Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
llvm/include/llvm/Support/X86TargetParser.def | ||
---|---|---|
96 | FYI: gcc is going to land the sierraforest patch. They are going to add such E-core based CPUs to CPU_TYPE list but not this one, like the previous atom series e.g. tremont, silvermont and so on. It's functionally correct, so I'm ok to this change. WDYT |
llvm/include/llvm/Support/X86TargetParser.def | ||
---|---|---|
96 |
ping. |
llvm/lib/Target/X86/X86.td | ||
---|---|---|
1532 | SLMModel explicitly doesn't have support for AVX/AVX2 - I think you should investigate forking SLMModel and creating a AlderlakeEModel for all of these Also, should these be here or moved up below tremont and the other atom cores? |
llvm/lib/Target/X86/X86.td | ||
---|---|---|
1532 | Yeah, SLMModel's coverage is not right here. From view of coverage, AlderlakePModel is better here. Maybe we could use AlderlakePModel first? When sierraforest and grandridge's instruction perf info is more detailed or we can test from a real machine, we can then add a new Model and replaced here. Same for SRFTuning and GRRTuning. |
gcc has landed related patch: Sierraforest, Graniterapids and Grandrdige. We can follow their CPU list order now.
llvm/lib/Target/X86/X86.td | ||
---|---|---|
949 | Why is this differnt from FeaturesGraniteRapids in X86TargetParser.cpp? |
Drop trailing period from subject.
clang/test/Preprocessor/predefined-arch-macros.c | ||
---|---|---|
1792 | Prefer --target= to legacy -target for new tests. Consider refactoring the long CHECK lines to use delta encoding: RUN: ... | FileCheck %s --check-prefixes=A RUN: ... | FileCheck %s --check-prefixes=A,B RUN: ... | FileCheck %s --check-prefixes=A,B,C for better maintenance. |
Prefer --target= to legacy -target for new tests.
Consider refactoring the long CHECK lines to use delta encoding:
for better maintenance.