R600/SI: Add basic support for more integer vector types.
v1i32, v2i32, v8i32 and v16i32.
Only add VGPR register classes for integer vector types, to avoid attempts
copying from VGPR to SGPR registers, which is not possible.
Patch By: Michel Dänzer
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 174632
diff --git a/llvm/lib/Target/R600/SIInstructions.td b/llvm/lib/Target/R600/SIInstructions.td
index d33e113..dd779cf 100644
--- a/llvm/lib/Target/R600/SIInstructions.td
+++ b/llvm/lib/Target/R600/SIInstructions.td
@@ -584,7 +584,7 @@
//defm V_CVT_F32_U32 : VOP1_32 <0x00000006, "V_CVT_F32_U32", []>;
//defm V_CVT_U32_F32 : VOP1_32 <0x00000007, "V_CVT_U32_F32", []>;
defm V_CVT_I32_F32 : VOP1_32 <0x00000008, "V_CVT_I32_F32",
- [(set VReg_32:$dst, (fp_to_sint AllReg_32:$src0))]
+ [(set (i32 VReg_32:$dst), (fp_to_sint AllReg_32:$src0))]
>;
defm V_MOV_FED_B32 : VOP1_32 <0x00000009, "V_MOV_FED_B32", []>;
////def V_CVT_F16_F32 : VOP1_F16 <0x0000000a, "V_CVT_F16_F32", []>;
@@ -1000,17 +1000,17 @@
//def S_CBRANCH_G_FORK : SOP2_ <0x0000002b, "S_CBRANCH_G_FORK", []>;
def S_ABSDIFF_I32 : SOP2_32 <0x0000002c, "S_ABSDIFF_I32", []>;
-class V_MOV_IMM <Operand immType, SDNode immNode> : InstSI <
+class V_MOV_IMM <ValueType type, Operand immType, SDNode immNode> : InstSI <
(outs VReg_32:$dst),
(ins immType:$src0),
"V_MOV_IMM",
- [(set VReg_32:$dst, (immNode:$src0))]
+ [(set VReg_32:$dst, (type immNode:$src0))]
>;
let isCodeGenOnly = 1, isPseudo = 1 in {
-def V_MOV_IMM_I32 : V_MOV_IMM<i32imm, imm>;
-def V_MOV_IMM_F32 : V_MOV_IMM<f32imm, fpimm>;
+def V_MOV_IMM_I32 : V_MOV_IMM<i32, i32imm, imm>;
+def V_MOV_IMM_F32 : V_MOV_IMM<f32, f32imm, fpimm>;
def S_MOV_IMM_I32 : InstSI <
(outs SReg_32:$dst),
@@ -1227,8 +1227,12 @@
def : Insert_Element <f32, v4f32, VReg_32, VReg_128, 6, sub2>;
def : Insert_Element <f32, v4f32, VReg_32, VReg_128, 7, sub3>;
+def : Vector1_Build <v1i32, VReg_32, i32, VReg_32>;
+def : Vector2_Build <v2i32, VReg_64, i32, VReg_32>;
def : Vector_Build <v4f32, VReg_128, f32, VReg_32>;
-def : Vector_Build <v4i32, SReg_128, i32, SReg_32>;
+def : Vector_Build <v4i32, VReg_128, i32, VReg_32>;
+def : Vector8_Build <v8i32, VReg_256, i32, VReg_32>;
+def : Vector16_Build <v16i32, VReg_512, i32, VReg_32>;
def : BitConvert <i32, f32, SReg_32>;
def : BitConvert <i32, f32, VReg_32>;