[llvm] [AArch64] SME definitions for C1-Nano scheduling model (PR #207174)

[Asher Dobrescu via llvm-commits]

================
@@ -105,12 +119,26 @@
 # STREAMING-NEXT:     1      4     1.00                        index	z0.s, w0, #1
 # STREAMING-NEXT:     1      4     1.00                        index	z0.s, #0, w0
 # STREAMING-NEXT:     1      4     1.00                        index	z0.s, w0, w1
-# STREAMING-NEXT:     1      8     1.00                        insr	z0.s, w0
+# STREAMING-NEXT:     1      4     0.20                        insr	z0.s, w0
 # STREAMING-NEXT:     1      4     0.33                        lasta	d0, p7, z31.d
 # STREAMING-NEXT:     1      16    1.50                        faddv	d0, p0, z1.d
-# STREAMING-NEXT:     1      4     0.50                        fmaxnmv	d0, p0, z1.d
+# STREAMING-NEXT:     1      16    1.50                        fmaxnmv	d0, p0, z1.d
 # STREAMING-NEXT:     1      8     0.50                        sqcvtn	z0.h, { z0.s, z1.s }
 # STREAMING-NEXT:     1      12    1.50                        uaddv	d0, p7, z31.b
+# STREAMING-NEXT:     1      4     0.33                        clasta	z0.b, p7, z0.b, z31.b
+# STREAMING-NEXT:     1      4     0.33                        clastb	z0.b, p7, z0.b, z31.b
+# STREAMING-NEXT:     1      4     0.33                        compact	z31.s, p7, z31.s
+# STREAMING-NEXT:     1      4     0.33                        splice	z29.b, p7, { z30.b, z31.b }
+# STREAMING-NEXT:     1      4     1.50                        cmpeq	p0.b, p0/z, z0.b, #-16
+# STREAMING-NEXT:     1      4     1.50                        cmpge	p0.b, p0/z, z0.b, #-16
+# STREAMING-NEXT:     1      4     1.50                        cmpgt	p0.b, p0/z, z0.b, #-16
+# STREAMING-NEXT:     1      4     1.50                        cmphi	p0.d, p0/z, z0.d, #127
+# STREAMING-NEXT:     1      4     1.50                        cmphs	p0.b, p0/z, z0.b, #0
+# STREAMING-NEXT:     1      4     1.50                        cmple	p0.b, p0/z, z0.b, #-16
+# STREAMING-NEXT:     1      4     1.50                        cmplo	p0.b, p0/z, z0.b, #0
+# STREAMING-NEXT:     1      4     1.50                        cmpls	p0.b, p0/z, z0.b, #0
+# STREAMING-NEXT:     1      4     1.50                        cmplt	p0.b, p0/z, z0.b, #-16
+# STREAMING-NEXT:     1      4     1.50                        cmpne	p0.b, p0/z, z0.b, #-16
----------------
Asher8118 wrote:

nit: I think it's fine to choose one or two instructions from a group as a test case. What's important is that the instruction group itself gets tested. They will all use the same definition after all.

https://github.com/llvm/llvm-project/pull/207174