[clang] [llvm] [HLSL] implement elementwise firstbithigh hlsl builtin (PR #111082)
Farzon Lotfi via cfe-commits
cfe-commits at lists.llvm.org
Wed Oct 30 13:06:55 PDT 2024
================
@@ -2626,6 +2671,151 @@ Register SPIRVInstructionSelector::buildPointerToResource(
MIRBuilder);
}
+bool SPIRVInstructionSelector::selectFirstBitHigh16(Register ResVReg,
+ const SPIRVType *ResType,
+ MachineInstr &I,
+ bool IsSigned) const {
+ unsigned Opcode = IsSigned ? SPIRV::OpSConvert : SPIRV::OpUConvert;
+ // zero or sign extend
+ Register ExtReg = MRI->createVirtualRegister(GR.getRegClass(ResType));
+ bool Result =
+ selectUnOpWithSrc(ExtReg, ResType, I, I.getOperand(2).getReg(), Opcode);
+ return Result & selectFirstBitHigh32(ResVReg, ResType, I, ExtReg, IsSigned);
+}
+
+bool SPIRVInstructionSelector::selectFirstBitHigh32(Register ResVReg,
+ const SPIRVType *ResType,
+ MachineInstr &I,
+ Register SrcReg,
+ bool IsSigned) const {
+ unsigned Opcode = IsSigned ? GL::FindSMsb : GL::FindUMsb;
+ return BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(SPIRV::OpExtInst))
----------------
farzonl wrote:
Nvmd upon looking only llvm IR has a `setOperand` I don't see one in MIR so this is fine.
https://github.com/llvm/llvm-project/pull/111082
More information about the cfe-commits
mailing list