Some image ops return three or five dwords. Previously, we modeled that
with a 4 or 8 dword register class. The register allocator could
cleverly spot that some subregs were dead and allocate something else
there, but that caused the de-optimization that waitcnt insertion would
think that the result was used immediately.
This commit allows such an image op to have a result with a three or
five dword result, avoiding the above de-optimization.
Differential Revision: https://reviews.llvm.org/D58905
Change-Id: I3651211bbd7ed22721ee7b9fefd7bcc60a809d8b
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356757 91177308-0d34-0410-b5e6-96231b3b80d8
Now we have vec3 MVTs, this commit implements dwordx3 variants of the
buffer intrinsics.
On gfx6, a dwordx3 buffer load intrinsic is implemented as a dwordx4
instruction, and a dwordx3 buffer store intrinsic is not supported.
We need to support the dwordx3 load intrinsic because it is generated by
subtarget-unaware code in InstCombine.
Differential Revision: https://reviews.llvm.org/D58904
Change-Id: I016729d8557b98a52f529638ae97c340a5922a4e
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356755 91177308-0d34-0410-b5e6-96231b3b80d8
The AArch64 test was broken since the result register already had a
set register class, so this test was a no-op. The mapping verify call
would fail because the result size is not the same as the inputs like
in a copy or phi.
The AMDGPU testcases are half broken and introduce illegal VGPR->SGPR
copies which need much more work to handle correctly (same for phis),
but add them as a baseline.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356713 91177308-0d34-0410-b5e6-96231b3b80d8
Added support for dwordx3 for most load/store types, but not DS, and not
intrinsics yet.
SI (gfx6) does not have dwordx3 instructions, so they are not enabled
there.
Some of this patch is from Matt Arsenault, also of AMD.
Differential Revision: https://reviews.llvm.org/D58902
Change-Id: I913ef54f1433a7149da8d72f4af54dbb13436bd9
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356659 91177308-0d34-0410-b5e6-96231b3b80d8
Machine DCE cannot remove a dead definition if there are non-dbg uses.
A use however can be in the same instruction:
dead %0 = INST %0
Such instructions sometimes created by Detect dead lanes pass.
Allow this instruction to be deleted despite the use if the only use
belongs to the same instruction.
Differential Revision: https://reviews.llvm.org/D59565
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356619 91177308-0d34-0410-b5e6-96231b3b80d8
The constantness shouldn't change the register bank choice. We also
don't need to restrict this to only indexing VGPRs, since it's
possible to index SGPRs (but SelectionDAG made using this
difficult). Allow directly indexing SGPRs when appropriate.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356611 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
- Should use `targetconstant` instead of `constant` operand for clamp
bit, which is expected as an immediate operand. Under certain
conditions, such as a common `i1 false` constant is used in other
place and selected before the instruction with clamp bit, register
operand may be added instead of immediate one. Use `targetcosntant` to
enforce that.
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59608
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356608 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
PAL metadata now supports both the old linear reg=val pairs format and
the new MsgPack format.
The MsgPack format uses YAML as its textual representation. On output to
YAML, a mnemonic name is provided for some hardware registers.
Differential Revision: https://reviews.llvm.org/D57028
Change-Id: I2bbaabaaca4b3574f7e03b80fbef7c7a69d06a94
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356591 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This commit introduces a new AMDGPUPALMetadata class that:
* is inside the AMDGPU target;
* keeps an in-memory representation of PAL metadata;
* provides a method to read the frontend-supplied metadata from LLVM IR;
* provides methods for the asm printer to set metadata items;
* provides methods to write the metadata as a binary blob to put in a
.note record or as an asm directive;
* provides a method to read the metadata as a binary blob from a .note
record.
Because llvm-readobj cannot call directly into a target, I had to remove
llvm-readobj's ability to dump PAL metadata, pending a resolution to
https://reviews.llvm.org/D52821
Differential Revision: https://reviews.llvm.org/D57027
Change-Id: I756dc830894fcb6850324cdcfa87c0120eb2cf64
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356582 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
If an MIMG instruction has managed to get through to adjustWritemask in isel but
has no uses (and doesn't enable TFC) then prevent an assertion by not attempting
to adjust the writemask.
The instruction will be removed anyway.
Change-Id: I9a5dba6bafe1f35ac99c1b73df390936e2ac27a7
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58964
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356540 91177308-0d34-0410-b5e6-96231b3b80d8
I found this really weird WWM-related case whereby through the WWM
transformations our isel lowering was trying to promote 2 min's into a
min3 for the i8 type, which our hardware doesn't support.
The new min3_i8.ll test case would previously spew the error:
PromoteIntegerResult #0: t69: i8 = SMIN3 t70, Constant:i8<0>, t68
Before the simple fix to our isel lowering to not do it for i8 MVT's.
Differential Revision: https://reviews.llvm.org/D59543
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356464 91177308-0d34-0410-b5e6-96231b3b80d8
Allow the clamp modifier on vop3 int arithmetic instructions in assembly
and disassembly.
This involved adding a clamp operand to the affected instructions in MIR
and MC, and thus having to fix up several places in codegen and MIR
tests.
Differential Revision: https://reviews.llvm.org/D59267
Change-Id: Ic7775105f02a985b668fa658a0cd7837846a534e
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356399 91177308-0d34-0410-b5e6-96231b3b80d8
This commit allows v_cndmask_b32_e64 with abs, neg source
modifiers on src0, src1 to be assembled and disassembled.
This does appear to be allowed, even though they are floating point
modifiers and the operand type is b32.
To do this, I added src0_modifiers and src1_modifiers to the
MachineInstr, which involved fixing up several places in codegen and mir
tests.
Differential Revision: https://reviews.llvm.org/D59191
Change-Id: I69bf4a8c73ebc65744f6110bb8fc4e937d79fbea
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356398 91177308-0d34-0410-b5e6-96231b3b80d8
Add an experimental buffer fat pointer address space that is currently
unhandled in the backend. This commit reserves address space 7 as a
non-integral pointer repsenting the 160-bit fat pointer (128-bit buffer
descriptor + 32-bit offset) that is heavily used in graphics workloads
using the AMDGPU backend.
Differential Revision: https://reviews.llvm.org/D58957
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356373 91177308-0d34-0410-b5e6-96231b3b80d8
There are a few different issues, mostly stemming from using
generation based checks for anything instead of subtarget
features. Stop adding flat-address-space as a feature for HSA, as it
should only be a device property. This was incorrectly allowing flat
instructions to select for SI.
Increase the default generation for HSA to avoid the encoding error
when emitting objects. This has some other side effects from various
checks which probably should be separate subtarget features (in the
cost model and for dealing with the DS offset folding issue).
Partial fix for bug 41070. It should probably be an error to try using
amdhsa without flat support.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356347 91177308-0d34-0410-b5e6-96231b3b80d8
AMDGPU would like to have MVTs for v3i32, v3f32, v5i32, v5f32. This
commit does not add them, but makes preparatory changes:
* Fixed assumptions of power-of-2 vector type in kernel arg handling,
and added v5 kernel arg tests and v3/v5 shader arg tests.
* Added v5 tests for cost analysis.
* Added vec3/vec5 arg test cases.
Some of this patch is from Matt Arsenault, also of AMD.
Differential Revision: https://reviews.llvm.org/D58928
Change-Id: I7279d6b4841464d2080eb255ef3c589e268eabcd
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356342 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
At the exit of the loop, the compiler uses a register to remember and accumulate
the number of threads that have already exited. When all active threads exit the
loop, this register is used to restore the exec mask, and the execution continues
for the post loop code.
When there is a "continue" in the loop, the compiler made a mistake to reset the
register to 0 when the "continue" backedge is taken. This will result in some
threads not executing the post loop code as they are supposed to.
This patch fixed the issue.
Reviewers:
nhaehnle, arsenm
Differential Revision:
https://reviews.llvm.org/D59312
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356298 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
- During the fixing of SGPR copying from VGPR, ensure users of SCC is
properly propagated, i.e.
* only propagate through live def of SCC,
* skip the SCC-def inst itself, and
* stop the propagation on the other SCC-def inst after checking its
SCC-use first.
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59362
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356258 91177308-0d34-0410-b5e6-96231b3b80d8
This has been a very painful missing feature that has made producing
reduced testcases difficult. In particular the various registers
determined for stack access during function lowering were necessary to
avoid undefined register errors in a large percentage of
cases. Implement a subset of the important fields that need to be
preserved for AMDGPU.
Most of the changes are to support targets parsing register fields and
properly reporting errors. The biggest sort-of bug remaining is for
fields that can be initialized from the IR section will be overwritten
by a default initialized machineFunctionInfo section. Another
remaining bug is the machineFunctionInfo section is still printed even
if empty.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356215 91177308-0d34-0410-b5e6-96231b3b80d8
This is consistent with what SelectionDAG does and is much easier to
work with than the extract sequence with an artificial wide register.
For the AMDGPU control flow intrinsics, this was producing an s128 for
the i64, i1 tuple return. Any legalization that should apply to a real
s128 value would badly obscure the direct values that need to be seen.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356147 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
MsgPackDocument is the lighter-weight replacement for MsgPackTypes. This
commit switches AMDGPU HSA metadata processing to use MsgPackDocument
instead of MsgPackTypes.
Differential Revision: https://reviews.llvm.org/D57024
Change-Id: I0751668013abe8c87db01db1170831a76079b3a6
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356081 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
A number of optimizations are inhibited by single-use TokenFactors not
being merged into the TokenFactor using it. This makes we consider if
we can do the merge immediately.
Most tests changes here are due to the change in visitation causing
minor reorderings and associated reassociation of paired memory
operations.
CodeGen tests with non-reordering changes:
X86/aligned-variadic.ll -- memory-based add folded into stored leaq
value.
X86/constant-combiners.ll -- Optimizes out overlap between stores.
X86/pr40631_deadstore_elision -- folds constant byte store into
preceding quad word constant store.
Reviewers: RKSimon, craig.topper, spatel, efriedma, courbet
Reviewed By: courbet
Subscribers: dylanmckay, sdardis, nemanjai, jvesely, nhaehnle, javed.absar, eraman, hiraditya, kbarton, jrtc27, atanasyan, jsji, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59260
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@356068 91177308-0d34-0410-b5e6-96231b3b80d8
This indicates an intrinsic parameter is required to be a constant,
and should not be replaced with a non-constant value.
Add the attribute to all AMDGPU and generic intrinsics that comments
indicate it should apply to. I scanned other target intrinsics, but I
don't see any obvious comments indicating which arguments are intended
to be only immediates.
This breaks one questionable testcase for the autoupgrade. I'm unclear
on whether the autoupgrade is supposed to really handle declarations
which were never valid. The verifier fails because the attributes now
refer to a parameter past the end of the argument list.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355981 91177308-0d34-0410-b5e6-96231b3b80d8
Targets can potentially emit more efficient code if they know address
computations never overflow. For example ILP32 code on AArch64 (which only has
64-bit address computation) can ignore the possibility of overflow with this
extra information.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355926 91177308-0d34-0410-b5e6-96231b3b80d8
Narrow Scalar G_MUL for MIPS32.
Revisit NarrowScalar implementation in LegalizerHelper.
Introduce new helper function multiplyRegisters.
It performs generic multiplication of values held in multiple registers.
Generated instructions use only types NarrowTy and i1.
Destination can be same or two times size of the source.
Differential Revision: https://reviews.llvm.org/D58824
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355814 91177308-0d34-0410-b5e6-96231b3b80d8
When matching half of the build_vector to a load, there could still be
a hidden dependency on the other half of the build_vector the pattern
wouldn't detect. If there was an additional chain dependency on the
other value, a cycle could be introduced.
I don't think a tablegen pattern is capable of matching the necessary
conditions, so move this into PreprocessISelDAG. Check isPredecessorOf
for the other value to avoid a cycle. This has a warning that it's
expensive, so this should probably be moved into an MI pass eventually
that will have more freedom to reorder instructions to help match
this. That is currently complicated by the lack of a computeKnownBits
type mechanism for the selected function.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355731 91177308-0d34-0410-b5e6-96231b3b80d8
This avoids breaking possible value dependencies when sorting loads by
offset.
AMDGPU has some load instructions that write into the high or low bits
of the destination register, and have a tied input for the other input
bits. These can easily have the same base pointer, but be a swizzle so
the high address load needs to come first. This was inserting glue
forcing the opposite ordering, producing a cycle the InstrEmitter
would assert on. It may be potentially expensive to look for the
dependency between the other loads, so just skip any where this could
happen.
Fixes bug 40936 by reverting r351379, which added a hacky attempt to
fix this by adding chains in this case, which I think was just working
around broken glue before the InstrEmitter. The core of the patch is
re-implementing the fix for that problem.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355728 91177308-0d34-0410-b5e6-96231b3b80d8
This was checking the wrong operands for the base register and the
offsets. The indexes are shifted by the number of output registers
from the machine instruction definition, and the chain is moved to the
end.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355722 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
This adds support for 64 bit buffer atomic arithmetic instructions but does not include
cmpswap as that depends on a fix to the way the register pairs are handled
Change-Id: Ib207ea65fb69487ccad5066ea647ae8ddfe2ce61
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, jfb, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58918
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355520 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
In some cases the KILL was causing a hazard to be introduced as these were
scheduled into hazard slots, but don't result in an instruction.
KILL shouldn't be considered for hazard recognition.
Change-Id: Ib6d2a2160f8c94cd0ce611ab198c7e4f46aeffcf
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58898
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355384 91177308-0d34-0410-b5e6-96231b3b80d8
These were not recognized as potential atomics by memory legalizer.
The test was working not because legalizer did a right thing, but
because it has skipped all these instructions. When I have fixed
DS desciption test started to fail because region address has
changed from 4 to 2 a while ago.
Differential Revision: https://reviews.llvm.org/D58802
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355179 91177308-0d34-0410-b5e6-96231b3b80d8
We have to at least tolerate calls to kernels, possibly with a
mismatched calling convention on the callsite.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355049 91177308-0d34-0410-b5e6-96231b3b80d8
Tim Renouf