The first trivial example I tried failed to merge due to the user scan
logic. Remove the complicated scan of users handling with distance
thresholds, with a same block restriction. The actual expansion of
sincos is basically the same size as sin or cos individually. Copy the
technique the generic optimization uses, which is to just use the
input instruction as the insert point or just insert at the start of
the entry block.
https://reviews.llvm.org/D156706
For G_ABS with type v2s16 and sgpr inputs break down into two s32 G_ABS
instructions.
Patch by: Acim Maravic
Differential Revision: https://reviews.llvm.org/D155867
There is no need to increase the size of odd sized vectors if they are
going to be scalarized by a different rule.
Patch by: Acim Maravic
Differential Revision: https://reviews.llvm.org/D155865
It's not a libcall so doesn't really belong here to begin
with. Relying on checking the target name and explicit features isn't
particularly sound either. The library doesn't use the intrinsic
anymore, so it doesn't matter anyway.
This reverts commit a496c8be6e638ae58bb45f13113dbe3a4b7b23fd.
The workaround in c26dfc81e254c78dc23579cf3d1336f77249e1f6 should work
around the underlying problem with SUBREG_TO_REG.
Introduced the convergent equivalent of the existing G_INTRINSIC opcodes:
- G_INTRINSIC_CONVERGENT
- G_INTRINSIC_CONVERGENT_W_SIDE_EFFECTS
Out of the targets that currently have some support for GlobalISel, the patch
assumes that the convergent intrinsics only relevant to SPIRV and AMDGPU.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D154766
Change the scheduler's physical register dependency tracking from
registers-and-their-aliases to regunits. This has a couple of advantages
when subregisters are used:
- The dependency tracking is more accurate and creates fewer useless
edges in the dependency graph. An AMDGPU example, edited for clarity:
SU(0): $vgpr1 = V_MOV_B32 $sgpr0
SU(1): $vgpr1 = V_ADDC_U32 0, $vgpr1
SU(2): $vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 $vgpr0_vgpr1, 0, 0
There is a data dependency on $vgpr1 from SU(0) to SU(1) and from
SU(1) to SU(2). But the old dependency tracking code also added a
useless edge from SU(0) to SU(2) because it thought that SU(0)'s def
of $vgpr1 aliased with SU(2)'s use of $vgpr0_vgpr1.
- On targets like AMDGPU that make heavy use of subregisters, each
register can have a huge number of aliases - it can be quadratic in
the size of the largest defined register tuple. There is a much lower
bound on the number of regunits per register, so iterating over
regunits is faster than iterating over aliases.
The LLVM compile-time tracker shows a tiny overall improvement of 0.03%
on X86. I expect a larger compile-time improvement on targets like
AMDGPU.
Differential Revision: https://reviews.llvm.org/D156552
In SelectionDAG InstrEmitter automatically puts dead flags on unused
physreg defs everywhere. The generated selectors should also set dead
on physreg defs that were not used in the pattern.
This is required for many trees produced in practice for i8 CodeGen.
Differential Revision: https://reviews.llvm.org/D155864
Change-Id: Iac01d183d9998b15138bdc7a5051e3bed338e7d9
The patterns were ripped out in
a4a3ac10cb1a40ccebed4e81cd7e94f1eb71602d so this always needs to be
custom lowered. I absolutely hate how difficult it is to write tests
for these, I have no doubt there are more of these hidden.
Fixes#64142
And dependent commits.
Details in D150388.
This reverts commit 825b7f0ca5f2211ec3c93139f98d1e24048c225c.
This reverts commit 7a98f084c4d121244ef7286bc6503b6a181d446e.
This reverts commit b4a62b1fa546312d882fa12dfdcd015177d66826.
This reverts commit b7836d856206ec39509d42529f958c920368166b.
No conflicts in the code, few tests had conflicts in autogenerated CHECKs:
llvm/test/CodeGen/Thumb2/mve-float32regloops.ll
llvm/test/CodeGen/AMDGPU/fix-frame-reg-in-custom-csr-spills.ll
Reviewed By: alexfh
Differential Revision: https://reviews.llvm.org/D156381
The previous condition was incorrect in some cases, like storing <2 x i32>
into a double. If IndexVal was >0, we ended up never storing anything.
Reviewed By: #amdgpu, arsenm
Differential Revision: https://reviews.llvm.org/D156308
In code object 5 (https://llvm.org/docs/AMDGPUUsage.html#code-object-v5-metadata) the AMDGPU backend added the .uses_dynamic_stack bit to the kernel meta data to identity kernels which have compile time indeterminable stack usage (indirect function calls and recursion mainly). This patch adds this information to the output of the kernel-resource-usage remarks.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D156040
Author: Corbin Robeck <corbin.robeck@amd.com>
Correct AMDGPU strictfp tests to follow the rules documented in the LangRef:
https://llvm.org/docs/LangRef.html#constrained-floating-point-intrinsics
Mostly these tests just needed the strictfp attribute on function
definitions. I've also removed the strictfp attribute from uses
of the constrained intrinsics because it comes by default since
D154991, but I only did this in tests I was changing anyway.
I also removed attributes added to declare lines of intrinsics. The
attributes of intrinsics cannot be changed in a test so I eliminated
attempts to do so.
Test changes verified with D146845.
rocm-device-libs and llpc were avoiding using f64 sqrt
intrinsics in favor of their own expansions. Port the
expansion into the backend. Both of these users should be
updated to call the intrinsic instead.
The library and llpc expansions are slightly different.
llpc uses an ldexp to do the scale; the library uses a multiply.
Use ldexp to do the scale instead of the multiply.
I believe v_ldexp_f64 and v_mul_f64 are always the same number of
cycles, but it's cheaper to materialize the 32-bit integer constant
than the 64-bit double constant.
The libraries have another fast version of sqrt which will
be handled separately.
I am tempted to do this in an IR expansion instead. In the IR
we could take advantage of computeKnownFPClass to avoid
the 0-or-inf argument check.
This allows PromoteAlloca to not be reliant on a second SROA run to remove the alloca completely. It just does the full transformation directly.
Note PromoteAlloca is still reliant on SROA running first to
canonicalize the IR. For instance, PromoteAlloca will no longer handle aggregate types because those should be simplified by SROA before reaching the pass.
Reviewed By: #amdgpu, arsenm
Differential Revision: https://reviews.llvm.org/D152706
In this example an implicit def had live-out undef subrange
defs. After coalescing with the def from a previous block, the
undef-defed lanes are no longer live out of the block in the new
interval. An empty subrange was tenatively created for these lanes,
but it must be deleted.
A live out implicit_def wasn't deleted, but the subranges weren't
correctly updated. The main range was correct but the def
corresponding to the initial main range def instruction was missing
from the lanes redefined in another block.
The written lanes are not quite the same as the valid lanes in the
case of an implicit_def.
Fixes verifier error in blender. There is an additional verifier in
some of the testcase variants where an empty subrange remains.
When input to intrinsic is uniform value, reduced value is
same as input whereas if input value is divergent we need
to iterate over all active lanes of WaveFront to perform
the reduction.
The control flow for a `loop` has been set up, which
iterates over `only` active lanes to perform reduction.
Introduced WAVE_REDUCE_UMIN_PSEUDO_U32 and
WAVE_REDUCE_UMAX_PSEUDO_U32 Pseudos which
are lowered Post-ISel (in `EmitInstrWithCustomInserter `).
Reviewed By: arsenm, #amdgpu
Differential Revision: https://reviews.llvm.org/D154858
Extends the new frexp scaled reciprocal to the general case. The
reciprocal case is just the same thing when frexp of 1 is constant
folded. Could probably clean up the code to rely on that constant
folding.
Improves results for the IEEE path for the default OpenCL division. We
used to only emit the fdiv.fast intrinsic with a 2.5 ulp accuracy
threshold with DAZ, which uses explicit range checks. This gives us a
better fast option with the default IEEE behavior.
NFC-ish. Does trigger some reordering of the fdiv scalarization. Also
skips scalarizing in more cases where nothing was going to happen. We
can still scalarize in some no-op edge cases.
https://reviews.llvm.org/D155740
The highlight change is a new denormal safe 1ulp lowering which uses
rcp after using frexp to perform input scaling. This saves 2
instructions compared to other implementations which performed an
explicit denormal range change. This improves the OpenCL default, and
requires a flag for HIP. I don't believe there's any flag wired up for
OpenMP to emit the necessary fpmath metadata.
This provides several improvements and changes that were hard to
separate without regressing one case or another. Disturbingly the
OpenCL conformance test seems to have the reciprocal test commented
out. I locally hacked it back in to test this.
Starts introducing f32 rsq intrinsics in AMDGPUCodeGenPrepare. Like
the rcp case, we could do this in codegen if !fpmath were preserved
(although we would lose some computeKnownFPClass tricks). Start
requiring contract flags to form rsq. The rsq fusion actually improves
the result from ~2ulp to ~1ulp. We have some older fusion in codegen
which only keys off unsafe math which should be refined.
Expand rsq patterns by checking for denormal inputs and pre/post
multiplying like the current library code does. We also take advantage
of computeKnownFPClass to avoid the scaling when we can statically
prove the input cannot be a denormal. We could do the same for the rcp
case, but unlike rsq a large input can underflow to denormal. We need
additional upper bound exponent checks on the input in order to do the
same for rcp.
This rsq handling also now starts handling the negated case. We
introduce rsq with an fneg. In the case the fneg doesn't fold into its
user, it's a neutral change but provides improvement if it is foldable
as a source modifier.
Also starts respecting the arcp attribute properly, and more strictly
interprets afn. We were previously interpreting afn as implying you
could do the reciprocal expansion of an fdiv. The codegen handling of
these also needs to be revisited.
This also effectively introduces the optimization
combineRepeatedFPDivisors enables, just done in the IR instead (and
only for f32).
This is almost across the board better. The one minor regression is
for gfx6/buggy frexp case where for multiple reciprocals, we could
previously reuse rematerialized constants per instance (it's neutral
for a single rcp).
The fdiv.fast and sqrt handling need to be revisited next.
https://reviews.llvm.org/D155593
We kind of have to have multiple implementations of fdiv split between
the two selectors with some pre-processing. Add yet another test to
check for consistency of interpretation of flag combinations. We have
quite a bit of test redundancy here already, but there are so many
possible interesting permutations it's unwieldy to cover every detail
in any one of them. We have a number of overlapping fdiv tests but
it's hard to follow everything going on as it is.
This isn't always folded to fneg for a freestanding fsub depending on
the denormal mode. When matching source modifiers, we're implicitly
canonicalizing the input so we can fold it here.
Doesn't bother handling the VOP3P case since it's only relevant with
DAZ, which nobody really uses with f16.
For f64, tests show an existing bug where DAGCombiner tries to respect
the denormal mode for fsub -0, x, but not after it's lowered to fadd
-0, (fneg x). Either the fold is wrong or we shouldn't restrict the
fsub case based on the denormal mode.
https://reviews.llvm.org/D155652
