[llvm] Change `fp128` lowering to use `f128` functions by default (PR #76558)

[Trevor Gross via llvm-commits]

================
@@ -425,14 +425,19 @@ define fp128 @qp_sincos(ptr nocapture readonly %a) nounwind {
 ; CHECK-NEXT:    mflr r0
 ; CHECK-NEXT:    stdu r1, -64(r1)
 ; CHECK-NEXT:    std r0, 80(r1)
-; CHECK-NEXT:    addi r5, r1, 48
-; CHECK-NEXT:    addi r6, r1, 32
-; CHECK-NEXT:    lxv v2, 0(r3)
-; CHECK-NEXT:    bl sincosf128
+; CHECK-NEXT:    stxv v31, 48(r1) # 16-byte Folded Spill
+; CHECK-NEXT:    stxv v30, 32(r1) # 16-byte Folded Spill
+; CHECK-NEXT:    lxv v31, 0(r3)
+; CHECK-NEXT:    vmr v2, v31
+; CHECK-NEXT:    bl cosf128
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    vmr v30, v2
+; CHECK-NEXT:    vmr v2, v31
+; CHECK-NEXT:    bl sinf128
 ; CHECK-NEXT:    nop
-; CHECK-NEXT:    lxv v2, 48(r1)
-; CHECK-NEXT:    lxv v3, 32(r1)
-; CHECK-NEXT:    xsmulqp v2, v3, v2
+; CHECK-NEXT:    xsmulqp v2, v30, v2
+; CHECK-NEXT:    lxv v31, 48(r1) # 16-byte Folded Reload
+; CHECK-NEXT:    lxv v30, 32(r1) # 16-byte Folded Reload
----------------
tgross35 wrote:

I don't follow this, why was the sin+cos->sincos optimization lost here?

https://github.com/llvm/llvm-project/pull/76558