[llvm] [SPIR-V] Implement SPV_KHR_float_controls2 (PR #146941)

[Nathan Gauër via llvm-commits]

================
@@ -606,6 +664,73 @@ void SPIRVAsmPrinter::outputAnnotations(const Module &M) {
   }
 }
 
+void SPIRVAsmPrinter::outputFPFastMathDefaultInfo() {
+  // Collect the SPIRVTypes that are OpTypeFloat.
+  std::vector<const MachineInstr *> SPIRVFloatTypes;
+  for (const MachineInstr *MI : MAI->getMSInstrs(SPIRV::MB_TypeConstVars)) {
+    // Skip if the instruction is not OpTypeFloat.
+    if (MI->getOpcode() != SPIRV::OpTypeFloat)
+      continue;
+
+    // Collect the SPIRV type.
+    SPIRVFloatTypes.push_back(MI);
+  }
+
+  for (const auto &[Func, FPFastMathDefaultInfoVec] :
+       MAI->FPFastMathDefaultInfoMap) {
+    if (FPFastMathDefaultInfoVec.empty())
+      continue;
+
+    for (const MachineInstr *MI : SPIRVFloatTypes) {
+      unsigned OpTypeFloatSize = MI->getOperand(1).getImm();
+      unsigned Index = computeFPFastMathDefaultInfoVecIndex(OpTypeFloatSize);
+      assert(Index < FPFastMathDefaultInfoVec.size() &&
+             "Index out of bounds for FPFastMathDefaultInfoVec");
+      const auto &FPFastMathDefaultInfo = FPFastMathDefaultInfoVec[Index];
+      assert(FPFastMathDefaultInfo.Ty &&
+             "Expected target type for FPFastMathDefaultInfo");
+      assert(FPFastMathDefaultInfo.Ty->getScalarSizeInBits() ==
+                 OpTypeFloatSize &&
+             "Mismatched float type size");
+      MCInst Inst;
+      Inst.setOpcode(SPIRV::OpExecutionMode);
+      MCRegister FuncReg = MAI->getFuncReg(Func);
+      assert(FuncReg.isValid());
+      Inst.addOperand(MCOperand::createReg(FuncReg));
+      Inst.addOperand(
+          MCOperand::createImm(SPIRV::ExecutionMode::FPFastMathDefault));
+      MCRegister TypeReg =
+          MAI->getRegisterAlias(MI->getMF(), MI->getOperand(0).getReg());
+      Inst.addOperand(MCOperand::createReg(TypeReg));
+      unsigned Flags = FPFastMathDefaultInfo.FastMathFlags;
+      if (FPFastMathDefaultInfo.ContractionOff &&
+          (Flags & SPIRV::FPFastMathMode::AllowContract) &&
+          FPFastMathDefaultInfo.FPFastMathDefault)
----------------
Keenuts wrote:

Might be missing something.
`ContractionOff + FastMath = illegal` from the spec point of view. But as you mention when using `float_controls_2`, `ContractionOff` is not emitted. Instead the `FastMathDefault` mask is used. So If I get it, even if the spec states it's illegal, here we are fine since in the end we won't emit the `ContractionOff` decoration. We are just converting the old decorations to the new bitmask format.

If that's the case, seems the condition should then be independent of the result of `FPFastMathDefaultInfo.FPFastMathDefault` but be more `if we have both NoContraction and AllowContract bit is set, something is wrong`, independently from the `FPFastMathDefault`
And we might want to assert that if any bit is set in the FPFastMathDefaultInfo, `FPFastMathDefault` should be true no?

https://github.com/llvm/llvm-project/pull/146941