Jason Ekstrand [Sat, 27 Feb 2016 17:26:04 +0000 (09:26 -0800)]
anv/gen7: Only try to get the depth format the surface has depth
Jason Ekstrand [Wed, 24 Feb 2016 20:50:31 +0000 (12:50 -0800)]
anv/image: Use isl for filling brw_image_param
Jason Ekstrand [Wed, 24 Feb 2016 20:45:42 +0000 (12:45 -0800)]
isl: Add helpers for filling out brw_image_param
Jason Ekstrand [Wed, 24 Feb 2016 19:38:14 +0000 (11:38 -0800)]
anv: Fill out image_param structs at view creation time
Jason Ekstrand [Thu, 25 Feb 2016 03:49:12 +0000 (19:49 -0800)]
anv/image: Add a ussage_mask field to image_view_init
This allows us to avoid doing some unneeded work on the meta paths where we
know that the image view will be used for exactly one thing. The meta
paths also sometimes do things that aren't quite valid like setting the
array slice on a 3-D texture and we want to limit the number of paths that
need to be able to sensibly handle the lies.
Jason Ekstrand [Wed, 24 Feb 2016 19:14:15 +0000 (11:14 -0800)]
isl: Move isl_image.c to isl_storage_image.c
Jason Ekstrand [Tue, 23 Feb 2016 00:54:25 +0000 (16:54 -0800)]
anv: Use isl to fill buffer surface states
Jason Ekstrand [Tue, 23 Feb 2016 00:48:19 +0000 (16:48 -0800)]
isl: Add a helper for filling a buffer surface state
Jason Ekstrand [Sun, 21 Feb 2016 05:40:25 +0000 (21:40 -0800)]
anv: Remove unneeded fiels from anv_image_view
Jason Ekstrand [Sun, 21 Feb 2016 05:39:46 +0000 (21:39 -0800)]
anv/state: Remove unused fill_surface_state functions
Jason Ekstrand [Sat, 20 Feb 2016 19:45:50 +0000 (11:45 -0800)]
anv: Use ISL to fill out surface states
Jason Ekstrand [Sat, 27 Feb 2016 00:22:47 +0000 (16:22 -0800)]
anv/device: Store the default MOCS in the device
Jason Ekstrand [Sat, 20 Feb 2016 17:32:43 +0000 (09:32 -0800)]
isl: Add a function for filling out a surface state
Jason Ekstrand [Sat, 20 Feb 2016 17:22:07 +0000 (09:22 -0800)]
isl: Create per-gen helper libraries for gens 7, 8, and 9
Jason Ekstrand [Sat, 27 Feb 2016 00:11:33 +0000 (16:11 -0800)]
genxml: Add MOCS fields to RENDER_SURFACE_STATE
This allows us to set MOCS as a single uint32_t on all platforms.
Jason Ekstrand [Wed, 24 Feb 2016 23:51:36 +0000 (15:51 -0800)]
gen/genX_pack: Add genxml to the pack header path
If you have an out-of-tree build, gen8_pack.h and friends will not be in
the same folder as genX_pack.h so this will be a problem. We fixed
out-of-tree earlier by adding the genxml folder to the includes for the
vulkan driver. However, this is not a good long-term solution because we
want to use it in ISL as well.
Nanley Chery [Wed, 24 Feb 2016 20:10:44 +0000 (12:10 -0800)]
isl: Fix isl_surf_get_image_intratile_offset_el()
Consecutive tiles are separated by the size of the tile, not by the
logical tile width.
v2: Remove extra subtraction (Ville)
Add parenthesis (Jason)
v3: Update the unit tests for the function
Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Thomas Hindoe Paaboel Andersen [Thu, 25 Feb 2016 21:00:09 +0000 (22:00 +0100)]
anv: remove stray ; after if
Both logic and indentation suggests that the ; were not intended here.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Jason Ekstrand [Fri, 26 Feb 2016 20:37:04 +0000 (12:37 -0800)]
anv/gen8: Emit the 3DSTATE_PS_BLEND packet
Jason Ekstrand [Thu, 25 Feb 2016 16:52:35 +0000 (08:52 -0800)]
anv/device: Properly handle apiVersion == 0
From the Vulkan 1.0 spec section 3.2:
"If apiVersion is 0 the implementation must ignore it"
Kristian Høgsberg Kristensen [Wed, 24 Feb 2016 20:50:27 +0000 (12:50 -0800)]
Merge remote-tracking branch 'origin/master' into vulkan
Kristian Høgsberg Kristensen [Fri, 12 Feb 2016 05:11:48 +0000 (21:11 -0800)]
anv: Set max_hs_threads/max_ds_threads
Kenneth Graunke [Tue, 23 Feb 2016 01:28:22 +0000 (17:28 -0800)]
anv: Allocate more push constant space.
Previously we allocated 4kB of push constant space for VS, GS, and PS
(for a total of 12kB) no matter what. This works, but doesn't fully
utilize the space - we have 16kB or 32kB of space.
This makes anv use the same method as brw - divide up the space evenly
among all active shader stages. This means HS and DS would get space,
if those shader stages existed.
In the future, we can probably do better by inspecting how many push
constants each shader stage uses, and weight things accordingly. But
this is strictly better than the old code, and ideally we'd justify
a fancier solution with actual performance data.
Kenneth Graunke [Tue, 23 Feb 2016 01:26:15 +0000 (17:26 -0800)]
anv: Properly size the push constant L3 area.
We were assuming it was 32kB everywhere, reducing the available URB
space. It's actually 16kB on Ivybridge, Baytrail, and Haswell GT1-2.
Kenneth Graunke [Mon, 22 Feb 2016 23:46:23 +0000 (15:46 -0800)]
anv: Emit 3DSTATE_PUSH_CONSTANT_ALLOC_* via a loop.
Now we're emitting HS and DS packets as well.
Kenneth Graunke [Mon, 22 Feb 2016 23:23:06 +0000 (15:23 -0800)]
anv: Emit 3DSTATE_URB_* via a loop.
Rather than keeping separate {vs,hs,ds,gs}_start fields, we now store an
array indexed by the shader stage (MESA_SHADER_*). The 3DSTATE_URB_*
commands are also sequentially numbered. This makes it easy to just
emit them in a loop.
This simplifies the code a little, and also will make it easier to add
more credible HS and DS code later.
Brian Paul [Tue, 23 Feb 2016 20:46:28 +0000 (13:46 -0700)]
mesa: replace for loop with bitshifting in supported_buffer_bitmask()
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 20:41:27 +0000 (13:41 -0700)]
mesa: updates some comments in buffers.c
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 20:35:32 +0000 (13:35 -0700)]
mesa: make _mesa_draw_buffers() static
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 20:34:46 +0000 (13:34 -0700)]
mesa: make _mesa_draw_buffer() static
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 20:32:21 +0000 (13:32 -0700)]
mesa: make _mesa_read_buffer() static
Not called from any other file. Remove _mesa_ prefix and update comments.
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 20:25:43 +0000 (13:25 -0700)]
mesa: move declaration of buffer var in handle_first_current()
Declare the var in the scopes where it's used.
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 20:24:11 +0000 (13:24 -0700)]
mesa: use gl_buffer_index in a few places
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 20:19:29 +0000 (13:19 -0700)]
st/mesa: remove useless break statement
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 20:12:56 +0000 (13:12 -0700)]
st/mesa: rename st_readpixels to st_ReadPixels
To match the convention of other device driver functions.
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Brian Paul [Tue, 23 Feb 2016 23:15:34 +0000 (16:15 -0700)]
st/mesa: fix frontbuffer glReadPixels regressions
The change "mesa/readpix: Don't clip in _mesa_readpixels()" caused a
few piglit regressions. The failing tests use glReadPixels to read
from the front color buffer. The problem is we were trying to read
from a non-existant front color buffer. The front color buffer is
created on demand in st/mesa. Since the missing buffer bounds were
effectively 0 x 0 the glReadPixels was totally clipped and returned
early.
The fix involves creating the real front color buffer when we're about
to try reading from it.
Tested with llvmpipe and VMware driver on Linux, Windows.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94253
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94254
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94257
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Jason Ekstrand [Wed, 24 Feb 2016 06:02:41 +0000 (22:02 -0800)]
nir/spirv: Allow but warn for a few capabilities
Unfortunately, glslang gives us cull/clip distance and GS streams even if
the shader doesn't use it whenever a shader is declared as version 450.
This is a glslang bug, but we can easily enough ignore it for now.
Jason Ekstrand [Wed, 24 Feb 2016 05:23:06 +0000 (21:23 -0800)]
anv/descriptor_set: Use the correct size for the descriptor pool
The descriptor sizes array gives the total number of each type of
descriptor that will ever be allocated from the pool, not the total amount
that may be in any particular set. In our case, this simply means that we
have to sum a bunch of things up and there we go.
Jason Ekstrand [Wed, 24 Feb 2016 05:01:00 +0000 (21:01 -0800)]
nir/spirv: Add more capabilities
Jason Ekstrand [Wed, 24 Feb 2016 01:04:19 +0000 (17:04 -0800)]
anv/meta: Allocate descriptor pools on-the-fly
We can't use a global descriptor pool like we were because it's not
thread-safe. For now, we'll allocate them on-the-fly and that should work
fine. At some point in the future, we could do something where we
stack-allocate them or allocate them out of one of the state streams.
Oded Gabbay [Tue, 23 Feb 2016 14:33:45 +0000 (16:33 +0200)]
gallium/radeon: Correctly translate colorswaps for big endian
The current code in r600_translate_colorswap uses the swizzle information
to determine which colorswap to use.
This works for BE & LE when the nr_channels is <4, but when nr_channels==4
(e.g. PIPE_FORMAT_A8R8G8B8_UNORM), this method can not be used for both BE
and LE, because the swizzle info is the same for both of them.
As a result, r600g doesn't support 24bit color formats, only 16bit, which
forces the user to choose 16bit color in X server.
This patch fixes this bug by separating the checks for LE and BE and
adapting the swizzle conditions in the BE part of the checks.
Tested on an Evergreen GPU (Cedar GL FirePro 2270) running inside POWER7
Big-Endian Machine.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
CC: "11.2" "11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Thomas Hindoe Paaboel Andersen [Mon, 22 Feb 2016 22:58:00 +0000 (15:58 -0700)]
mesa: use sizeof on the correct type
Before the luminance stride was based on the size of GL_FLOAT
which is just the type constant (0x1406). Change it to use the
size of GLfloat.
Reviewed-by: Brian Paul <brianp@vmware.com>
Marek Olšák [Mon, 22 Feb 2016 23:37:39 +0000 (00:37 +0100)]
tgsi/scan: handle holes between VS inputs, assert-fail in other cases
"st/mesa: overhaul vertex setup for clearing, glDrawPixels, glBitmap"
added a vertex shader declaring IN[0] and IN[2], but not IN[1].
Drivers relying on tgsi_shader_info can't handle holes in declarations,
because tgsi_shader_info doesn't track that.
This is just a quick workaround meant for stable that will work for vertex
shaders.
This fixes radeonsi DrawPixels and CopyPixels crashes.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Brian Paul <brianp@vmware.com>
Jason Ekstrand [Tue, 23 Feb 2016 05:39:14 +0000 (21:39 -0800)]
anv/descriptor_set: Set descriptor type for immuatable samplers
Jason Ekstrand [Tue, 23 Feb 2016 05:20:04 +0000 (21:20 -0800)]
intel/genxml: Update macro documentation
Francisco Jerez [Mon, 22 Feb 2016 23:48:17 +0000 (15:48 -0800)]
docs: Mark off GL_OES_shader_image_atomic as done.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Francisco Jerez [Sun, 16 Aug 2015 22:52:19 +0000 (01:52 +0300)]
i965/fs: Return result of image atomic in a register of the expected type.
So the result is of float type if we're implementing the float
overload of imageAtomicExchange. This is the only back-end change
required to support OES_shader_image_atomic AFAICT.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Francisco Jerez [Fri, 21 Aug 2015 11:12:22 +0000 (14:12 +0300)]
glsl: Implement the required built-in functions when OES_shader_image_atomic is enabled.
This is basically just the same atomic functions exposed by
ARB_shader_image_load_store, with one exception:
"highp float imageAtomicExchange(
coherent IMAGE_PARAMS,
float data);"
There's no float atomic exchange overload in the original
ARB_shader_image_load_store or GL 4.2, so this seems like new
functionality that requires specific back-end support and a separate
availability condition in the built-in function generator.
v2: Move image availability predicate logic into a separate static
function for clarity. Had to pull out the image_function_flags
enum from the builtin_builder class for that to be possible.
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Francisco Jerez [Mon, 22 Feb 2016 23:06:45 +0000 (15:06 -0800)]
glsl: Add usual extension boilerplate for OES_shader_image_atomic.
v2: No need for extension enable bits (Ilia).
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Francisco Jerez [Mon, 22 Feb 2016 23:04:37 +0000 (15:04 -0800)]
mesa: Add extension table entry for OES_shader_image_atomic.
v2: No need for extension enable bits (Ilia).
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Jason Ekstrand [Tue, 23 Feb 2016 03:19:00 +0000 (19:19 -0800)]
anv/state: Replace a bunch of ANV_GEN with GEN_GEN
Jason Ekstrand [Tue, 23 Feb 2016 01:13:46 +0000 (17:13 -0800)]
anv/descriptor_set: Stop marking everything as having dynamic offsets
Kristian Høgsberg Kristensen [Fri, 12 Feb 2016 06:46:28 +0000 (22:46 -0800)]
anv: Implement descriptor pools
Descriptor pools are an optimization that lets applications allocate
descriptor sets through an externally synchronized object (that is,
unlocked). In our case it's also plugging a memory leak, since we
didn't track all allocated sets and failed to free them in
vkResetDescriptorPool() and vkDestroyDescriptorPool().
Kristian Høgsberg Kristensen [Mon, 22 Feb 2016 18:19:43 +0000 (10:19 -0800)]
anv/x11: Free swapchain images and memory on destroy
Samuel Pitoiset [Sun, 21 Feb 2016 10:58:35 +0000 (11:58 +0100)]
nvc0: rename 3d binding points to NVC0_BIND_3D_XXX
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Sun, 21 Feb 2016 10:49:25 +0000 (11:49 +0100)]
nvc0: rename 3d dirty flags to NVC0_NEW_3D_XXX
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Sun, 21 Feb 2016 10:15:43 +0000 (11:15 +0100)]
nvc0: prefix compute macros with _CP_ instead of _COMPUTE_
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Sun, 21 Feb 2016 10:13:01 +0000 (11:13 +0100)]
nvc0: rename NVXX_COMPUTE to NVXX_CP
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Sun, 21 Feb 2016 10:06:27 +0000 (11:06 +0100)]
nvc0: rename nvc0_context::dirty to nvc0_context::dirty_3d
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Mon, 22 Feb 2016 20:11:33 +0000 (21:11 +0100)]
nvc0/ir: add missing emission of locked load predicate
Like unlocked store on shared memory, locked store can fail and the
second dest which is a predicate must be emitted.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
Samuel Pitoiset [Sun, 21 Feb 2016 14:28:16 +0000 (15:28 +0100)]
nvc0/ir: add ld lock/st unlock emission on GK104
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Sun, 21 Feb 2016 18:40:01 +0000 (19:40 +0100)]
nv50/ir: restore OP_SELP to be a regular instruction
Actually OP_SELP doesn't need to be a compare instruction. Instead we
just need to set the NOT modifier when building the instruction.
While we are at it, fix the dst register type and use a GPR.
Suggested by Ilia Mirkin.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Mark Janes [Mon, 22 Feb 2016 19:31:15 +0000 (11:31 -0800)]
vulkan: fix out-of-tree builds
Brian Paul [Mon, 22 Feb 2016 15:53:47 +0000 (08:53 -0700)]
svga: unbind index buffer when drawing non-indexed primitives
Silences a warning reported by the svga3d device.
v2: also null-out the index buffer pointer
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Kristian Høgsberg Kristensen [Mon, 22 Feb 2016 17:14:25 +0000 (09:14 -0800)]
intel/genxml: Add README
I've had people ask about the design of the pack functions, for example,
why aren't we using bitfields. I wrote up a bit of background on why and
how we ended up with the current design and we might as well keep that
with the code.
Nanley Chery [Thu, 18 Feb 2016 22:05:31 +0000 (14:05 -0800)]
anv/meta_blit: Handle compressed textures in anv_CmdCopyImage
As with anv_CmdCopyBufferToImage, compressed textures require special
handling during copies.
Reviewed-by: Kristian Høgsberg Kristensen <kristian.h.kristensen@intel.com>
Ilia Mirkin [Wed, 17 Feb 2016 12:36:58 +0000 (07:36 -0500)]
mesa: add GL_EXT_texture_border_clamp support
This extension is identical to GL_OES_texture_border_clamp. But dEQP has
tests that want the EXT variant.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Ilia Mirkin [Tue, 16 Feb 2016 00:09:15 +0000 (19:09 -0500)]
mesa: add GL_OES_texture_border_clamp support
Only minor differences to the existing ARB_texture_border_clamp support.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Ilia Mirkin [Mon, 22 Feb 2016 14:06:59 +0000 (09:06 -0500)]
mesa: bump version
11.2 has been branched, we're on 11.3 now.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Emil Velikov [Mon, 22 Feb 2016 10:22:21 +0000 (10:22 +0000)]
nouveau: update the Makefile.sources list
Reflect the nv50->g80 change and the new gm107_texture header.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Jason Ekstrand [Mon, 22 Feb 2016 05:42:33 +0000 (21:42 -0800)]
nir/spirv: Add support for multisampled textures
Marek Olšák [Thu, 11 Feb 2016 14:49:34 +0000 (15:49 +0100)]
radeonsi: implement binary shaders & shader cache in memory (v2)
v2: handle _mesa_hash_table_insert failure
other cosmetic changes
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 11 Feb 2016 22:33:54 +0000 (23:33 +0100)]
gallium/radeon: remove unused radeon_shader_binary_free_* functions
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 11 Feb 2016 22:20:38 +0000 (23:20 +0100)]
radeonsi: make radeon_shader_reloc name string fixed-sized
This will simplify implementations of binary shaders.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 11 Feb 2016 20:09:38 +0000 (21:09 +0100)]
radeonsi: move some struct si_shader members to new struct si_shader_info
This will be part of shader binaries.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 11 Feb 2016 20:06:33 +0000 (21:06 +0100)]
radeonsi: use smaller types for some si_shader members
in order to decrease the shader size for a shader cache.
v2: add & use SI_MAX_VS_OUTPUTS
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 28 Jan 2016 00:29:59 +0000 (01:29 +0100)]
radeonsi: enable compiling one variant per shader
Shader stats from VERDE:
Default scheduler:
Totals:
SGPRS: 491272 -> 488672 (-0.53 %)
VGPRS: 289980 -> 311093 (7.28 %)
Code Size:
11091656 ->
11219948 (1.16 %) bytes
LDS: 97 -> 97 (0.00 %) blocks
Scratch:
1732608 ->
2246656 (29.67 %) bytes per wave
Max Waves: 78063 -> 77352 (-0.91 %)
Wait states: 0 -> 0 (0.00 %)
Looking at some of the worst regressions, I get:
- The VGPR increase seems to be caused by the fact that if PS has used less
than 16 VGPRs, now it will always use 16 VGPRs and sometimes even 20.
However, the wave count remains at 10 if VGPRs <= 24, so no harm there.
- The scratch increase seems to be caused by SGPR spilling.
The unnecessary SGPR spilling has been an ongoing issue with the compiler
and it's completely fixable by rematerializing s_loads or reordering
instructions.
SI scheduler:
Totals:
SGPRS: 374848 -> 374576 (-0.07 %)
VGPRS: 284456 -> 307515 (8.11 %)
Code Size:
11433068 ->
11535452 (0.90 %) bytes
LDS: 97 -> 97 (0.00 %) blocks
Scratch: 509952 -> 522240 (2.41 %) bytes per wave
Max Waves: 79456 -> 78217 (-1.56 %)
Wait states: 0 -> 0 (0.00 %)
VGPRs - same story as before. The SI scheduler doesn't spill SGPRs so much
and generally spills way less than the default scheduler.
(522240 spills vs
2246656 spills)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Fri, 5 Feb 2016 22:30:01 +0000 (23:30 +0100)]
radeonsi: print full shader name before disassembly
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 28 Jan 2016 01:53:13 +0000 (02:53 +0100)]
radeonsi: compile non-GS middle parts of shaders immediately if enabled
Still disabled.
Only prologs & epilogs are compiled in draw calls, but each variant of those
is compiled only once per process.
VS is always compiled as hw VS.
TES is always compiled as hw VS.
LS and ES stages are always compiled on demand.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Wed, 27 Jan 2016 01:59:23 +0000 (02:59 +0100)]
radeonsi: rework polygon stippling for PS prolog
Don't use the pstipple module.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Mon, 15 Feb 2016 22:57:54 +0000 (23:57 +0100)]
radeonsi: add PS prolog
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 23:50:29 +0000 (00:50 +0100)]
radeonsi: add PS epilog
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 23:38:03 +0000 (00:38 +0100)]
radeonsi: add TCS epilog
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 23:33:07 +0000 (00:33 +0100)]
radeonsi: add VS epilog
It only exports the primitive ID.
Also used by TES when it's compiled as VS.
The VS input location of the primitive ID input is v2.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 23:29:53 +0000 (00:29 +0100)]
radeonsi: add VS prolog
This is disabled with use_monolithic_shaders = true.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Mon, 15 Feb 2016 22:16:32 +0000 (23:16 +0100)]
radeonsi: first bits for non-monolithic shaders
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 21:15:27 +0000 (22:15 +0100)]
radeonsi: add code for dumping all shader parts together (v2)
v2: unify some code into si_get_shader_binary_size
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Marek Olšák [Tue, 26 Jan 2016 21:12:26 +0000 (22:12 +0100)]
radeonsi: add code for combining and uploading shaders from 3 shader parts
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Thu, 11 Feb 2016 21:04:35 +0000 (22:04 +0100)]
radeonsi: fail compilation if non-GS non-CS shaders have rodata
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 21:05:16 +0000 (22:05 +0100)]
radeonsi: separate 2 pieces of code from create_function
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 20:57:57 +0000 (21:57 +0100)]
radeonsi: add samplemask parameter to si_export_mrt_color
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 20:55:28 +0000 (21:55 +0100)]
radeonsi: add start_instance parameter to get_instance_index_for_fetch
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 26 Jan 2016 17:46:31 +0000 (18:46 +0100)]
radeonsi: separate out shader key bits for prologs & epilogs
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Sun, 3 Jan 2016 02:26:28 +0000 (03:26 +0100)]
radeonsi: compute how many input VGPRs fragment shaders have
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Fri, 1 Jan 2016 19:37:38 +0000 (20:37 +0100)]
radeonsi: compute how many input SGPRs and VGPRs shaders have
Prologs (shader binaries inserted before the API shader binary) need to
know this, so that they won't change the input registers unintentionally.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Marek Olšák [Tue, 22 Dec 2015 19:55:53 +0000 (20:55 +0100)]
gallium/radeon: add basic code for setting shader return values
LLVMBuildInsertValue will be used on return_value.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Samuel Pitoiset [Fri, 5 Feb 2016 12:32:59 +0000 (13:32 +0100)]
nvc0: enable compute shaders on Fermi
Kepler compute support is really different than Fermi and it's not
ready yet.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Tue, 26 Jan 2016 00:32:23 +0000 (01:32 +0100)]
nv50/ir: add atomics support on shared memory for Fermi
Changes from v3:
- move the previous OP_SELP change to the previous commit
Changes from v2:
- make sure the op is OP_SELP when emitting the predicate and add one
assert
- use bld.getSSA() for mkOp2()
- add cross edge between tryLockAndSetBB and joinBB
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Fri, 5 Feb 2016 14:09:07 +0000 (15:09 +0100)]
nv50/ir: make OP_SELP a compare instruction
This OP_SELP insn will be used to handle compare and swap subops.
Changes from v2:
- fix logic for GK110+
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Samuel Pitoiset [Mon, 25 Jan 2016 20:43:13 +0000 (21:43 +0100)]
nv50/ir: add lock/unlock subops for load/store
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>