[PATCH] D93791: [AArch64] Add Fujitsu A64FX scheduling model
KAWASHIMA Takahiro via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Jan 4 19:57:58 PST 2021
kawashima-fj marked an inline comment as done.
kawashima-fj added inline comments.
================
Comment at: llvm/lib/Target/AArch64/AArch64SchedA64FX.td:102
+
+def A64FXAny : ProcResGroup<[A64FXIPFLA, A64FXIPPR, A64FXIPEXA, A64FXIPFLB,
+ A64FXIPEXB, A64FXIPEAGA, A64FXIPEAGB, A64FXIPBR]> {
----------------
dmgreen wrote:
> Is this used anywhere? If so, for what?
`A64FXAny` is not used elsewhere explicitly but removing it causes the following TableGen error. `CodeGenSchedModels::verifyProcResourceGroups` function in `llvm/utils/TableGen/CodeGenSchedule.cpp` emits this error.
```
$ $build/NATIVE/bin/llvm-tblgen -gen-subtarget
-I $src/llvm/lib/Target/AArch64 -I$build/include -I$src/llvm/include
-I $src/llvm/lib/Target $src/llvm/lib/Target/AArch64/AArch64.td
--write-if-changed -o lib/Target/AArch64/AArch64GenSubtargetInfo.inc
-d lib/Target/AArch64/AArch64GenSubtargetInfo.inc.d
Included from $src/llvm/lib/Target/AArch64/AArch64.td:556:
$src/llvm/lib/Target/AArch64/AArch64SchedA64FX.td:80:1:
error: proc resource group overlaps with A64FXGI02 but no supergroup contains both.
def A64FXGI01 : ProcResGroup<[A64FXIPFLA, A64FXIPPR]>;
^
```
Does anyone know why this overlaping check exists?
`llvm/lib/Target/AArch64/AArch64SchedThunderX2T99.td` also has `THX2T99Any` and `llvm/lib/Target/X86/X86Sched*.td` also have `*PortAny`. I copied it.
================
Comment at: llvm/lib/Target/AArch64/AArch64Schedule.td:35
def WriteID64 : SchedWrite; // 64-bit Divide
+def WriteID512 : SchedWrite; // 512-bit Divide
def ReadID : SchedRead; // 32/64-bit Divide
----------------
dmgreen wrote:
> What is this for? It doesn't appear to be used in any instructions.
It was my mistake. `WriteID512` was not needed. I removed it in the latest patch. Thanks.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D93791/new/
https://reviews.llvm.org/D93791
More information about the llvm-commits
mailing list