summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
* gallium: add PIPE_CAP_TGSI_VOTE for when the VOTE ops are allowedIlia Mirkin2016-06-0617-0/+17
| | | | | Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Dave Airlie <[email protected]>
* gallium: add VOTE_* opcodes to implement GL_ARB_shader_group_voteIlia Mirkin2016-06-063-1/+26
| | | | | Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Dave Airlie <[email protected]>
* mesa: hook up core bits of GL_ARB_shader_group_voteIlia Mirkin2016-06-0612-1/+67
| | | | | | Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Ian Romanick <[email protected]> Reviewed-by: Dave Airlie <[email protected]>
* glsl: Make opt_copy_propagation_elements actually propagate into loops.Kenneth Graunke2016-06-061-2/+17
| | | | | | | | | | | | | | | | | | | | | | We've had a FINISHME here since Eric originally wrote the code in 2011. This patch implements his suggested approach, which makes us actually able to copy propagate into the loops, at the unfortunate cost of making this pass even more expensive. The shader-db statistics are basically a wash: No change in instruction counts. total cycles in shared programs: 78685980 -> 78680730 (-0.01%) cycles in affected programs: 2102646 -> 2097396 (-0.25%) helped: 48 HURT: 83 I figured if we're going to do this for one copy propagation pass, we may as well do it in both. Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* glsl: Make opt_copy_propagation actually propagate into loops.Kenneth Graunke2016-06-061-6/+24
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | We've had a FINISHME here since Eric originally wrote the code in 2010. This patch implements his suggested approach, which makes us actually able to copy propagate into the loops, at the unfortunate cost of making this pass even more expensive. The shader-db statistics are not terribly impressive: total instructions in shared programs: 9008589 -> 9008613 (0.00%) instructions in affected programs: 4293 -> 4317 (0.56%) helped: 0 HURT: 10 total cycles in shared programs: 78550978 -> 78575760 (0.03%) cycles in affected programs: 655426 -> 680208 (3.78%) helped: 75 HURT: 88 GAINED: 2 Most of the "regressions" appear to be us successfully copy propagating uniforms, which i965 is loading as pull constants instead of push, so we occasionally have two pulls instead of one. That doesn't seem like this pass's job - it's propagating correctly, and we should be smarter about pull loads in the backend. This patch is also useful for a couple of reasons: 1. It can clean up copies created by varying packing (previously, we couldn't if the uses were inside a loop). This fixes a bug when interpolateAt*() is used on a packed varying inside a loop: glsl_to_nir struggles to see through the extra copy and mistakenly believed the variable was not an input. 2. It will help propagate uniform array access created by lower_const_array_to_uniforms(). Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* nv50/ir: use round toward 0 when converting doubles to integersSamuel Pitoiset2016-06-061-1/+3
| | | | | | | | | | | | Like floats, we should use the round toward 0 mode instead of the nearest one (which is the default) for doubles to integers. This fixes all arb_gpu_shader_fp64 piglits which convert doubles to integers (16 tests). Signed-off-by: Samuel Pitoiset <[email protected]> Reviewed-by: Ilia Mirkin <[email protected]> Cc: "11.2 12.0" <[email protected]>
* gallium/radeon: don't re-set BO metadata after CMASK deallocationMarek Olšák2016-06-061-1/+0
| | | | | | CMASK has no effect on metadata, because it's not sharable. Reviewed-by: Nicolai Hähnle <[email protected]>
* st/mesa: change SQRT lowering to fix the game RisenMarek Olšák2016-06-061-7/+9
| | | | | | | Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94627 (against nouveau) Acked-by: Ilia Mirkin <[email protected]>
* radeonsi: add a performance tweak for 4 SE partsMarek Olšák2016-06-061-0/+11
| | | | | | Ported from Vulkan. Reviewed-by: Nicolai Hähnle <[email protected]>
* radeonsi: simplify PRIMGROUP_SIZE computation for tessellationMarek Olšák2016-06-061-9/+1
| | | | | | | | Ported from Vulkan. v2: keep the comment Reviewed-by: Nicolai Hähnle <[email protected]>
* r600g: use hw MSAA resolve for non-trivial resolvesMarek Olšák2016-06-061-9/+53
| | | | | | This improves MSAA resolve performance. Reviewed-by: Nicolai Hähnle <[email protected]>
* radeonsi: use hw MSAA resolve for non-trivial resolvesMarek Olšák2016-06-061-10/+54
| | | | | | This improves MSAA resolve performance. Reviewed-by: Nicolai Hähnle <[email protected]>
* mesa/program_resource: return -1 for index if no location.Dave Airlie2016-06-071-3/+16
| | | | | | | | | | | | | | | | | The GL4.5 spec quote seems clear on this: "The value -1 will be returned by either command if an error occurs, if name does not identify an active variable on programInterface, or if name identifies an active variable that does not have a valid location assigned, as described above." This fixes: GL45-CTS.program_interface_query.output-built-in [airlied: use _mesa_program_resource_location_index as suggested by Eduardo] Reviewed-by: Eduardo Lima Mitev <[email protected]> Cc: "12.0" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* radeonsi: set descriptor dirty mask on shader buffer unbindNicolai Hähnle2016-06-061-0/+1
| | | | | | | | Found randomly while skimming the code. This might have caused VM faults in robustness tests. Cc: 12.0 <[email protected]> Reviewed-by: Marek Olšák <[email protected]>
* st/mesa: fix resource leak in try_pbo_readpixelsNicolai Hähnle2016-06-061-1/+1
| | | | | | | Found by inspection after seeing https://bugs.freedesktop.org/show_bug.cgi?id=96343 Reviewed-by: Marek Olšák <[email protected]>
* tgsi: fix mixed data type comparison in tgsi_point_sprite.cCharmaine Lee2016-06-061-3/+3
| | | | | | | | | | | | | | | | Cast the unsigned semantic index to integer datatype before comparing to max_generic, otherwise, max_generic which is initialized to -1 will be converted to unsigned int before the comparison, causing a wrong semantic index to be assigned to a shader output. Fixes the assert running TurboCAD_gl.trace. (VMware bug 1667265) Also tested with glretrace, mesa demos pointblast, spriteblast and pointcoord. v2: use the original max_generic variable but add the (int) cast to the semantic index, as suggested by Brian. Reviewed-by: Brian Paul <[email protected]>
* svga: print shader linkage info when tgsi debug bit is onCharmaine Lee2016-06-061-2/+5
| | | | | | | | When TGSI debug flag is enabled, print the shader linkage info as well. Tested with mesa demos with SVGA_DEBUG=tgsi Reviewed-by: Brian Paul <[email protected]>
* st/mesa: check shader image format support before using PBO downloadIlia Mirkin2016-06-061-0/+6
| | | | | | | | | | | ARB_shader_image_load_store only requires a very fixed list of formats to be supported, while textures may be in all kinds of formats, like BGRA which are presently not supported on at least Kepler. Signed-off-by: Ilia Mirkin <[email protected]> Tested-by: Samuel Pitoiset <[email protected]> Reviewed-by: Nicolai Hähnle <[email protected]> Reviewed-by: Marek Olšák <[email protected]>
* tgsi: use truncf in micro_truncLars Hamre2016-06-061-4/+4
| | | | | | | | | | | | | | | | | | | | | | | | | Switches to using truncf in micro_trunc. Fixes the following piglit tests (for softpipe): /spec/glsl-1.30/execution/built-in-functions/... fs-trunc-float fs-trunc-vec2 fs-trunc-vec3 fs-trunc-vec4 vs-trunc-float vs-trunc-vec2 vs-trunc-vec3 vs-trunc-vec4 /spec/glsl-1.50/execution/built-in-functions/... gs-trunc-float gs-trunc-vec2 gs-trunc-vec3 gs-trunc-vec4 Signed-off-by: Lars Hamre <[email protected]> Reviewed-by: Roland Scheidegger <[email protected]>
* i965/gs/scalar: Fix load input for doublesSamuel Iglesias Gonsálvez2016-06-061-18/+54
| | | | | | Signed-off-by: Samuel Iglesias Gonsálvez <[email protected]> Reviewed-by: Timothy Arceri <[email protected]> Cc: "12.0" <[email protected]>
* i965/fs: fix offset when loading double vector input varyingsSamuel Iglesias Gonsálvez2016-06-061-1/+21
| | | | | | | | | | | | | | | | | When we are not packing a double input varying, we might need to read its data in a non-aligned to 64-bit offset, so we read the wrong data. This is happening when using explicit locations in varyings because Mesa disables packing varying for that case. const_index is in 32-bit size units but offset() is multiplying it by destination type size units. When operating with double input varyings, const_index value could be not aligned to 64 bits. To fix it, we load the double vector as if it was a float based vector with twice the number of components. Signed-off-by: Samuel Iglesias Gonsálvez <[email protected]> Reviewed-by: Timothy Arceri <[email protected]> Cc: "12.0" <[email protected]>
* i965/fs: fix FS_OPCODE_CINTERP for unpacked double input varyingsSamuel Iglesias Gonsálvez2016-06-061-1/+12
| | | | | | | | | | | Data starts at suboffet 3 in 32-bit units (12 bytes), so it is not 64-bit aligned and the current implementation fails to read the data properly. Instead, when there is is a double input varying, read it as vector of floats with twice the number of components. Signed-off-by: Samuel Iglesias Gonsálvez <[email protected]> Reviewed-by: Timothy Arceri <[email protected]> Cc: "12.0" <[email protected]>
* glsl: geom shader max_vertices layout must match.Dave Airlie2016-06-063-9/+13
| | | | | | | | | | | | | From GLSL 4.5 spec, "4.4.2.3 Geometry Outputs". "all geometry shader output vertex count declarations in a program must declare the same count." Fixes: GL45-CTS.geometry_shader.output.conflicted_output_vertices_max Reviewed-by: Alejandro Piñeiro <[email protected]> Cc: "11.2 12.0" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* anv/pipeline: Add support for caching the push constant mapJason Ekstrand2016-06-061-4/+29
| | | | | Signed-off-by: Jason Ekstrand <[email protected]> Reviewed-by: Kristian Høgsberg <[email protected]>
* glsl: use enum glsl_interface_packing in more places. (v2)Dave Airlie2016-06-0611-36/+46
| | | | | | | | | | | | | | | Although the glsl_types.h stores this in a bitfield, we should hide that from everyone else. Hide the cast in an accessor method and use the enum everywhere. This makes things a bit nicer in gdb, and improves type safety. v2: fix a few pieces of interface I missed that caused some piglit regressions. Signed-off-by: Dave Airlie <[email protected]> Reviewed-by: Ilia Mirkin <[email protected]>
* i965: don't use NumLayers for 3D textures.Dave Airlie2016-06-061-1/+1
| | | | | | | | | | | | For 3D textures we shouldn't be using NumLayers, we need to get it from the depth. This fixes: GL45-CTS.geometry_shader.layered_framebuffer.clear_call_support Reviewed-by: Eduardo Lima Mitev <[email protected]> Cc: "12.0" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* glsl: for anonymous struct matching use without_array() (v3)Dave Airlie2016-06-061-3/+20
| | | | | | | | | | | | | | | | | With tessellation shaders we can have cases where we have arrays of anon structs, so make sure we match using without_array(). Fixes: GL45-CTS.tessellation_shader.tessellation_control_to_tessellation_evaluation.gl_in v2: test lengths match as well (Ilia) v3: descend array lengths to check for matches as well (Ilia) Reviewed-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* glsl/ast: don't crash when func_name is NULLDave Airlie2016-06-061-0/+4
| | | | | | | | | | | | This fixes a crash in GL43-CTS.shader_subroutine.subroutines_not_allowed_as_variables_constructors_and_argument_or_return_types If we can't find the func_name in one of these paths, we have emitted an earlier error so just return here. Reviewed-by: Timothy Arceri <[email protected]> Cc: "11.2 12.0" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* glsl: handle ast_aggregate in has_sequence_subexpression. (v2)Dave Airlie2016-06-061-1/+1
| | | | | | | | | | | | | | | GL43-CTS.compute_shader.work-group-size does uniform uint g_uniform[gl_WorkGroupSize.z + 20] = { 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24 }; The initializer triggers the GLSL 4.30/GLES3 tests for constant sequence subexpressions, so it doesn't happen unless you are using those, so just return false as this path is now reachable. v2: update commit msg with diagnosis Acked-by: Timothy Arceri <[email protected]> Cc: "11.2 12.0" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* mesa: Try to unbreak the MSVC build.Kenneth Graunke2016-06-052-0/+8
| | | | | | | PATH_MAX is apparently not a thing on Windows. Borrow the hack from pipe_loader.c to try and make this work. Signed-off-by: Kenneth Graunke <[email protected]>
* mesa: Add MESA_SHADER_CAPTURE_PATH for writing .shader_test files.Kenneth Graunke2016-06-053-0/+77
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This writes linked shader programs to .shader_test files to $MESA_SHADER_CAPTURE_PATH in the format used by shader-db (http://cgit.freedesktop.org/mesa/shader-db). It supports both GLSL shaders and ARB programs. All stages that are linked together are written in a single .shader_test file. This eliminates the need for shader-db's split-to-files.py, as Mesa produces the desired format directly. It's much more reliable than parsing stdout/stderr, as those may contain extraneous messages, or simply be closed by the application and unavailable. We have many similar features already, but this is a bit different: - MESA_GLSL=dump writes to stdout, not files. - MESA_GLSL=log writes each stage to separate files (rather than all linked shaders in one file), at draw time (not link time), with uniform data and state flag info. - Tapani's shader replacement mechanism (MESA_SHADER_DUMP_PATH and MESA_SHADER_READ_PATH) also uses separate files per shader stage, but allows reading in files to replace an app's shader code. v2: Dump ARB programs too, not just GLSL. v3: Don't dump bogus 0.shader_test file. v4: Add "GL_ARB_separate_shader_objects" to the [require] block. v5: Print "GLSL 4.00" instead of "GLSL 4.0" in the [require] block. v6: Don't hardcode /tmp/mesa. v7: Fix memoization of getenv(). v8: Also print "SSO ENABLED" (suggested by Timothy). v9: Also handle ES shaders (suggested by Ilia). v10: Guard against MESA_SHADER_CAPTURE_PATH being too long; add _mesa_warning calls on error handling (suggested by Ben). v11: Fix crash when variable is unset introduced in v10. Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* nv50,nvc0: fix BGR10_A2UI vertex formatIlia Mirkin2016-06-051-1/+1
| | | | | | | | This is mostly academic as this is not reachable from GL, which only has the packed RGB10_A2UI vertex format. Signed-off-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]>
* nvc0: do not clear surfaces bins in the validate functionSamuel Pitoiset2016-06-052-5/+2
| | | | | | | | | We should not call nouveau_bufctx_reset() inside a validate function. This only affects Fermi where images are aliased between 3D and CP. Signed-off-by: Samuel Pitoiset <[email protected]> Reviewed-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]>
* nvc0: re-validate images after launching a grid on FermiSamuel Pitoiset2016-06-051-0/+3
| | | | | | | | | | | | | | | | | | | | | Images invalidation is a bit weird on Fermi and there is already a hack which forces invalidating all images when launching a computer shader to help in fixing 3D<->CP interaction. However, we need to re-validate images for compute because nvc0_compute_invalidate_surfaces() will destroy the previous binding. This is not really good for performance purposes but this might be improved later. This fixes the following piglits: - spec/arb_compute_shader/execution/basic-uniform-access - spec/arb_compute_shader/execution/mutiple-texture-reading - spec/arb_compute_shader/execution/multiple-workgroups - spec/glsl-4.30/execution/built-in-functions/cs-* (207 tests) Signed-off-by: Samuel Pitoiset <[email protected]> Reviewed-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]>
* radeonsi: fix images with level > 0Marek Olšák2016-06-051-1/+1
| | | | | | | | | | This should fix spec@arb_shader_image_load_store@level. Broken by: Commit: 95c5bbae66af3ca1f805d94f6fe8d8e4ba2c9c43 radeonsi: set some image descriptor fields at bind time Reviewed-by: Bas Nieuwenhuizen <[email protected]>
* nvc0: reduce overhead from always marking images dirtyIlia Mirkin2016-06-041-9/+36
| | | | | | | | | | We would revalidate images when anything was touched at all. Which is unfortunate, since the state tracker does not use CSO's to reduce the workload. So instead implement a protocol to ensure that something has changed before revalidating all the images. Signed-off-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]>
* nvc0: reduce overhead from always marking buffers dirtyIlia Mirkin2016-06-041-6/+20
| | | | | | | | | | We would revalidate buffers when anything was touched at all. Which is unfortunate, since the state tracker does not use CSO's to reduce the workload. So instead implement a protocol to ensure that something has changed before revalidating all the SSBOs. Signed-off-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]>
* nvc0: fix memory barrier flag handlingIlia Mirkin2016-06-041-9/+16
| | | | | Signed-off-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]>
* nvc0: mark bound buffer range validIlia Mirkin2016-06-043-0/+9
| | | | | Signed-off-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]>
* anv/entrypoints: don't go using wayland/xcb unless they are configuredDave Airlie2016-06-051-6/+9
| | | | | | | | | | | | | | The fix in: anv: let anv_entrypoints_gen.py generate proper Wayland/Xcb guards breaks things if wayland headers aren't installed. Separate things out properly to avoid that problem. [airlied: fixed up to put in pre-existing sections]. Reported-by: Arjan van de Ven Reviewed-by: Jason Ekstrand <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* gallium/radeon: don't use the DMA ring for pipelined buffer uploadsMarek Olšák2016-06-041-5/+4
| | | | | | | | | | | | | | | | | | | | Submitting a DMA IB flushes the GFX IB and all GPU caches. Vedran Miletić said: "On Tonga 380X, this improves The Talos Principle from 8.3 fps to 28.3 fps (all graphics settings Ultra, 4xAA, 1080p resolution with downsampling from 1200p)." Some anonymous dude said: R9 390 results: Tomb Raider (normal settings): 80 -> 88 FPS Talos Principle (custom settings): 23 -> 56 FPS Metro Last Light Redux (default benchmark settings): 39 -> 40 FPS Reviewed-by: Alex Deucher <[email protected]> Tested-by: Vedran Miletić <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* r600g: don't flush caches when binding shader resourcesMarek Olšák2016-06-044-31/+26
| | | | | | Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* r600g: only do necessary cache flushes in cp_dma_copy_bufferMarek Olšák2016-06-041-14/+1
| | | | | | | | | The main impact is that {upload, draw, upload, draw, ..} doesn't flush framebuffer caches before every upload. Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* r600g: only do necessary cache flushes in cp_dma_clear_bufferMarek Olšák2016-06-042-14/+18
| | | | | | | | The main impact is that fast color clear doesn't flush TC, CONST, DB. Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* r600g: remove a CP DMA workaround that's not needed anymoreMarek Olšák2016-06-041-6/+0
| | | | | | Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* r600g: fix CP DMA hazard with index buffer fetches (v3)Marek Olšák2016-06-047-7/+93
| | | | | | | | | v3: use PFP_SYNC_ME on EG-CM only when supported by the kernel, otherwise use MEM_WRITE + WAIT_REG_MEM to emulate that Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* r600g: properly sync CP with CP DMA on R6xxMarek Olšák2016-06-041-1/+8
| | | | | | | | This will allow removing useless cache & IB flushes. Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* r600g: write WAIT_UNTIL in the correct placeMarek Olšák2016-06-041-8/+11
| | | | | | | | | | This has been wrong all along. Fixing this will allow removing useless cache flushes. Cc: 11.1 11.2 12.0 <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* gallium/radeon: rename allocator_so_filled_size -> allocator_zeroed_memoryMarek Olšák2016-06-043-6/+6
| | | | | | Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
* gallium/u_suballoc: allow different alignment for each allocationMarek Olšák2016-06-048-21/+20
| | | | | | | | | Just move the alignment parameter from u_suballocator_create to u_suballocator_alloc. Reviewed-by: Alex Deucher <[email protected]> Tested-by: Grazvydas Ignotas <[email protected]> Tested-by: Dieter Nützel <[email protected]>
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-24 10:55:53 +0000