Avx512 upper regs2 #4
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This allows for more registers to be encoded in the register allocator.
Does this happen even if |
anthonycanino
pushed a commit
that referenced
this pull request
Jul 6, 2023
…tnet#87189) This fixes a startup crash on Big Sur: > error: * Assertion at /Users/runner/work/1/s/src/mono/mono/utils/mono-hwcap-arm64.c:35, condition `res == 0' not met Because sysctl can't find some of these options: $ sysctl hw.optional.armv8_crc32 hw.optional.armv8_crc32: 1 $ sysctl hw.optional.arm.FEAT_RDM sysctl: unknown oid 'hw.optional.arm.FEAT_RDM' $ sysctl hw.optional.arm.FEAT_DotProd sysctl: unknown oid 'hw.optional.arm.FEAT_DotProd' $ sysctl hw.optional.arm.FEAT_SHA1 sysctl: unknown oid 'hw.optional.arm.FEAT_SHA1' $ sysctl hw.optional.arm.FEAT_SHA256 sysctl: unknown oid 'hw.optional.arm.FEAT_SHA256' $ sysctl hw.optional.arm.FEAT_AES sysctl: unknown oid 'hw.optional.arm.FEAT_AES' Full stack trace: * thread #1, queue = 'com.apple.main-thread', stop reason = breakpoint 2.1 * frame #0: 0x0000010ef37560 libmonosgen-2.0.dylib`monoeg_assertion_message frame #1: 0x0000010ef375cc libmonosgen-2.0.dylib`mono_assertion_message + 32 frame #2: 0x0000010ef40d6c libmonosgen-2.0.dylib`mono_hwcap_arch_init + 544 frame #3: 0x0000010ef54bd8 libmonosgen-2.0.dylib`mono_hwcap_init + 72 frame #4: 0x0000010ee14dc0 libmonosgen-2.0.dylib`parse_optimizations + 52 frame #5: 0x0000010edbed48 libmonosgen-2.0.dylib`mono_init frame #6: 0x0000010ee18968 libmonosgen-2.0.dylib`mono_jit_init_version frame #7: 0x0000010f48a300 libxamarin-dotnet-debug.dylib`xamarin_bridge_initialize + 216 frame #8: 0x0000010f4900a4 libxamarin-dotnet-debug.dylib`xamarin_main + 376
anthonycanino
pushed a commit
that referenced
this pull request
Dec 18, 2024
* bug #1: don't allow for values out of the SerializationRecordType enum range * bug #2: throw SerializationException rather than KeyNotFoundException when the referenced record is missing or it points to a record of different type * bug #3: throw SerializationException rather than FormatException when it's being thrown by BinaryReader (or sth else that we use) * bug #4: document the fact that IOException can be thrown * bug #5: throw SerializationException rather than OverflowException when parsing the decimal fails * bug #6: 0 and 17 are illegal values for PrimitiveType enum * bug #7: throw SerializationException when a surrogate character is read (so far an ArgumentException was thrown)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I'm creating this PR so I have a place to document a few of the changes I've made and some of the existing issues.
Changes:
emitxarch.cppetc.xmm16 - xmm31to associated register definition files (register.handtargetamd64.h). In addition, add a stress modeJitStressRegs=4that will force the upper simd registers to be allocated.JitStressEVEXEncoding=1must be enabled for the upper regs to be used.RBM_NONEis used as themaskduring much of the lsra build phase, at which point a use will select the set of all registers available for a register type ifRBM_NONEis specified (dotnet@f8a04a3#diff-ad66a6bcf1fd550d5ad10d995c03218afbbc39463d36e1f2a224f9ca070a2f99R560-R574).Because of node containment, it wasn't so easy to specific the set of low simd registers during the lsra build code because a tbe build up of a contained node didn't necessarily use the same register set as its containing node (a limitation which is noted in the comment). I get around this with the
RBM_LOWSIMDmarker.I've added
zmm16:zmm31to the flt callee trash register set. Before I made this change, I was getting some issues with some functions using and not preserving the registers correctly.Because the set of available registers is now dependant upon available ISA and not arch, there is some code to adjust the available float registers based on the available AVX512 ISA. Likewise, I think a lot of the trash set macros needs to be adjusted as well (
RBM_STOP_FOR_GC_TRASHas an example).Currently, most of the JIT tests pass and no asserts on superpmi but quite a lot of diffs in the prolog/epilog.