[llvm] r245764 - AMDGPU: Minor cleanups to help with f16 support
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Fri Aug 21 16:49:52 PDT 2015
Author: arsenm
Date: Fri Aug 21 18:49:51 2015
New Revision: 245764
URL: http://llvm.org/viewvc/llvm-project?rev=245764&view=rev
Log:
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.
Modified:
llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td
Modified: llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td?rev=245764&r1=245763&r2=245764&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td (original)
+++ llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td Fri Aug 21 18:49:51 2015
@@ -965,28 +965,30 @@ class getNumSrcArgs<ValueType Src1, Valu
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]>;
More information about the llvm-commits
mailing list