Have mono handle ConvertToIntegerNative for Double and Single #112206
Conversation
93fb02f to
87d57e2
Compare
87d57e2 to
8bf0780
Compare
| case MONO_TYPE_I2: | ||
| opcode = OP_FCONV_TO_I2; | ||
| break; | ||
| #if TARGET_SIZEOF_VOID_P == 4 |
There was a problem hiding this comment.
nit: in other places we use SIZEOF_REGISTER for these kinds of checks.
There was a problem hiding this comment.
looks like it varies based on a file, e.g., in simd-intrinsics.c, the TARGET_SIZEOF_VOID_P looks to be the dominant way.
There was a problem hiding this comment.
yeah its mixed, in this file SIZEOF_REGISTER seems to dominate. An interesting observation is that we have some code like this:
#if (SIZEOF_REGISTER > TARGET_SIZEOF_VOID_P) && (G_BYTE_ORDER == G_BIG_ENDIAN)
and we have this comment in mono-machine.h
SIZEOF_REGISTER is usually the same as TARGET_SIZEOF_VOID_P, except when MONO_ARCH_ILP32 is defined
There was a problem hiding this comment.
Conversions care about the size of a pointer, not the size of the register; hence why this one was selected. It matches how other code paths, such as method-to-ir, typically deal with MONO_TYPE_I.
|
Little curious if this should still be excluded on Mono (same for double), ConvertToIntegerNative ? For direct calls we would optimize calls using the new intrinsic implemented by this PR, but I would assume that indirect calls might end up in this implementation on Mono as well. |
|
Looks like this needs to explicitly add support to Mono Interpreter as well. |
c5e3af0 to
5fba9bf
Compare
5fba9bf to
c0d450e
Compare
a90f670 to
4a263b9
Compare
|
Cc. @BrzVlad for the mono mini side |
* main: Code clean up in AP for NonNull* (dotnet#112027) JIT: Invalidate LSRA's DFS tree if we aren't running new layout phase (dotnet#112364) Update dependencies from https://github.com/dotnet/source-build-reference-packages build 20250204.2 (dotnet#112339) Add doc on OS onboarding (dotnet#112026) Add `TypeName` APIs to simplify metadata lookup. (dotnet#111598) Internal monitor impl not using coop mutex causing deadlocks on Android. (dotnet#112358) Do not run NAOT arm64 OSX testing on all PRs (dotnet#112342) Special-case empty enumerables in AsyncEnumerable (dotnet#112321) Have mono handle ConvertToIntegerNative for Double and Single (dotnet#112206) Update dependencies from https://github.com/dotnet/arcade build 20250206.4 (dotnet#112338) System.Configuration.ConfigurationManager.Tests: use Assembly.Location to determine ThisApplicationPath. (dotnet#112231) Force write of local file header when "version needed to extract" changes (dotnet#112032) JIT: Don't reorder handler blocks (dotnet#112292) [RISC-V] Synthesize some floating constants inline (dotnet#111529) Enable `SA1000`: Spacing around keywords (dotnet#112302) Fix relocs for linux-riscv64 AOT (dotnet#112331)
|
Thank you @tannergooding for the fix. The regressions got resolved |
This should resolve dotnet/perf-autofiling-issues#49586