AMDGPU: Cleanup lowering actions

These are kind of a mess and hard to follow, particularly
for loads and stores. Fix various redundant, unnecessary
and dead settings.

llvm-svn: 270307

2 files changed, 131 insertions, 32 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/global-extload-i8.ll b/llvm/test/CodeGen/AMDGPU/global-extload-i8.ll
index b31d5361d5a..85739476944 100644
--- a/llvm/test/CodeGen/AMDGPU/global-extload-i8.ll
+++ b/llvm/test/CodeGen/AMDGPU/global-extload-i8.ll
@@ -1,4 +1,4 @@
-; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs< %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -verify-machineinstrs< %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs< %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -march=r600 -mcpu=cypress < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 
@@ -262,26 +262,26 @@ define void @sextload_global_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <16
   ret void
 }
 
-; XFUNC-LABEL: {{^}}zextload_global_v32i8_to_v32i64:
-; XSI: s_endpgm
-; define void @zextload_global_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i8> addrspace(1)* nocapture %in) nounwind {
-;   %load = load <32 x i8>, <32 x i8> addrspace(1)* %in
-;   %ext = zext <32 x i8> %load to <32 x i64>
-;   store <32 x i64> %ext, <32 x i64> addrspace(1)* %out
-;   ret void
-; }
+; FUNC-LABEL: {{^}}zextload_global_v32i8_to_v32i64:
+; SI: s_endpgm
+define void @zextload_global_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i8> addrspace(1)* nocapture %in) nounwind {
+  %load = load <32 x i8>, <32 x i8> addrspace(1)* %in
+  %ext = zext <32 x i8> %load to <32 x i64>
+  store <32 x i64> %ext, <32 x i64> addrspace(1)* %out
+  ret void
+}
 
-; XFUNC-LABEL: {{^}}sextload_global_v32i8_to_v32i64:
-; XSI: s_endpgm
-; define void @sextload_global_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i8> addrspace(1)* nocapture %in) nounwind {
-;   %load = load <32 x i8>, <32 x i8> addrspace(1)* %in
-;   %ext = sext <32 x i8> %load to <32 x i64>
-;   store <32 x i64> %ext, <32 x i64> addrspace(1)* %out
-;   ret void
-; }
+; FUNC-LABEL: {{^}}sextload_global_v32i8_to_v32i64:
+; SI: s_endpgm
+define void @sextload_global_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i8> addrspace(1)* nocapture %in) nounwind {
+  %load = load <32 x i8>, <32 x i8> addrspace(1)* %in
+  %ext = sext <32 x i8> %load to <32 x i64>
+  store <32 x i64> %ext, <32 x i64> addrspace(1)* %out
+  ret void
+}
 
-; XFUNC-LABEL: {{^}}zextload_global_v64i8_to_v64i64:
-; XSI: s_endpgm
+; ; XFUNC-LABEL: {{^}}zextload_global_v64i8_to_v64i64:
+; ; XSI: s_endpgm
 ; define void @zextload_global_v64i8_to_v64i64(<64 x i64> addrspace(1)* %out, <64 x i8> addrspace(1)* nocapture %in) nounwind {
 ;   %load = load <64 x i8>, <64 x i8> addrspace(1)* %in
 ;   %ext = zext <64 x i8> %load to <64 x i64>
@@ -289,8 +289,8 @@ define void @sextload_global_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <16
 ;   ret void
 ; }
 
-; XFUNC-LABEL: {{^}}sextload_global_v64i8_to_v64i64:
-; XSI: s_endpgm
+; ; XFUNC-LABEL: {{^}}sextload_global_v64i8_to_v64i64:
+; ; XSI: s_endpgm
 ; define void @sextload_global_v64i8_to_v64i64(<64 x i64> addrspace(1)* %out, <64 x i8> addrspace(1)* nocapture %in) nounwind {
 ;   %load = load <64 x i8>, <64 x i8> addrspace(1)* %in
 ;   %ext = sext <64 x i8> %load to <64 x i64>
diff --git a/llvm/test/CodeGen/AMDGPU/store-v3i64.ll b/llvm/test/CodeGen/AMDGPU/store-v3i64.ll
index e0c554ad2c1..b4d7505e0a8 100644
--- a/llvm/test/CodeGen/AMDGPU/store-v3i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/store-v3i64.ll
@@ -1,29 +1,128 @@
-; XFAIL: *
-; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs< %s | FileCheck -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs< %s | FileCheck -check-prefix=SI %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
+; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 
-; SI-LABEL: {{^}}global_store_v3i64:
-; SI: buffer_store_dwordx4
-; SI: buffer_store_dwordx4
+; GCN-LABEL: {{^}}global_store_v3i64:
+; GCN-DAG: buffer_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, off, s{{\[[0-9]+:[0-9]+\]}}, 0 offset:16
+; GCN-DAG: buffer_store_dwordx4 v{{\[[0-9]+:[0-9]+\]}}, off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
 define void @global_store_v3i64(<3 x i64> addrspace(1)* %out, <3 x i64> %x) {
   store <3 x i64> %x, <3 x i64> addrspace(1)* %out, align 32
   ret void
 }
 
-; SI-LABEL: {{^}}global_store_v3i64_unaligned:
+; GCN-LABEL: {{^}}global_store_v3i64_unaligned:
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
+; GCN: buffer_store_byte
 define void @global_store_v3i64_unaligned(<3 x i64> addrspace(1)* %out, <3 x i64> %x) {
   store <3 x i64> %x, <3 x i64> addrspace(1)* %out, align 1
   ret void
 }
 
-; SI-LABEL: {{^}}local_store_v3i64:
+; GCN-LABEL: {{^}}local_store_v3i64:
+; GCN: ds_write_b64
+; GCN: ds_write_b64
+; GCN: ds_write_b64
 define void @local_store_v3i64(<3 x i64> addrspace(3)* %out, <3 x i64> %x) {
   store <3 x i64> %x, <3 x i64> addrspace(3)* %out, align 32
   ret void
 }
 
-; SI-LABEL: {{^}}local_store_v3i64_unaligned:
-define void @local_store_v3i64_unaligned(<3 x i64> addrspace(1)* %out, <3 x i64> %x) {
-  store <3 x i64> %x, <3 x i64> addrspace(1)* %out, align 1
+; GCN-LABEL: {{^}}local_store_v3i64_unaligned:
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+; GCN: ds_write_b8
+define void @local_store_v3i64_unaligned(<3 x i64> addrspace(3)* %out, <3 x i64> %x) {
+  store <3 x i64> %x, <3 x i64> addrspace(3)* %out, align 1
+  ret void
+}
+
+; GCN-LABEL: {{^}}global_truncstore_v3i64_to_v3i32:
+; GCN-DAG: buffer_store_dwordx2
+; GCN-DAG: buffer_store_dword v
+define void @global_truncstore_v3i64_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i64> %x) {
+  %trunc = trunc <3 x i64> %x to <3 x i32>
+  store <3 x i32> %trunc, <3 x i32> addrspace(1)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}global_truncstore_v3i64_to_v3i16:
+; GCN-DAG: buffer_store_short
+; GCN-DAG: buffer_store_dword v
+define void @global_truncstore_v3i64_to_v3i16(<3 x i16> addrspace(1)* %out, <3 x i64> %x) {
+  %trunc = trunc <3 x i64> %x to <3 x i16>
+  store <3 x i16> %trunc, <3 x i16> addrspace(1)* %out
+  ret void
+}
+
+
+; GCN-LABEL: {{^}}global_truncstore_v3i64_to_v3i8:
+; GCN-DAG: buffer_store_short
+; GCN-DAG: buffer_store_byte v
+define void @global_truncstore_v3i64_to_v3i8(<3 x i8> addrspace(1)* %out, <3 x i64> %x) {
+  %trunc = trunc <3 x i64> %x to <3 x i8>
+  store <3 x i8> %trunc, <3 x i8> addrspace(1)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}global_truncstore_v3i64_to_v3i1:
+; GCN-DAG: buffer_store_byte v
+; GCN-DAG: buffer_store_byte v
+; GCN-DAG: buffer_store_byte v
+define void @global_truncstore_v3i64_to_v3i1(<3 x i1> addrspace(1)* %out, <3 x i64> %x) {
+  %trunc = trunc <3 x i64> %x to <3 x i1>
+  store <3 x i1> %trunc, <3 x i1> addrspace(1)* %out
   ret void
 }