[X86] AMD Zen 3 sched model: FMA ops have inverse throughput of 0.5

Now that exegesis produces meaningful snippets to measure throughtput
for instructions with tied operands:
2ffe225
the measurements clearly show these instructions to have
more optimistic throughtput.

There's still some noise in the reports, especially around instructions
with memory operands. I'm not sure if we measure those correctly.

Fixes #59325

llvm/lib/Target/X86/X86ScheduleZnver3.td

@@ -937,9 +937,9 @@ defm : Zn3WriteResXMMPair<WriteFRsqrt, [Zn3FPFDiv], 3, [1], 1>;  // Floating poi
 defm : Zn3WriteResXMMPair<WriteFRsqrtX, [Zn3FPFDiv], 3, [1], 1>; // Floating point reciprocal square root estimate (XMM).
 defm : Zn3WriteResYMMPair<WriteFRsqrtY, [Zn3FPFDiv], 3, [1], 1>; // Floating point reciprocal square root estimate (YMM).
 defm : X86WriteResPairUnsupported<WriteFRsqrtZ>; // Floating point reciprocal square root estimate (ZMM).
-defm : Zn3WriteResXMMPair<WriteFMA, [Zn3FPFMul01], 4, [2], 1>;  // Fused Multiply Add.
-defm : Zn3WriteResXMMPair<WriteFMAX, [Zn3FPFMul01], 4, [2], 1>; // Fused Multiply Add (XMM).
-defm : Zn3WriteResYMMPair<WriteFMAY, [Zn3FPFMul01], 4, [2], 1>; // Fused Multiply Add (YMM).
+defm : Zn3WriteResXMMPair<WriteFMA, [Zn3FPFMul01], 4, [1], 1>;  // Fused Multiply Add.
+defm : Zn3WriteResXMMPair<WriteFMAX, [Zn3FPFMul01], 4, [1], 1>; // Fused Multiply Add (XMM).
+defm : Zn3WriteResYMMPair<WriteFMAY, [Zn3FPFMul01], 4, [1], 1>; // Fused Multiply Add (YMM).
 defm : X86WriteResPairUnsupported<WriteFMAZ>; // Fused Multiply Add (ZMM).
 defm : Zn3WriteResXMMPair<WriteDPPD, [Zn3FPFMul01], 9, [6], 3, /*LoadUOps=*/2>; // Floating point double dot product.
 defm : Zn3WriteResXMMPair<WriteDPPS, [Zn3FPFMul01], 15, [8], 8, /*LoadUOps=*/2>; // Floating point single dot product.