shader_storage_blocks_write_access was computed using the buffer indices
in the program but ShaderStorageBlocksWriteAccess is used with the shader
buffers.
So if a VS had 3 SSBOs and a FS had 4, the mask for VS was 0x3 (correct) but
the mask for the FS was 0x78 instead of 0x15.
Fix this by substracting the index of the first shader buffer in the program's
buffers.
Fixes: 79127f8d5b ("glsl: set ShaderStorageBlocksWriteAccess in the nir linker")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6184
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15552>
In general, an atomic intrinsic may perform separate atomics for every
enabled SIMD channel, as each channel may operate on different memory.
However, an extremely common case is for all channels to access the same
memory location. In this case, we can simply perform a reduction/scan
across the subgroup, and perform one atomic for the whole subgroup,
rather than one per channel. For example, if an intrinsic says to take
the minimum value of the existing memory and the value in each channel,
we can do a thread-local minimum of all enabled channels, then do a
single atomic to take the minimum of that and the existing memory.
Our hardware doesn't optimize the case where multiple channels ask for
atomics on the same memory location; it assumes the compiler will do so.
nir_opt_uniform_atomics() uses divergence analysis to detect this case,
adds the necessary subgroup operations, and moves the atomic inside a
conditional that disables all but a single invocation. It even detects
cases where the shader code already performs this kind of optimization,
and avoids doing it a second time.
This may not be the optimal solution for us. In the backend, we could
detect this case and emit send(1) instructions with NoMask, rather than
generating if...send(16)...endif, and a lot of unnecessary ALU ops. But
it's simple to do, reuses the same path as ACO, and still provides most
of the benefit by cutting up to 16x atomics down to a single atomic,
which is more merciful to the memory bus.
Improves performance of Shadow of the Tomb Raider by 5.5% on XeHP.
Improves performance of a customer-internal benchmark on XeHP at
3840x2160 and low settings by approximately 30%.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15484>
We haven't exposed this intrinsic as it doesn't directly correspond to
anything in SPIR-V. However, it's used internally by some NIR passes,
namely nir_opt_uniform_atomics().
We reuse most of the infrastructure in brw_find_live_channel, but with
LZD/ADD instead of FBL. A new SHADER_OPCODE_FIND_LAST_LIVE_CHANNEL is
like SHADER_OPCODE_FIND_LIVE_CHANNEL but from the other side.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15484>
- load_reloc_const is just an immediate constant load, it's convergent.
- nir_intrinsic_load_global_const_block_intel should be convergent,
it says the address must be uniform, and we uniformize the predicate
- Lowered image intrinsics: image_deref_load_param_intel just reads
information about an image, as long as the image variable is
convergent it should be too. load_raw_intel...if the address we
come up with is convergent, it ought to be as well.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15484>
Instead of reusing the in/out slot mechanism, use a separated NIR
variable mode. This will make easier later to implement staging the
output in shared memory (and storing all at the end to the URB).
Note to get 64-bit type support we currently rely on the
brw_nir_lower_mem_access_bit_sizes() pass.
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15022>
The only ways that function can return NULL are:
- the xcb connection was closed
- the window for the swapchain was destroyed
- the special event listener was unregistered from another thread
- malloc failure
All of these are permanent errors, the swapchain is no longer in a
usable state, so we should treat this as VK_ERROR_SURFACE_LOST_KHR.
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15558>
Logic is lifted from bi_layout.c, adapted to work on instructions (not
clauses) and for Valhall's off-by-one semantic which is annoyingly
different than Bifrost. (But the same as Midgard -- Bifrost was
annoyingly different than Midgard!)
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15223>
We already collect enums in the ISA description XML. Export them for use in the
compiler backend, particularly the packing code.
Usually we'd use Mako for templating. In this case, the script is so trivial a
template engine didn't seem worth it. (The obvious version with Mako was about
10 lines longer than just prints and f-strings used here.)
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Suggested-by: Icecream95 <ixn@disroot.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15223>
Georg Lehmann