This fixes a serious performance issue with DXVK:
https://github.com/doitsujin/dxvk/issues/937
This was caused by a recent change that to improve performance on RADV
which back-fired on ANV and killed performance for some apps:
https://github.com/doitsujin/dxvk/commit/e5a06d3f4a103a54cd4eb51970fedee405d1d698
Throwing in this bit of lowering lets us come along and CSE those UBO
loads (or copy-prop for SSBO load) and get one load where we previously
would have gotten several.
VkPipeline-db results on Kaby Lake:
total instructions in shared programs: 5115361 -> 5073185 (-0.82%)
instructions in affected programs: 1754333 -> 1712157 (-2.40%)
helped: 5331
HURT: 63
total cycles in shared programs: 2544501169 -> 2481144545 (-2.49%)
cycles in affected programs: 2531058653 -> 2467702029 (-2.50%)
helped: 9202
HURT: 4323
total loops in shared programs: 3340 -> 3331 (-0.27%)
loops in affected programs: 9 -> 0
helped: 9
HURT: 0
total spills in shared programs: 3246 -> 3053 (-5.95%)
spills in affected programs: 384 -> 191 (-50.26%)
helped: 10
HURT: 5
total fills in shared programs: 4626 -> 4452 (-3.76%)
fills in affected programs: 439 -> 265 (-39.64%)
helped: 10
HURT: 5
All of the shaders with hurt spilling were in Rise of the Tomb Raider
which also had shaders solidly helped in the spilling department. Not
shown in those results (because I've not had success dumping the
shaders) is Witcher 3 where this reduces spilling and improves over-all
perf by around 20-25%. There were no shader-db changes. Apparently,
this just isn't a pattern that happens in OpenGL.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Cc: "19.0" mesa-stable@lists.freedesktop.org
Instead, we do UBO and SSBO deref lowering in NIR after we've given it a
chance to optimize SSBO access:
Shader-db results on Kaby Lake:
total instructions in shared programs: 15235775 -> 15235484 (<.01%)
instructions in affected programs: 14992 -> 14701 (-1.94%)
helped: 19
HURT: 20
total cycles in shared programs: 339220331 -> 339027307 (-0.06%)
cycles in affected programs: 79831981 -> 79638957 (-0.24%)
helped: 540
HURT: 602
total loops in shared programs: 4402 -> 4348 (-1.23%)
loops in affected programs: 186 -> 132 (-29.03%)
helped: 27
HURT: 0
total spills in shared programs: 23261 -> 23234 (-0.12%)
spills in affected programs: 38 -> 11 (-71.05%)
helped: 1
HURT: 0
total fills in shared programs: 31442 -> 31371 (-0.23%)
fills in affected programs: 98 -> 27 (-72.45%)
helped: 1
HURT: 0
LOST: 12
GAINED: 12
Most of the help and hurt in instruction counts was just churn caused by
re-ordering of optimizations and the fact that the NIR deref lowering
code is emitting slightly different instructions. Nothing was hurt by
more than three instructions and most things weren't helped by more than
four. The primary exception to this is one Car Chase shader:
shaders/non-free/gfxbench4/carchase/341.shader_test CS SIMD32: 1144 -> 821 (-28.23%)
There is also one compute shader in Manhattan 3.1 and a fragment shader
in the UE4 Shooter Game demo that now get a loop partially unrolled.
Those showed up in the results as hurt instructions but were manually
removed to get the results above.
The lost/gained was a dozen Car Chase shaders that went from SIMD8 to
SIMD16 thanks to improved register pressure:
shaders/non-free/gfxbench4/carchase/366.shader_test CS
shaders/non-free/gfxbench4/carchase/368.shader_test CS
shaders/non-free/gfxbench4/carchase/370.shader_test CS
shaders/non-free/gfxbench4/carchase/372.shader_test CS
shaders/non-free/gfxbench4/carchase/376.shader_test CS
shaders/non-free/gfxbench4/carchase/378.shader_test CS
shaders/non-free/gfxbench4/carchase/380.shader_test CS
shaders/non-free/gfxbench4/carchase/382.shader_test CS
shaders/non-free/gfxbench4/carchase/384.shader_test CS
shaders/non-free/gfxbench4/carchase/388.shader_test CS
shaders/non-free/gfxbench4/carchase/4.shader_test CS
shaders/non-free/gfxbench4/carchase/6.shader_test CS
Given how much it appeared to be improved, I ran Car Chase on my laptop.
Unfortunately, I wasn't able to see any measurable improvement. It
might be helped by 1-2% but it's in the noise. It does render correctly
as far as I can tell so the improvement is legitimate.
All of the loops that got delete were in dolphin uber shaders. I've had
no opportunity to test them for correctness or performance.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
It's just a 32-bit index and offset. We're going to want to use it in
GL as well so stop talking about Vulkan.
Reviewed-by: Kristian H. Kristensen <hoegsberg@chromium.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We don't set it on HSW and earlier in i965 and disabling it appears to
make derivatives somewhat more reliable.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
We've been fairly inconsistent about this so we should really choose
whether we're going to use VK_TRUE/FALSE or the C boolean values. The
Vulkan #defines are set to 1 and 0 respectively so it's the same value
as C gives you when you cast a boolean expression to an integer. Since
there are several places where we set a VkBool32 to a C logical
expression, let's just embrace C booleans and stop using the VK defines.
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Due to lack of write mask in SPIR-V store, generators may produce
multiple stores to the same vector but using different array derefs.
Use the combining store pass to clean this up. For example,
layout(binding = 3) buffer block {
vec4 v;
};
void main() {
v.x = 11;
v.y = 22;
}
after going to SPIR-V and NIR, ends up with in two store_derefs to
v[0] and v[1]
vec2 32 ssa_4 = deref_struct &ssa_3->field0 (ssbo vec4) /* &((block *)ssa_2)->field0 */
vec2 32 ssa_6 = deref_array &(*ssa_4)[0] (ssbo float) /* &((block *)ssa_2)->field0[0] */
intrinsic store_deref (ssa_6, ssa_7) (1, 0) /* wrmask=x */ /* access=0 */
vec1 32 ssa_13 = load_const (0x00000001 /* 0.000000 */)
vec2 32 ssa_14 = deref_array &(*ssa_4)[1] (ssbo float) /* &((block *)ssa_2)->field0[1] */
intrinsic store_deref (ssa_14, ssa_15) (1, 0) /* wrmask=x */ /* access=0 */
producing two different sends instructions in skl. The combining pass
transform the snippet above into
vec2 32 ssa_4 = deref_struct &ssa_3->field0 (ssbo vec4) /* &((block *)ssa_2)->field0 */
vec4 32 ssa_18 = vec4 ssa_7, ssa_15, ssa_16, ssa_17
intrinsic store_deref (ssa_4, ssa_18) (3, 0) /* wrmask=xy */ /* access=0 */
producing a single sends instruction.
v2: Move this from spirv_to_nir into the general optimization pass for
intel compiler. (Jason)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
v2: (all from Jason)
Reuse existing function for the end of the block combinations.
Check the SSA values are coming from the right place in tests.
Document the case when the store to array_deref is reused.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We were not copying the saturate bit from the original instruction
to the new replacement instruction. This caused major misrendering
in DiRT Rally on iris, where comparisons leading to discards failed
due to the missing saturate, causing lots of extra garbage pixels to
be drawn in text rendering, trees, and so on.
This did not show up on i965 because st/nir performs a more aggressive
version of nir_opt_peephole_select, yielding more b32csel operations.
Fixes: 52c7df1643 i965/fs: Merge CMP and SEL into CSEL on Gen8+
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The IO scalarization pass that we run to help with linking end up
turning some shader I/O such as that for tessellation and geometry
shaders into many scalar URB operations rather than one vector one. To
alleviate this, we now vectorize the I/O once again. This fixes a 10%
performance regression in the GfxBench tessellation test that was caused
by scalarizing.
Shader-db results on Kaby Lake:
total instructions in shared programs: 15224023 -> 15220871 (-0.02%)
instructions in affected programs: 342009 -> 338857 (-0.92%)
helped: 1236
HURT: 443
total spills in shared programs: 23471 -> 23465 (-0.03%)
spills in affected programs: 6 -> 0
helped: 1
HURT: 0
total fills in shared programs: 31770 -> 31766 (-0.01%)
fills in affected programs: 4 -> 0
helped: 1
HURT: 0
Cycles was just a lot of churn do to moves being different places. Most
of the pure churn in instructions was +/- one or two instructions in
fragment shaders.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107510
Fixes: 4434591bf5 "intel/nir: Call nir_lower_io_to_scalar_early"
Fixes: 8d8222461f "intel/nir: Enable nir_opt_find_array_copies"
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
This reverts commit 47fc359822.
Reason is that patch did not take in to account situation where we might
have both OpenGL and Vulkan using glsl_types at the same time.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
As stated in Vulkan spec:
"Resetting a descriptor pool recycles all of the resources from all
of the descriptor sets allocated from the descriptor pool back to
the descriptor pool, and the descriptor sets are implicitly freed."
This fixes dEQP-VK.api.descriptor_pool.*
Fixes: 14f6275c92 "anv/descriptor_set: add reference counting for..."
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Clayton Craft <clayton.a.craft@intel.com>
Fixes leaks for each glsl_type generated:
==32470== 384 bytes in 3 blocks are possibly lost in loss record 18 of 18
==32470== at 0x483880B: malloc (vg_replace_malloc.c:309)
==32470== by 0x4C43F4A: ralloc_size (ralloc.c:119)
==32470== by 0x4C44014: rzalloc_size (ralloc.c:151)
==32470== by 0x4C44258: rzalloc_array_size (ralloc.c:215)
==32470== by 0x4D38957: glsl_type::glsl_type(glsl_struct_field const*, unsigned int, char const*) (glsl_types.cpp:114)
==32470== by 0x4D3BEED: glsl_type::get_struct_instance(glsl_struct_field const*, unsigned int, char const*) (glsl_types.cpp:1146)
==32470== by 0x4D42ECC: glsl_struct_type (nir_types.cpp:501)
==32470== by 0x4CDB5A1: vtn_handle_type (spirv_to_nir.c:1269)
==32470== by 0x4CE53DD: vtn_handle_variable_or_type_instruction (spirv_to_nir.c:4018)
==32470== by 0x4CD8CFF: vtn_foreach_instruction (spirv_to_nir.c:365)
==32470== by 0x4CE5E6B: spirv_to_nir (spirv_to_nir.c:4490)
==32470== by 0x497AF10: anv_shader_compile_to_nir (anv_pipeline.c:173)
v2: move release call to vkDestroyInstance
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes following leak:
==21853== 32 bytes in 1 blocks are definitely lost in loss record 2 of 20
==21853== at 0x483AB1A: calloc (vg_replace_malloc.c:762)
==21853== by 0x4C4DD7F: util_vma_heap_free (vma.c:221)
==21853== by 0x4C4D647: util_vma_heap_init (vma.c:46)
==21853== by 0x4957B9F: anv_CreateDescriptorPool (anv_descriptor_set.c:578)
Fixes: c520f4dec9 ("anv: Add a concept of a descriptor buffer")
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Patch maintains a list of sets in the pool and destroys possible
remaining sets when pool is destroyed.
As stated in Vulkan spec:
"When a pool is destroyed, all descriptor sets allocated from
the pool are implicitly freed and become invalid."
This fixes memory leaks spotted with valgrind:
==19622== 96 bytes in 1 blocks are definitely lost in loss record 2 of 3
==19622== at 0x483880B: malloc (vg_replace_malloc.c:309)
==19622== by 0x495B67E: default_alloc_func (anv_device.c:547)
==19622== by 0x4955E05: vk_alloc (vk_alloc.h:36)
==19622== by 0x4956A8F: anv_multialloc_alloc (anv_private.h:538)
==19622== by 0x4956A8F: anv_CreateDescriptorSetLayout (anv_descriptor_set.c:217)
Fixes: 14f6275c92 ("anv/descriptor_set: add reference counting for descriptor set layouts")
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Now that nir_opt_copy_prop_vars can properly handle array derefs on
vectors, it's safe to move UBO and SSBO lowering to late in the
pipeline. This should allow NIR to actually start optimizing SSBO
access.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
It doesn't really matter where this pass goes as long as it's after we
call nir_lower_explicit_io and before we go into the back-end. Putting
it brw_postprocess_nir lets us move nir_lower_explicit_io significantly
later in the pipeline.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
In order to be used for OpenGL (right now for ARB_gl_spirv).
This commit adds two new structures:
* nir_xfb_varying_info: that identifies each individual varying. For
each one, we need to know the type, buffer and xfb_offset
* nir_xfb_buffer_info: as now for each buffer, in addition to the
stride, we need to know how many varyings are assigned to it.
For this patch, the only case where num_outputs != num_varyings is
with the case of doubles, that for dvec3/4 could require more than one
output. There are more cases though (like aoa), that will be handled
on following patches.
v2: updated after new nir general XFB support introduced for "anv: Add
support for VK_EXT_transform_feedback"
v3: compute num_varyings beforehand for allocating, instead of relying
on num_outputs as approximate value (Timothy Arceri)
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Found out that some base64 data matched the '---' identifier. We can
avoid this by adding the surrounding spaces.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
The error state contains several kind of BOs, including the context
image which we will want to write in a later commit. Because it can
come later in the error state than the user buffers and because we
need to write it first in the aub file, we have to first build a list
of BOs and then write them in the appropriate order.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Unlike most of the cases in which we do this by hand, the new helper
properly handles non-32-bit pointers.
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This is necessary for legacy texture buffer object formats, where we'll
need to use a swizzle to fake e.g. luminance.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We allocate GGTT entries and physical addresses are we create engines
rather than having a fixed layout.
Context images now receive a parameter argument which is used to setup
pml4 & ring buffer addresses.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
We'll make them more parameterized in a later commit.
As this is just a transitional commit, we allow ourself to leak the
context images allocated in get_context_init(). We'll fix this in the
next commit.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Prepare aub write to deal with multiple engine instances. We don't
pass the instance number yet this could be done in the future by
having a 2 dimensional array of struct engine.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Rafael Antognolli <rafael.antognolli@intel.com>
In the future we'll want error2aub to reuse the context image saved by
i915 instead of the default one we write in intel_dump_gpu.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
IGT has a test to hang the GPU that works by having a batch buffer
jump back into itself, trigger an infinite loop on the command stream.
As our implementation of the decoding is "perfectly" mimicking the
hardware, our decoder also "hangs". This change limits the number of
batch buffer we'll decode before we bail to 100.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
An MI_BATCH_BUFFER_START in the ring buffer acts as a second level
batchbuffer (aka jump back to ring buffer when running into a
MI_BATCH_BUFFER_END).
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Fixes leaks from anv_device_upload_nir:
==7345== 8,192 bytes in 2 blocks are definitely lost in loss record 24 of 24
==7345== at 0x4C2ED78: malloc (vg_replace_malloc.c:308)
==7345== by 0x4C31393: realloc (vg_replace_malloc.c:836)
==7345== by 0x54E0848: grow_to_fit (blob.c:67)
==7345== by 0x54E0BE5: blob_reserve_bytes (blob.c:166)
==7345== by 0x54E0C7C: blob_reserve_intptr (blob.c:186)
==7345== by 0x54704A7: nir_serialize (nir_serialize.c:1091)
==7345== by 0x512F97D: anv_device_upload_nir (anv_pipeline_cache.c:756)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Jason Ekstrand