Non-entry functions were unconditionally aligned to 4 bytes with no
architecture-specific preferred alignment, and setAlignment() was used
instead of ensureAlignment(), overwriting any explicit IR attributes.
Add instruction cache line size and fetch alignment data to GCNSubtarget
for each generation (GFX9: 64B/32B, GFX10: 64B/4B, GFX11+: 128B/4B). Use
this to call setPrefFunctionAlignment() in SITargetLowering, aligning
non-entry functions to the cache line size by default. Change
setAlignment to ensureAlignment in AMDGPUAsmPrinter so explicit IR align
attributes are respected.
Empirical thread trace analysis on gfx942, gfx1030, gfx1100, and gfx1200
showed that only GFX9 exhibits measurable fetch stalls when functions
cross the 32-byte fetch window boundary. GFX10+ showed no alignment
sensitivity. A hidden option -amdgpu-align-functions-for-fetch-only is
provided to use the fetch granularity instead of cache line size.
Assisted-by: Claude Opus
Returns uint64_t to simplify callers. The goal is eventually replace
getValueType with this query, which should return the known minimum
reference-able size, as provided (instead of a Type) during create.
Additionally the common isSized query would be replaced with an
isExactKnownSize query to test if that size is an exact definition.
AMDGPUMCExpr lives in the MC layer it should not depend on Function.h or
GCNSubtarget.h
Move the function that needed GCNSubtarget to the one file that called
it.
According to `llvm-project/llvm/docs/AMDGPUUsage.rst::L5212` the
`GRANULATED_WAVEFRONT_SGPR_COUNT`, which is `compute_pgm_rsrc1[6:9]` has
to be 0 for gfx10+ arch
---------
Co-authored-by: Matt Arsenault <Matthew.Arsenault@amd.com>
For each function with the AMDGPU_CS_Chain calling convention, with
dynamic VGPRs enabled, add a _dvgpr$ symbol, with the value of the
function symbol, plus an offset encoding one less than the number of
VGPR blocks used by the function (16 VGPRs per block, no more than 128)
in bits 5..3 of the symbol value. This is used by a front-end to have
functions that are chained rather than called, and a dispatcher that
dynamically resizes the VGPR count before dispatching to a function.
When computing the number of registers required by entry functions, the
`AMDGPUAsmPrinter` needs to take into account both the register usage
computed by the `AMDGPUResourceUsageAnalysis` pass, and the number
of registers initialized by the hardware. At the moment, the way it
computes the latter is different for graphics vs compute, due to differences in
the implementation. For kernels, all the information needed is available in
the `SIMachineFunctionInfo`, but for graphics shaders we would iterate over
the `Function` arguments in the `AMDGPUAsmPrinter`. This pretty much
repeats some of the logic from instruction selection.
This patch introduces 2 new members to `SIMachineFunctionInfo`, one
for SGPRs and one for VGPRs. Both will be computed during instruction
selection and then used during `AMDGPUAsmPrinter`, removing the need
to refer to the `Function` when printing assembly.
This patch is NFC except for the fact that we now add the extra SGPRs
(VCC, XNACK etc) to the number of SGPRs computed for graphics entry points.
I'm not sure why these weren't included before. It would be nice if
someone could confirm if that was just an oversight or if we have some docs
somewhere that I haven't managed to find. Only one test is affected (its SGPR
usage increases because we now take into account the XNACK registers).
Performance testing shows no significant gains or losses on graphics
workloads, so this is mostly to make the behavior consistent across all
supported OSes instead of special-casing HSA.
## Purpose
This patch is one in a series of code-mods that annotate LLVM’s public
interface for export. This patch annotates the `llvm/Target` library.
These annotations currently have no meaningful impact on the LLVM build;
however, they are a prerequisite to support an LLVM Windows DLL (shared
library) build.
## Background
This effort is tracked in #109483. Additional context is provided in
[this
discourse](https://discourse.llvm.org/t/psa-annotating-llvm-public-interface/85307),
and documentation for `LLVM_ABI` and related annotations is found in the
LLVM repo
[here](https://github.com/llvm/llvm-project/blob/main/llvm/docs/InterfaceExportAnnotations.rst).
A sub-set of these changes were generated automatically using the
[Interface Definition Scanner (IDS)](https://github.com/compnerd/ids)
tool, followed formatting with `git clang-format`.
The bulk of this change is manual additions of `LLVM_ABI` to
`LLVMInitializeX` functions defined in .cpp files under llvm/lib/Target.
Adding `LLVM_ABI` to the function implementation is required here
because they do not `#include "llvm/Support/TargetSelect.h"`, which
contains the declarations for this functions and was already updated
with `LLVM_ABI` in a previous patch. I considered patching these files
with `#include "llvm/Support/TargetSelect.h"` instead, but since
TargetSelect.h is a large file with a bunch of preprocessor x-macro
stuff in it I was concerned it would unnecessarily impact compile times.
In addition, a number of unit tests under llvm/unittests/Target required
additional dependencies to make them build correctly against the LLVM
DLL on Windows using MSVC.
## Validation
Local builds and tests to validate cross-platform compatibility. This
included llvm, clang, and lldb on the following configurations:
- Windows with MSVC
- Windows with Clang
- Linux with GCC
- Linux with Clang
- Darwin with Clang
…133242)"
This reverts commit 130080fab11cde5efcb338b77f5c3b31097df6e6 because it
causes issues in testcases similar to coalescer_remat.ll [1], i.e. when
we use a VGPR tuple but only write to its lower parts. The high VGPRs
would then not be included in the vgpr_count, and accessing them would
be an out of bounds violation.
[1]
https://github.com/llvm/llvm-project/blob/main/llvm/test/CodeGen/AMDGPU/coalescer_remat.ll
Don't count register uses when determining the maximum number of
registers used by a function. Count only the defs. This is really an
underestimate of the true register usage, but in practice that's not
a problem because if a function uses a register, then it has either
defined it earlier, or some other function that executed before has
defined it.
In particular, the register counts are used:
1. When launching an entry function - in which case we're safe because
the register counts of the entry function will include the register
counts of all callees.
2. At function boundaries in dynamic VGPR mode. In this case it's safe
because whenever we set the new VGPR allocation we take into account
the outgoing_vgpr_count set by the middle-end.
The main advantage of doing this is that the artificial VGPR arguments
used only for preserving the inactive lanes when using the
llvm.amdgcn.init.whole.wave intrinsic are no longer counted. This
enables us to allocate only the registers we need in dynamic VGPR mode.
---------
Co-authored-by: Thomas Symalla <5754458+tsymalla@users.noreply.github.com>
Register assembly printer passes in the pass registry.
This makes it possible to use `llc -start-before=<target>-asm-printer ...` in tests.
Adds a `char &ID` parameter to the AssemblyPrinter constructor to allow
targets to use the `INITIALIZE_PASS` macros and register the pass in the
pass registry. This currently has a default parameter so it won't break
any targets that have not been updated.
This implements the result of the discussion at:
https://discourse.llvm.org/t/rfc-report-fatal-error-and-the-default-value-of-gencrashdialog/73587
There are two different use cases for report_fatal_error, so replace it
with two functions reportFatalInternalError() and
reportFatalUsageError(). The former indicates a bug in LLVM and
generates a crash dialog. The latter does not. The names have been
suggested by rnk and people seemed to like them.
This replaces a lot of the usages that passed an explicit value for
GenCrashDiag. I did not bulk replace remaining report_fatal_error usage
-- they probably require case by case review for which function to use.
The CWSR trap handler needs to save and restore the VGPRs. When dynamic
VGPRs are in use, the fixed function hardware will only allocate enough
space for one VGPR block. The rest will have to be stored in scratch, at
offset 0.
This patch allocates the necessary space by:
- generating a prologue that checks at runtime if we're on a compute
queue (since CWSR only works on compute queues); for this we will have
to check the ME_ID bits of the ID_HW_ID2 register - if that is non-zero,
we can assume we're on a compute queue and initialize the SP and FP with
enough room for the dynamic VGPRs
- forcing all compute entry functions to use a FP so they can access
their locals/spills correctly (this isn't ideal but it's the quickest to
implement)
Note that at the moment we allocate enough space for the theoretical
maximum number of VGPRs that can be allocated dynamically (for blocks of
16 registers, this will be 128, of which we subtract the first 16, which
are already allocated by the fixed function hardware). Future patches
may decide to allocate less if they can prove the shader never allocates
that many blocks.
Also note that this should not affect any reported stack sizes (e.g. PAL
backend_stack_size etc).
This represents a hardware mode supported only for wave32 compute
shaders. When enabled, we set the `.dynamic_vgpr_en` field of
`.compute_registers` to true in the PAL metadata.
This will be changed to use an attribute after downstream consumers
have been migrated.
From GFX10 onwards it is possible to employ benevolent scheduling of
waves. This patch unconditionally enables, for the `amdhsa` OS, the bit
which controls that capability, as it is beneficial for algorithms that
rely on more complex concurrent coordination and it is generally
performance neutral otherwise.
Occupancy (i.e., the number of waves per EU) depends, in addition to
register usage, on per-workgroup LDS usage as well as on the range of
possible workgroup sizes. Mirroring the latter, occupancy should
therefore be expressed as a range since different group sizes generally
yield different achievable occupancies.
`getOccupancyWithLocalMemSize` currently returns a scalar occupancy
based on the maximum workgroup size and LDS usage. With respect to the
workgroup size range, this scalar can be the minimum, the maximum, or
neither of the two of the range of achievable occupancies. This commit
fixes the function by making it compute and return the range of
achievable occupancies w.r.t. workgroup size and LDS usage; it also
renames it to `getOccupancyWithWorkGroupSizes` since it is the range of
workgroup sizes that produces the range of achievable occupancies.
Computing the achievable occupancy range is surprisingly involved.
Minimum/maximum workgroup sizes do not necessarily yield maximum/minimum
occupancies i.e., sometimes workgroup sizes inside the range yield the
occupancy bounds. The implementation finds these sizes in constant time;
heavy documentation explains the rationale behind the sometimes
relatively obscure calculations.
As a justifying example, consider a target with 10 waves / EU, 4 EUs/CU,
64-wide waves. Also consider a function with no LDS usage and a flat
workgroup size range of [513,1024].
- A group of 513 items requires 9 waves per group. Only 4 groups made up
of 9 waves each can fit fully on a CU at any given time, for a total of
36 waves on the CU, or 9 per EU. However, filling as much as possible
the remaining 40-36=4 wave slots without decreasing the number of groups
reveals that a larger group of 640 items yields 40 waves on the CU, or
10 per EU.
- Similarly, a group of 1024 items requires 16 waves per group. Only 2
groups made up of 16 waves each can fit fully on a CU ay any given time,
for a total of 32 waves on the CU, or 8 per EU. However, removing as
many waves as possible from the groups without being able to fit another
equal-sized group on the CU reveals that a smaller group of 896 items
yields 28 waves on the CU, or 7 per EU.
Therefore the achievable occupancy range for this function is not [8,9]
as the group size bounds directly yield, but [7,10].
Naturally this change causes a lot of test churn as instruction
scheduling is driven by achievable occupancy estimates. In most unit
tests the flat workgroup size range is the default [1,1024] which,
ignoring potential LDS limitations, would previously produce a scalar
occupancy of 8 (derived from 1024) on a lot of targets, whereas we now
consider the maximum occupancy to be 10 in such cases. Most tests are
updated automatically and checked manually for sanity. I also manually
changed some non-automatically generated assertions when necessary.
Fixes#118220.
Add a prologue to the kernel entry to handle cases where code designed
for kernarg preloading is executed on hardware equipped with
incompatible firmware. If hardware has compatible firmware the 256 bytes
at the start of the kernel entry will be skipped. This skipping is done
automatically by hardware that supports the feature.
A pass is added which is intended to be run at the very end of the
pipeline to avoid any optimizations that would assume the prologue is a
real predecessor block to the actual code start. In reality we have two
possible entry points for the function. 1. The optimized path that
supports kernarg preloading which begins at an offset of 256 bytes. 2.
The backwards compatible entry point which starts at offset 0.
Resource usage information would try to overwrite unnamed functions if
there are multiple within the same compilation unit. This aims to either
use the `MCSymbol` assigned to the unnamed function (i.e.,
`CurrentFnSym`), or, rematerialize the `MCSymbol` for the unnamed
function.
When compiling multiple pipelines, the `MCRegisterInfo` instance in
`AMDGPUAsmPrinter` gets re-used even after finalization, so it calls
`finalize()` multiple times.
Add a reset method and call it in
`AMDGPUAsmPrinter::doFinalization`.
Different approach would be to make it a `unique_ptr`.
---------
Co-authored-by: Thomas Symalla <tsymalla@amd.com>
Converts AMDGPUResourceUsageAnalysis pass from Module to MachineFunction
pass. Moves function resource info propagation to to MC layer (through
helpers in AMDGPUMCResourceInfo) by generating MCExprs for every
function resource which the emitters have been prepped for.
Fixes https://github.com/llvm/llvm-project/issues/64863
