B32
-; SI-CHECK: DS_READ_B32
+; SI-PROMOTE: DS_WRITE_B32
+; SI-PROMOTE: DS_WRITE_B32
+; SI-PROMOTE: DS_READ_B32
+; SI-PROMOTE: DS_READ_B32
+
+; SI-ALLOCA: BUFFER_STORE_DWORD v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s{{[0-9]+}}
+; SI-ALLOCA: BUFFER_STORE_DWORD v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s{{[0-9]+}}
 define void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
 entry:
   %stack = alloca [5 x i32], align 4
@@ -42,8 +46,9 @@ entry:
 ; this.
 
 ; FUNC-LABEL: @multiple_structs
-; R600-CHECK-NOT: MOVA_INT
-; SI-CHECK-NOT: V_MOVREL
+; R600-NOT: MOVA_INT
+; SI-NOT: V_MOVREL
+; SI-NOT: V_MOVREL
 %struct.point = type { i32, i32 }
 
 define void @multiple_structs(i32 addrspace(1)* %out) {
@@ -72,8 +77,8 @@ entry:
 ; MOVA instructions.
 
 ; FUNC-LABEL: @direct_loop
-; R600-CHECK-NOT: MOVA_INT
-; SI-CHECK-NOT: V_MOVREL
+; R600-NOT: MOVA_INT
+; SI-NOT: V_MOVREL
 
 define void @direct_loop(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
 entry:
@@ -109,9 +114,12 @@ for.end:
 
 ; FUNC-LABEL: @short_array
 
-; R600-CHECK: MOVA_INT
+; R600: MOVA_INT
 
-; SI-CHECK: V_MOVRELS_B32_e32
+; SI-PROMOTE: BUFFER_STORE_SHORT v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s{{[0-9]+}}
+; SI-PROMOTE: BUFFER_STORE_SHORT v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s{{[0-9]+}}
+; SI-PROMOTE-NOT: MOVREL
+; SI-PROMOTE: BUFFER_LOAD_SSHORT v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}] + v{{[0-9]+}} + s{{[0-9]+}}
 define void @short_array(i32 addrspace(1)* %out, i32 %index) {
 entry:
   %0 = alloca [2 x i16]
@@ -128,10 +136,10 @@ entry:
 
 ; FUNC-LABEL: @char_array
 
-; R600-CHECK: MOVA_INT
+; R600: MOVA_INT
 
-; SI-CHECK: V_OR_B32_e32 v{{[0-9]}}, 0x100
-; SI-CHECK: V_MOVRELS_B32_e32
+; SI-DAG: BUFFER_STORE_BYTE v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s{{[0-9]+}}, 0x0
+; SI-DAG: BUFFER_STORE_BYTE v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s{{[0-9]+}}, 0x1
 define void @char_array(i32 addrspace(1)* %out, i32 %index) {
 entry:
   %0 = alloca [2 x i8]
@@ -150,11 +158,11 @@ entry:
 ; Make sure we don't overwrite workitem information with private memory
 
 ; FUNC-LABEL: @work_item_info
-; R600-CHECK-NOT: MOV T0.X
+; R600-NOT: MOV T0.X
 ; Additional check in case the move ends up in the last slot
-; R600-CHECK-NOT: MOV * TO.X
+; R600-NOT: MOV * TO.X
 
-; SI-CHECK-NOT: V_MOV_B32_e{{(32|64)}} v0
+; SI-NOT: V_MOV_B32_e{{(32|64)}} v0
 define void @work_item_info(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0 = alloca [2 x i32]
@@ -175,8 +183,8 @@ entry:
 ; FUNC-LABEL: @no_overlap
 ; R600_CHECK: MOV
 ; R600_CHECK: [[CHAN:[XYZW]]]+
-; R600-CHECK-NOT: [[CHAN]]+
-; SI-CHECK: V_MOV_B32_e32 v3
+; R600-NOT: [[CHAN]]+
+; SI: V_MOV_B32_e32 v3
 define void @no_overlap(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0 = alloca [3 x i8], align 1
@@ -267,5 +275,19 @@ entry:
   %load = load i32* %gep2
   store i32 %load, i32 addrspace(1)* %out
   ret void
+}
 
+define void @select_private(i32 addrspace(1)* %out, i32 %in) nounwind {
+entry:
+  %tmp = alloca [2 x i32]
+  %tmp1 = getelementptr [2 x i32]* %tmp, i32 0, i32 0
+  %tmp2 = getelementptr [2 x i32]* %tmp, i32 0, i32 1
+  store i32 0, i32* %tmp1
+  store i32 1, i32* %tmp2
+  %cmp = icmp eq i32 %in, 0
+  %sel = select i1 %cmp, i32* %tmp1, i32* %tmp2
+  %load = load i32* %sel
+  store i32 %load, i32 addrspace(1)* %out
+  ret void
 }
+