[mono] Add basic Vector128 support for arm64. #47028
Merged
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
ghost
added
the
imhameed
added
area-Codegen-LLVM-mono
runtime-mono
imhameed
changed the title
imhameed
force-pushed
the
llvm-arm64-vector_t
branch
3 times, most recently
from
73d4d40 to
56d198c
Compare
vargaz
approved these changes
Jan 18, 2021
SIMD value types are preemptively loaded from LLVM IR memory into an LLVM IR value in the initial basic block, but OP_LDADDR wasn't aware of this, and assumed that the LLVM value associated with the argument's vreg was the argument's address.
In the ABI we use on arm and arm64, value types can be passed in and out of function calls in multiple GPRs. For function parameters on both arm and arm64, this is done by encoding the parameter as an LLVM array of machine-register-sized elements that has a byte length at least as large as the value type being passed. Converting this array to a SIMD vector in the initial BB ensures that `addresses [vreg]` and `values [vreg]` have consistent types.
imhameed
force-pushed
the
llvm-arm64-vector_t
branch
from
e3da617 to
20b3523
Compare
imhameed
requested review from
CoffeeFlux,
EgorBo,
SamMonoRT and
lambdageek
as code owners
EgorBo
approved these changes
Jan 20, 2021
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
3 participants
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Outdated suggestions cannot be applied.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.
This change:
Unpacks SIMD parameters passed via multiple GPRs (
LLVMArgVtypeAsIArgs) intolocal LLVM SSA vector-typed values on function entry. The calling convention we
use on arm and arm64 has "large" value types passed in multiple GPRs when
possible, and the way this is currently encoded in the LLVM IR we emit is by
using parameters with the LLVM type
[Size x {i32 or i64}], where theresulting size of this type, in bytes, is at least as large as the size of the
value type encoded within.
Packs SIMD return values into multiple GPRs (
LLVMArgVtypeAsScalar); the waythis is currently encoded in the LLVM IR we emit for arm64 (but not arm) is by
using a "wide" LLVM integer type (e.g.
i128).Makes
OP_LDADDRaware of our special-case handling of SIMD value types:SIMD value types passed via an indirection (
LLVMArgVtypeAddrorLLVMArgVtypeByRef) are loaded from memory to an LLVM SSA value in the initialbasic block, but
OP_LDADDRhad previously assumed that any value type passedvia an indirection had an associated LLVM SSA value containing an indirection
rather than the actual value.
Adds a debugging utility function (
mono_llvm_dump_type).Gives greppable names to some allocas.
Makes some formerly amd64-specific SIMD-related opcodes also work on arm64.
This code is all a bit messy (especially the parts surrounding parameter
passing); this change does not clean any of this up.