R600/SI: Allow conversion between v32i8 and v8i32

Patch by: Marek Olšák Signed-off-by: Marek Olšák <[email protected]> git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188420 91177308-0d34-0410-b5e6-96231b3b80d8
ornata · Aug 14, 2013 · a7b7ab3 · a7b7ab3
1 parent 9735dc6
commit a7b7ab3
Show file tree

Hide file tree

Showing 3 changed files with 28 additions and 2 deletions.
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
@@ -1510,6 +1510,11 @@ def : BitConvert <v2i32, v2f32, VReg_64>;
 def : BitConvert <v4f32, v4i32, VReg_128>;
 def : BitConvert <v4i32, v4f32, VReg_128>;
 
+def : BitConvert <v8i32, v32i8, SReg_256>;
+def : BitConvert <v32i8, v8i32, SReg_256>;
+def : BitConvert <v8i32, v32i8, VReg_256>;
+def : BitConvert <v32i8, v8i32, VReg_256>;
+
 /********** =================== **********/
 /********** Src & Dst modifiers **********/
 /********** =================== **********/

diff --git a/lib/Target/R600/SIRegisterInfo.td b/lib/Target/R600/SIRegisterInfo.td
@@ -159,7 +159,7 @@ def SReg_64 : RegisterClass<"AMDGPU", [v2i32, i64, i1], 64,
 
 def SReg_128 : RegisterClass<"AMDGPU", [v16i8, i128], 128, (add SGPR_128)>;
 
-def SReg_256 : RegisterClass<"AMDGPU", [v32i8], 256, (add SGPR_256)>;
+def SReg_256 : RegisterClass<"AMDGPU", [v32i8, v8i32, v8f32], 256, (add SGPR_256)>;
 
 def SReg_512 : RegisterClass<"AMDGPU", [v64i8], 512, (add SGPR_512)>;
 
@@ -174,7 +174,7 @@ def VReg_96 : RegisterClass<"AMDGPU", [untyped], 96, (add VGPR_96)> {
 
 def VReg_128 : RegisterClass<"AMDGPU", [v4i32, v4f32], 128, (add VGPR_128)>;
 
-def VReg_256 : RegisterClass<"AMDGPU", [v8i32, v8f32], 256, (add VGPR_256)>;
+def VReg_256 : RegisterClass<"AMDGPU", [v32i8, v8i32, v8f32], 256, (add VGPR_256)>;
 
 def VReg_512 : RegisterClass<"AMDGPU", [v16i32, v16f32], 512, (add VGPR_512)>;
 

diff --git a/test/CodeGen/R600/bitcast.ll b/test/CodeGen/R600/bitcast.ll
@@ -0,0 +1,21 @@
+; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck %s
+
+; This test just checks that the compiler doesn't crash.
+; CHECK-LABEL: @v32i8_to_v8i32
+; CHECK: S_ENDPGM
+
+define void @v32i8_to_v8i32(<32 x i8> addrspace(2)* inreg) #0 {
+entry:
+  %1 = load <32 x i8> addrspace(2)* %0
+  %2 = bitcast <32 x i8> %1 to <8 x i32>
+  %3 = extractelement <8 x i32> %2, i32 1
+  %4 = icmp ne i32 %3, 0
+  %5 = select i1 %4, float 0.0, float 1.0
+  call void @llvm.SI.export(i32 15, i32 1, i32 1, i32 0, i32 1, float %5, float %5, float %5, float %5)
+  ret void
+}
+
+declare void @llvm.SI.export(i32, i32, i32, i32, i32, float, float, float, float)
+
+attributes #0 = { "ShaderType"="0" }
+