Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
clang/test/CodeGen/target-builtin-noerror.c | ||
---|---|---|
131 | I think this list was once in alphabetical order | |
compiler-rt/lib/builtins/cpu_model.c | ||
110 ↗ | (On Diff #467672) | The raptorlake patch also puts a constant here. What is the correct order? Can you please rebase one patch on top of the other or merge them into a single patch. |
llvm/lib/Target/X86/X86.td | ||
---|---|---|
1445 | This should be with the alderlake/raptorlake defs below (and use ADL model/features/tuning) |
THX for review! Gcc is also recently reviewing related patches. So to align with the compiler-rt and libgcc, let's wait for their land first.
Please can you update the summary now that raptorlake + meteorlake are in the same patch?
clang/test/Preprocessor/predefined-arch-macros.c | ||
---|---|---|
2245 | (pedantic) Probably better to put these after the alderlake tests so its easier to find? |
clang/test/Driver/x86-march.c | ||
---|---|---|
128 | Move these after alderlake instead of the old atom cores? |
compiler-rt/lib/builtins/cpu_model.c | ||
---|---|---|
110 ↗ | (On Diff #468387) | typo ? (ZHAOXIN_FAM7H_LUJIAZUI) |
compiler-rt/lib/builtins/cpu_model.c | ||
---|---|---|
110 ↗ | (On Diff #468387) | I think not. That is to keep aligned with libgcc for compatibility by Craig's catch on raptorlake patch. See https://github.com/gcc-mirror/gcc/blob/master/gcc/common/config/i386/i386-cpuinfo.h#L94 We know gcc is also recently modify here. To wait for their landing and then aligning with them, I'll convert to draft for this patch. |
compiler-rt/lib/builtins/cpu_model.c | ||
---|---|---|
111 ↗ | (On Diff #468387) | I see. But if possible, could we split "ZHAOXIN_FAM7H_LUJIAZUI" to another patch? |
For saving capacity of ProcessorSubtypes, gcc decided to not support part of compiler features of these two cpus:
__builtin_cpu_is("meteorlake") __attribute__((target("arch=raptorlake"))) ... some others I don't know.
Gcc's related patch:
raptorlake and meteorlake
Updated to align with gcc first. Welcome opinions and review!
clang/test/Preprocessor/predefined-arch-macros.c | ||
---|---|---|
2245 | I merged them with the check-prefix. | |
compiler-rt/lib/builtins/cpu_model.c | ||
111 ↗ | (On Diff #468387) | Related change is removed. See my latest comment. |
478 ↗ | (On Diff #468387) | Good catch! While related changes are removed, see my latest comment. |
llvm/lib/Support/Host.cpp | ||
843 | Good catch! |
clang/test/Driver/x86-march.c | ||
---|---|---|
91 | You can just write --target=x86_64. Then the line is short enough that you do not need wrapping. Not necessary to follow other lines. |
Confirmed they still supported these two features. So I added X86_CPU_SUBTYPE_ALIAS to support so. Pls review.
clang/lib/CodeGen/CGBuiltin.cpp | ||
---|---|---|
12942–12943 | Mo difference with X86_CPU_SUBTYPE? |
Mo difference with X86_CPU_SUBTYPE?