VFPU: Fix vrot with the angle in a destination lane

The cosine is then taken of what vrot wrote to that lane: the sine, or zero.
The IR looked at the sine lane instead of the lane holding the angle, and the
legacy JITs ignored the overlap. The assembler refuses such a vrot, so those
now leave it to the interpreter, and don't pair one with the vrot before it.

Covered by the new cpu/vfpu/vrot test.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Core/MIPS/ARM/ArmCompVFPU.cpp

@@ -2250,6 +2250,16 @@ namespace MIPSComp
 		if (vd2 >= 0)
 			GetVectorRegs(dregs2, sz, vd2);
 		GetVectorRegs(&sreg, V_Single, vs);
+		// With the angle in a destination lane, the cosine is taken of what was written there.
+		// The assembler refuses that, so leave it to the interpreter, and don't pair such a vrot.
+		for (int i = 0; i < n; i++) {
+			if (dregs[i] == sreg) {
+				DISABLE;
+			}
+			if (vd2 >= 0 && dregs2[i] == sreg) {
+				vd2 = -1;
+			}
+		}
 
 		int imm = (op >> 16) & 0x1f;
 

Core/MIPS/ARM64/Arm64CompVFPU.cpp

@@ -2231,6 +2231,16 @@ namespace MIPSComp {
 		if (vd2 >= 0)
 			GetVectorRegs(dregs2, sz, vd2);
 		GetVectorRegs(&sreg, V_Single, vs);
+		// With the angle in a destination lane, the cosine is taken of what was written there.
+		// The assembler refuses that, so leave it to the interpreter, and don't pair such a vrot.
+		for (int i = 0; i < n; i++) {
+			if (dregs[i] == sreg) {
+				DISABLE;
+			}
+			if (vd2 >= 0 && dregs2[i] == sreg) {
+				vd2 = -1;
+			}
+		}
 
 		int imm = (op >> 16) & 0x1f;
 

Core/MIPS/IR/IRCompVFPU.cpp

@@ -2335,12 +2335,20 @@ namespace MIPSComp {
 				}
 				break;
 			case 'c':
-				if (IsOverlapSafe(n, dregs, 1, sreg))
+				if (IsOverlapSafe(n, dregs, 1, sreg)) {
 					ir.Write(IROp::FCos, dregs[i], sreg[0]);
-				else if (dregs[sineLane] == sreg[0])
-					ir.Write(IROp::FCos, dregs[i], IRVTEMP_0);
-				else
-					ir.WriteFC(IROp::SetConstF, dregs[i], 0, 0, 1.0f);
+				} else {
+					// The cosine is taken of what the source lane got: the sine, or zero.
+					int srcLane = 0;
+					while (dregs[srcLane] != sreg[0]) {
+						srcLane++;
+					}
+					if (broadcastSine || srcLane == sineLane) {
+						ir.Write(IROp::FCos, dregs[i], IRVTEMP_0);
+					} else {
+						ir.WriteFC(IROp::SetConstF, dregs[i], 0, 0, 1.0f);
+					}
+				}
 				break;
 			}
 		}

Core/MIPS/x86/CompVFPU.cpp

@@ -3759,6 +3759,16 @@ void Jit::Comp_VRot(MIPSOpcode op) {
 	if (vd2 >= 0)
 		GetVectorRegs(dregs2, sz, vd2);
 	GetVectorRegs(&sreg, V_Single, vs);
+	// With the angle in a destination lane, the cosine is taken of what was written there.
+	// The assembler refuses that, so leave it to the interpreter, and don't pair such a vrot.
+	for (int i = 0; i < n; i++) {
+		if (dregs[i] == sreg) {
+			DISABLE;
+		}
+		if (vd2 >= 0 && dregs2[i] == sreg) {
+			vd2 = -1;
+		}
+	}
 
 	// Flush SIMD.
 	fpr.SimpleRegsV(&sreg, V_Single, 0);

test.py

@@ -102,6 +102,7 @@ tests_good = [
   "cpu/vfpu/matrix",
   "cpu/vfpu/vavg",
   "cpu/vfpu/exact",
+  "cpu/vfpu/vrot",
   "cpu/icache/icache",
   "cpu/lsu/lsu",
   "cpu/lsu/llsc",