[llvm] [AArch64] Model late forwarding in Neoverse N1 (PR #177590)

[Asher Dobrescu via llvm-commits]

================
@@ -448,30 +448,30 @@
 # CHECK-NEXT:  1      2     1.00                        crc32ch	w25, w26, w16
 # CHECK-NEXT:  1      2     1.00                        crc32cw	w27, w12, w23
 # CHECK-NEXT:  1      2     1.00                        crc32cx	w21, w28, x5
-# CHECK-NEXT:  1      5     3.00                        smulh	x30, x29, x28
-# CHECK-NEXT:  1      5     3.00                        smulh	xzr, x27, x26
-# CHECK-NEXT:  1      5     3.00                        umulh	x30, x29, x28
-# CHECK-NEXT:  1      5     3.00                        umulh	x23, x30, xzr
+# CHECK-NEXT:  1      5     1.00                        smulh	x30, x29, x28
+# CHECK-NEXT:  1      5     1.00                        smulh	xzr, x27, x26
+# CHECK-NEXT:  1      5     1.00                        umulh	x30, x29, x28
+# CHECK-NEXT:  1      5     1.00                        umulh	x23, x30, xzr
 # CHECK-NEXT:  1      2     1.00                        madd	w1, w3, w7, w4
 # CHECK-NEXT:  1      2     1.00                        madd	wzr, w0, w9, w11
 # CHECK-NEXT:  1      2     1.00                        madd	w13, wzr, w4, w4
 # CHECK-NEXT:  1      2     1.00                        madd	w19, w30, wzr, w29
 # CHECK-NEXT:  1      2     1.00                        mul	w4, w5, w6
-# CHECK-NEXT:  1      4     3.00                        madd	x1, x3, x7, x4
-# CHECK-NEXT:  1      4     3.00                        madd	xzr, x0, x9, x11
-# CHECK-NEXT:  1      4     3.00                        madd	x13, xzr, x4, x4
-# CHECK-NEXT:  1      4     3.00                        madd	x19, x30, xzr, x29
-# CHECK-NEXT:  1      4     3.00                        mul	x4, x5, x6
+# CHECK-NEXT:  1      4     1.00                        madd	x1, x3, x7, x4
----------------
Asher8118 wrote:

The throughput here should stay the same as before, Multiply accumulate X-form has throughput of 1/3.

https://github.com/llvm/llvm-project/pull/177590