[PATCH] D154756: [AArch64] Add scheduling model for Neoverse V1

[Evandro Menezes via Phabricator via llvm-commits]

evandro marked an inline comment as done.
evandro added inline comments.


================
Comment at: llvm/lib/Target/AArch64/AArch64SchedNeoverseV1.td:29
+
+  list<Predicate> UnsupportedFeatures = !listconcat(SMEUnsupported.F, [HasMTE]);
+}
----------------
dmgreen wrote:
> Does it help to add SVE2 too?
The Neoverse V1 does not support Armv9.


================
Comment at: llvm/lib/Target/AArch64/AArch64SchedNeoverseV1.td:58
+                              V1UnitM0, V1UnitM1]>;   // Integer units
+def V1UnitJ   : ProcResGroup<[V1UnitS, V1UnitM0]>;    // Integer 0-2 units
+def V1UnitM   : ProcResGroup<[V1UnitM0, V1UnitM1]>;   // Integer multicycle units
----------------
dmgreen wrote:
> J is invented to model the instructions that set flags, that have a throughput of 3?
Correct, gleaned from similar AArch32 instructions.


================
Comment at: llvm/test/tools/llvm-mca/AArch64/Neoverse/512tvb-sve-instructions.s:4
 
 # Check the Neoverse N2 model is used.
 
----------------
tschuett wrote:
> Nit: V1 model?
Perhaps it deserves a more comprehensive test case than just a single instruction.


================
Comment at: llvm/test/tools/llvm-mca/AArch64/Neoverse/V1-neon-instructions.s:22
+addp v0.8b, v0.8b, v0.8b
+and v0.8b, v0.8b, v0.8b
+bic v0.4h, #15, lsl #8
----------------
dmgreen wrote:
> The Neoverse-V2 version of this file has some extra instructions like addv's, aes and some dup instructions, among others.
It has many more instructions, but SVE2 ones.  I'll add more coverage here.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D154756/new/

https://reviews.llvm.org/D154756