maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

AMDGPU: Minor cleanups to help with f16 support 

The main change is inverting the condition for the
operand class classes so that VT.Size == 16 uses VGPR_32
instead of 64.

llvm-svn: 245764
This commit is contained in:
Matt Arsenault 2015-08-21 23:49:51 +00:00
parent cd5b8a0235
commit f56872dc30
1 changed files with 11 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
llvm/lib/Target/AMDGPU/SIInstrInfo.td
View File

@ -965,28 +965,30 @@ class getNumSrcArgs<ValueType Src1, ValueType Src2> {
class getVALUDstForVT<ValueType VT> {
RegisterOperand ret = !if(!eq(VT.Size, 32), VOPDstOperand<VGPR_32>,
!if(!eq(VT.Size, 64), VOPDstOperand<VReg_64>,
VOPDstOperand<SReg_64>)); // else VT == i1
!if(!eq(VT.Size, 16), VOPDstOperand<VGPR_32>,
VOPDstOperand<SReg_64>))); // else VT == i1
}
// Returns the register class to use for source 0 of VOP[12C]
// instructions for the given VT.
class getVOPSrc0ForVT<ValueType VT> {
RegisterOperand ret = !if(!eq(VT.Size, 32), VSrc_32, VSrc_64);
RegisterOperand ret = !if(!eq(VT.Size, 64), VSrc_64, VSrc_32);
}
// Returns the register class to use for source 1 of VOP[12C] for the
// given VT.
class getVOPSrc1ForVT<ValueType VT> {
RegisterClass ret = !if(!eq(VT.Size, 32), VGPR_32, VReg_64);
RegisterClass ret = !if(!eq(VT.Size, 64), VReg_64, VGPR_32);
}
// Returns the register class to use for sources of VOP3 instructions for the
// given VT.
class getVOP3SrcForVT<ValueType VT> {
RegisterOperand ret = !if(!eq(VT.Size, 32), VCSrc_32, VCSrc_64);
RegisterOperand ret = !if(!eq(VT.Size, 64), VCSrc_64, VCSrc_32);
}
// Returns 1 if the source arguments have modifiers, 0 if they do not.
// XXX - do f16 instructions?
class hasModifiers<ValueType SrcVT> {
bit ret = !if(!eq(SrcVT.Value, f32.Value), 1,
!if(!eq(SrcVT.Value, f64.Value), 1, 0));
@ -1093,12 +1095,12 @@ class VOPProfile <list<ValueType> _ArgVT> {
// FIXME: I think these F16/I16 profiles will need to use f16/i16 types in order
// for the instruction patterns to work.
def VOP_F16_F16 : VOPProfile <[f32, f32, untyped, untyped]>;
def VOP_F16_I16 : VOPProfile <[f32, i32, untyped, untyped]>;
def VOP_I16_F16 : VOPProfile <[i32, f32, untyped, untyped]>;
def VOP_F16_F16 : VOPProfile <[f16, f16, untyped, untyped]>;
def VOP_F16_I16 : VOPProfile <[f16, i32, untyped, untyped]>;
def VOP_I16_F16 : VOPProfile <[i32, f16, untyped, untyped]>;
def VOP_F16_F16_F16 : VOPProfile <[f32, f32, f32, untyped]>;
def VOP_F16_F16_I16 : VOPProfile <[f32, f32, i32, untyped]>;
def VOP_F16_F16_F16 : VOPProfile <[f16, f16, f16, untyped]>;
def VOP_F16_F16_I16 : VOPProfile <[f16, f16, i32, untyped]>;
def VOP_I16_I16_I16 : VOPProfile <[i32, i32, i32, untyped]>;
def VOP_F32_F32 : VOPProfile <[f32, f32, untyped, untyped]>;