all_true requires the input to be of the respective type, but the
final result is always a Int32x4 with a single node. So keep
the replacement type of all_true (and any_true) nodes to be Int32x4,
and use a helper method to decide what SimdType the input should
be replaced with.
Also split up any_true and all_true for readability.
Bug: v8:10507
Change-Id: I58ca50ffffcbca3ec77bbae1371ddd179925fc96
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2405803
Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69890}
We need to construct Float32 nodes for f32x4, using Word32 operators
will cause the wrong register to be allocated, triggering a CHECK
failure.
Bug: v8:10507
Change-Id: I70842f1d61b90fed2407ee52af4bc5a6b1b82ba6
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2399050
Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69856}
Fix lowering of functions that returns the result of an
i8x16 or i16x8 operation.
Bug: v8:10507
Change-Id: Ia3b29e69cff7771f85dc5160937cbaf2bbc12b55
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2399049
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69832}
v128.const was incorrectly always lowered to 4 word32 nodes, regardless
of what the lowered type was set to be.
In the test case, v128.const was consumed by i8x16.eq, so the lowered
typed of v128.const node was set to SimdType::kInt8x16, but it was still
lowered as a SimdType::kInt32x4, and then later crashes when lowering
the comparisons.
Bug: v8:10507
Change-Id: I24f16c94968cd8b6c7cd5d400d1a0046da3d47da
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2391919
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69755}
Functions with v128 in their signatures are always lowered to 4 word32.
So if a return happens to be have an input that is a f32x4 operation, we
get a register allocator error because it tries to fit a float into a
general register. To fix that we need to do some checks when lowering
kReturn, and for each input node, if we are returning a v128, and it is
to be lowered into 4 f32 nodes, we bitcast the floats to ints.
Bug: v8:10507
Change-Id: Iea2fdfc4057304ebf0898e6f7091124629c589f0
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2391331
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69705}
The interpreter is not an execution tier in production any more. It's
only used in tests.
Thus, remove {ExecutionTier::kInterpreter} and instead add a
{TestExecutionTier} that still has {kInterpreter}.
If needed (in {TestingModuleBuilder::execution_tier()}), we translate
back from {TestExecutionTier} to {ExecutionTier} (for {kLiftoff} and
{kTurboFan} only).
The {TraceMemoryOperation} method, which is shared between interpreter
and production code, now receives a {base::Optional<ExecutionTier>}, and
we will just pass en empty optional if called from the interpreter.
R=thibaudm@chromium.org
Bug: v8:10389
Change-Id: Ibe133b91e8dca6d6edbfaee5ffa0d7fe72ed6d64
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2335186
Reviewed-by: Thibaud Michaud <thibaudm@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69260}
Implement conversion of an i8x16 node to a f32x4 node.
Bug: v8:10507
Change-Id: Ifefffb779dbf25b57eae278afe41c11b41c949ac
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2185472
Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#67659}
