summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
* i965: Fix cross-primitive scratch corruption when changing the per-thread ↵Francisco Jerez2016-06-1317-18/+31
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | allocation. I haven't found any mention of this in the hardware docs, but experimentally what seems to be going on is that when the per-thread scratch slot size is changed between two pipelined draw calls, shader invocations using the old and new scratch size setting may end up being executed in parallel, causing their scratch offset calculations to be based in a different partitioning of the scratch space, which can cause their thread-local scratch space to overlap leading to cross-thread scratch corruption. I've been experimenting with alternative workarounds, like emitting a PIPE_CONTROL with DC flush and CS stall between draw (or dispatch compute) calls using different per-thread scratch allocation settings, or avoiding reuse of the scratch BO if the per-thread scratch allocation doesn't exactly match the original. Both seem to be as effective as this workaround, but they have potential performance implications, while this should be basically for free. Fixes over 40 failures in our CI system with spilling forced on (including CTS, dEQP and Piglit failures) on a number of different platforms from Gen4 to Gen9. The 'glsl-max-varyings' piglit test seems to be able to reproduce this bug consistently in the vertex shader on at least Gen4, Gen8 and Gen9 with spilling forced on. Cc: <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* i965: Keep track of the per-thread scratch allocation in brw_stage_state.Francisco Jerez2016-06-138-49/+70
| | | | | | | | | | | | | | | | This will be used to find out what per-thread slot size a previously allocated scratch BO was used with in order to fix a hardware race condition without introducing additional stalls or memory allocations. Instead of calling brw_get_scratch_bo() manually from the various codegen functions, call a new helper function that keeps track of the per-thread scratch size and conditionally allocates a larger scratch BO. v2: Handle BO allocation manually instead of relying on brw_get_scratch_bo (Ken). Cc: <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* i965: Fix scratch overallocation if the original slot size was already a ↵Francisco Jerez2016-06-131-1/+1
| | | | | | | | | | | | | | power of two. The bitwise arithmetic trick used in brw_get_scratch_size() to clamp the scratch allocation to 1KB has the unintended side effect that it will cause us to allocate 2x the required amount of scratch space if the original per-thread scratch size happened to be already a power of two. Instead use the obvious MAX2 idiom to clamp the scratch allocation to the expected range. Cc: <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* mesa: Make TexSubImage check negative dimensions sooner.Kenneth Graunke2016-06-131-19/+43
| | | | | | | | | | | | | | | | | | | Two dEQP tests expect INVALID_VALUE errors for negative width/height parameters, but get INVALID_OPERATION because they haven't actually created a destination image. This is arguably not a bug in Mesa, as there's no specified ordering of error conditions. However, it's also really easy to make the tests pass, and there's no real harm in doing these checks earlier. Fixes: dEQP-GLES3.functional.negative_api.texture.texsubimage3d_neg_width_height dEQP-GLES31.functional.debug.negative_coverage.get_error.texture.texsubimage3d_neg_width_height v2: Drop redundant check (caught by Anuj Phogat). Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Anuj Phogat <[email protected]>
* util: update some assertions in util_resource_copy_region()Brian Paul2016-06-131-4/+8
| | | | | | | | To cope with copies of compressed images which are not multiples of the block size. Suggested by Jose. Reviewed-by: Jose Fonseca <[email protected]> Reviewed-by: Roland Scheidegger <sroland@[email protected]>
* i965: Fix encode_slm_size() to take a generation, not a device info.Kenneth Graunke2016-06-132-3/+4
| | | | | | | | | | | | | | | In the Vulkan driver, we have the generation number (a compile time constant) but not necessarily the brw_device_info struct. I meant to rework the function to take a generation number instead of a brw_device_info pointer to accomodate this. But I forgot, and left it taking a brw_device_info pointer, while making Vulkan pass the generation number (8, 9, ...) directly. This led to crashes. Brown paper bag fix for commit 87d062a94080373995170f51063a9649. Cc: "12.0" <[email protected]> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96504 Signed-off-by: Kenneth Graunke <[email protected]>
* i965: Don't leak scratch BOs for TCS/TES.Kenneth Graunke2016-06-131-0/+4
| | | | | | | These need to be freed too. Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]>
* anv/pipeline: Don't dereference NULL dynamic state pointersNanley Chery2016-06-131-22/+48
| | | | | | | | | | | | | | | | Add guards to prevent dereferencing NULL dynamic pipeline state. Asserts of pCreateInfo members are moved to the earliest points at which they should not be NULL. This fixes a segfault seen in the McNopper demo, VKTS_Example09. v3 (Jason Ekstrand): - Fix disabled rasterization check - Revert opaque detection of color attachment usage Signed-off-by: Nanley Chery <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]> Cc: "12.0" <[email protected]>
* anv: Document and rename anv_pipeline_init_dynamic_state()Nanley Chery2016-06-131-3/+16
| | | | | | | | | | | | To reduce confusion, clarify that the state being copied is not dynamic. This agrees with the Vulkan spec's usage of the term. Various sections specify that the various pipeline state which have VkDynamicState enums (e.g. viewport, scissor, etc.) may or may not be dynamic. Signed-off-by: Nanley Chery <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]> Cc: "12.0" <[email protected]>
* nvc0/ir: clamp the UBO index for compute on KeplerSamuel Pitoiset2016-06-131-1/+9
| | | | | | | | | | We already check that the address is not "too far", but we should also clamp the UBO index in order to avoid looking at the wrong place in the driver cb. This is a pretty rare situation though. Signed-off-by: Samuel Pitoiset <[email protected]> Reviewed-by: Ilia Mirkin <[email protected]> Cc: "12.0" <[email protected]>
* radeonsi: enable scratch coalescingMarek Olšák2016-06-131-2/+10
| | | | | | | | This makes one particular compute shader 8x faster. Latest LLVM git is required. Reviewed-by: Nicolai Hähnle <[email protected]>
* st/va: hardlink driver instances to gallium_drv_video.soJimmy Berry2016-06-131-0/+14
| | | | | | | | | | | | Removes the need to set LIBVA_DRIVER_NAME=gallium for supported targets and is consistent with vdpau and general gallium drivers. Note: some versions of libva can detect the gallium name and use the backend. Although that behaviour seems inconsistent since it only works for some platforms/backends. Cc: "12.0" <[email protected]> Reviewed-by: Emil Velikov <[email protected]>
* vl: Fix trivial sign compare warningsJan Vesely2016-06-137-18/+15
| | | | | | | | | v2: add whitepace fixes Signed-off-by: Jan Vesely <[email protected]> Acked-by: Jose Fonseca <[email protected]> [Emil Velikov: squash a few more whitespace issues] Reviewed-by: Emil Velikov <[email protected]>
* Android: move libdrm settings to top-level Android.common.mkRob Herring2016-06-1325-39/+18
| | | | | | | | | | | | | | Fix warnings like these due to HAVE_LIBDRM being inconsistently defined: external/libdrm/include/drm/drm.h:839:30: warning: redefinition of typedef 'drm_clip_rect_t' is a C11 feature [-Wtypedef-redefinition] typedef struct drm_clip_rect drm_clip_rect_t; HAVE_LIBDRM needs to be set project wide to fix this. This change also harmlessly links libdrm with everything, but simplifies the makefiles a bit. Signed-off-by: Rob Herring <[email protected]> Acked-by: Emil Velikov <[email protected]>
* Android: disable some noisy warningsRob Herring2016-06-131-0/+4
| | | | | | | | | Turn off warnings for -Wpointer-arith, -Wno-missing-field-initializers, -Wno-initializer-overrides, and -Wno-mismatched-tags. These are all deemed pointless, on purpose or no plans to fix. Signed-off-by: Rob Herring <[email protected]> Acked-by: Emil Velikov <[email protected]>
* st/mesa: inline _mesa_create_context() into its only callerEmil Velikov2016-06-133-46/+6
| | | | | | | | Inline the function into it's only caller. This way it's more obvious how the classic and gallium drivers (st/mesa) use _mesa_initialize_context. Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* st/mesa: remove unneeded break from st_api_create_context()Emil Velikov2016-06-131-1/+0
| | | | | | | | We have return on the previous line, thus the break will never be reached. Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* st/mesa: use c99 initializer for st_gl_apiEmil Velikov2016-06-131-13/+13
| | | | | Signed-off-by: Emil Velikov <[email protected]> Acked-by: Ian Romanick <[email protected]>
* gallium: remove st_api::get_proc_address hookEmil Velikov2016-06-132-19/+0
| | | | | | | It has been unused for a long time, plus makes the gallium dri modules require an extra glapi symbol relative to their classic counterparts. Signed-off-by: Emil Velikov <[email protected]>
* mesa: remove _mesa_init_get_hash()Emil Velikov2016-06-133-70/+0
| | | | | | | | | | | | | | The actual code of the function print_table_stats() is guarded by a ifdef GET_DEBUG, which was not been defined in years. The last fix in 2013 (7db6b5aa91a) indicates that it's rarely used/tested. Since the issue has gone unnoticed for a whole year (broken with 2ad4a475474). Let's remove it for now. We can always revive it at a later stage. Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* mesa: kill off _mesa_do_init_remap_table()Emil Velikov2016-06-131-18/+7
| | | | | | | ... and inline its contents in _mesa_init_remap_table(). Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* mesa: use native types when possibleEmil Velikov2016-06-131-5/+6
| | | | | | | | All of the functions and related data is internal, so there's no point if using the GL types. Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* mesa: make _mesa_map_function_spec() staticEmil Velikov2016-06-132-6/+3
| | | | | | | Used only locally. Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* mesa: remove used _mesa_get_function_spec() and gl_function_remapEmil Velikov2016-06-132-27/+0
| | | | | | | Final user was killed with last commit. Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* mesa: remove unused _mesa_map_function_array()Emil Velikov2016-06-132-47/+0
| | | | | | | | Unused as of commit 5a175127f38 ("dri: Remove all extension enabling utility functions") and the patch before the previous patch. Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* glapi: remap_helper.py: remove MESA_alt_functionsEmil Velikov2016-06-131-42/+0
| | | | | | | The final user was nuked with last commit. Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* mesa: remove unused function _mesa_map_static_functions()Emil Velikov2016-06-132-23/+0
| | | | | | | | Unused as of commit 5a175127f38 ("dri: Remove all extension enabling utility functions") Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* dri/common: remove unused libdri_test_stubs.laEmil Velikov2016-06-133-104/+1
| | | | | | | | | | ... and associated file(s). No longer needed since commit 057259655e7 ("i965: Don't link libmesa or libdri_test_stubs into tests") Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* swr: automake: add missing -I flagEmil Velikov2016-06-131-0/+1
| | | | | | | | | | | | | When building from a release tarball (where the generated/built files are in srcdir) in an OOT fashion we need to have both builddir and srcdir in the includes list. Otherwise we'll error out, as the file (header gen_knobs.h in this case) won't be in the location where we are looking. Cc: "12.0" <[email protected]> Cc: Tim Rowley <[email protected]> Signed-off-by: Emil Velikov <[email protected]>
* automake: add SWR to `make distcheck' gallium driversEmil Velikov2016-06-131-1/+1
| | | | | | | | | Will allows us to catch missing files and build issues before getting the tarball out for general consumption. Cc: "12.0" <[email protected]> Cc: Tim Rowley <[email protected]> Signed-off-by: Emil Velikov <[email protected]>
* configure.ac: strip out the llvm-config -march/mtune flagsEmil Velikov2016-06-131-0/+3
| | | | | | | | | | | | | Otherwise drivers such as SWR that depend on providing their own values will fail to build. v2: Add -mcpu for good measure (Chuck) Cc: "11.2 12.0" <[email protected]> Cc: Tim Rowley <[email protected]> Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Chuck Atkins <[email protected]> Tested-by: Chuck Atkins <[email protected]>
* swr: Add missing headers for package inclusionChuck Atkins2016-06-131-1/+9
| | | | | CC: "12.0" <[email protected]> Reviewed-by: Emil Velikov <[email protected]>
* automake: get in-tree `make distclean' working again.Emil Velikov2016-06-131-1/+1
| | | | | | | | | | | | | | With earlier commit we've handled the `make distclean' out of tree build, yet we failed to attribute that for in-tree builds the test condition will return 1. Thus effectively the target will be considered as "failed". Fixes: b7f7ec78435 ("mesa: automake: distclean git_sha1.h when building OOT") Cc: <[email protected]> Tested-by: Andy Furniss <[email protected]> Reported-by: Andy Furniss <[email protected]> Signed-off-by: Emil Velikov <[email protected]>
* gallivm: Fix trivial sign warningsJan Vesely2016-06-138-21/+22
| | | | | | | v2: include whitespace fixes Signed-off-by: Jan Vesely <[email protected]> Reviewed-by: Jose Fonseca <[email protected]>
* st/va: use proper temp pipe_video_buffer templateJulien Isorce2016-06-131-4/+4
| | | | | | | | | | Instead of changing the format on the existing template which makes error handling not nice and confuses coverity. CoverityID: 1337953 Signed-off-by: Julien Isorce <[email protected]> Reviewed-by: Christian König <[email protected]>
* st/va: it is valid to release the VABuffer of an exported resourceJulien Isorce2016-06-131-7/+1
| | | | | | | | | | | | pipe_resource_reference(&res, NULL) will decrement reference counting, i.e. p_atomic_dec(res->count). But the va surface still has the initial reference since it has created the resource. So calling vaDestroyImage on a derived image calls VaDestroyBuffer but the decrementation won't reach 0. It is just wrong for vlVaDestroyBuffer to rely on the export_refcount flag. Finally the vaapi intel driver has the same logic. Signed-off-by: Julien Isorce <[email protected]> Reviewed-by: Christian König <[email protected]>
* glsl: fix component overlap validation for doublesTimothy Arceri2016-06-121-5/+5
| | | | | | | | | | | This change makes sure to remove arrays when checking if type is a double. The check for the end of the first slot of a multi-slot double is also fixed by bumping the check to 4 rather than 3. Previously we were we not reserving the last component. Reviewed-by: Kenneth Graunke <[email protected]>
* glsl: fix max varyings count for ARB_enhanced_layoutsTimothy Arceri2016-06-123-20/+43
| | | | | | | | | | | Since this extension allows more than one varying to share a single location we can't just count the number of slots a varying takes and add it to the total. Instead we now reuse the reserved varyings bitfield to determine how many slots are reserved for explicit locations instead. Reviewed-by: Kenneth Graunke <[email protected]>
* i965: Use the correct number of threads for compute shaders.Kenneth Graunke2016-06-121-1/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | We were programming the number of threads per subslice, when we should have been programming the total number of threads on the GPU as a whole. Thanks to Curro and Jordan for helping track this down! On Skylake GT3e: - Improves performance in Unreal's Elemental Demo by roughly 1.5-1.7x. - Improves performance in Synmark's Gl43CSDof by roughly 3.7x. - Improves performance in Synmark's Gl43GSCloth by roughly 1.18x. On Broadwell GT2: - Improves performance in Unreal's Elemental Demo by roughly 1.2-1.5x. - Improves performance in Synmark's Gl43CSDof by roughly 2.0x. - Improves performance in Synmark's Gl43GSCloth by 1.47035% +/- 0.255654% (n=25). On Haswell GT3e: - Improves performance in Unreal's Elemental Demo (in GL 4.3 mode) by roughly 1.10x. - Improves performance in Synmark's Gl43CSDof by roughly 1.18x. - Decreases performance in Synmark's Gl43CSCloth by -1.99484% +/- 0.432771% (n=64). On Ivybridge GT2: - Improves performance in Unreal's Elemental Demo (in GL 4.2 mode) by roughly 1.03x. - Improves performance in Synmark's G/43CSDof by roughly 1.25x. - No change in Synmark's Gl43CSCloth (n=28). Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* i965: Assert that the scratch spaces are in range.Kenneth Graunke2016-06-121-0/+14
| | | | | | | | | | | | | | | I don't know that anything actually guarantees this, but if we exceed the limits, we may end up overflowing and trashing random buffers that happen to be nearby in the VMA space, leading to rendering corruption, hangs, or worse. We should really fix this properly. However, the pitfall has existed for ages, so for now we should at least detect it. Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* i965: Fix CS scratch size calculations on Ivybridge and Baytrail.Kenneth Graunke2016-06-122-2/+10
| | | | | | | | | These are linear, not powers of two, and much more limited. Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* i965: Fix Haswell CS per-thread scratch space encoding.Kenneth Graunke2016-06-122-3/+25
| | | | | | | | | | | | | | | | | | | | | | Most scratch stages use power of two sizes, in kilobytes, where 0 means 1kB. But compute shaders on Haswell have a minimum of 2kB, and use a representation where 0 = 2kB. This meant that we were effectively telling the hardware to allocate each thread twice as much space as we meant to, while simultaneously not allocating that much space in the buffer, leading to overflows. Note that the existing code is completely wrong for Ivybridge, but that will take additional work to sort out, so I've left it as is for now. A subsequent commit will take care of that. Together with the previous patches, this fixes rendering corruption on Synmark's Gl43CSDof on Haswell. Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* i965: Account for poor address calculations in Haswell CS scratch size.Kenneth Graunke2016-06-121-1/+20
| | | | | | | | | | | | | | | | Curro figured this out by investigating the simulator. Apparently there's also a workaround in the Windows driver. I'm not sure it's actually documented anywhere. We were underallocating the scratch buffer by a factor of 128/70. v2: Rename threads_per_subslice to scratch_ids_per_subslice (suggested by Jordan Justen). Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* i965: Allocate scratch space for the maximum number of compute threads.Kenneth Graunke2016-06-121-1/+3
| | | | | | | | | | | | | | | | | | | | We were allocating enough space for the number of threads per subslice, when we should have been allocating space for the number of threads in the entire GPU. Even though we currently run with a reduced thread count (due to a bug), we might still overflow the scratch buffer because the address calculation is based on the FFTID, which can depend on exactly which threads, EUs, and threads are executing. We need to allocate enough for every possible thread that could run. Fixes rendering corruption in Synmark's Gl43CSDof on Gen8+. Earlier platforms need additional bug fixes. Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* i965: Set subslice_total on Gen7/7.5 platforms.Kenneth Graunke2016-06-121-1/+4
| | | | | | | | | | | | We'll use this for compute shader thread counts and scratch space calculations shortly. Note that subslices are referred to as "half slices" on Ivybridge. Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* i965: Fix shared local memory size for Gen9+.Kenneth Graunke2016-06-124-27/+37
| | | | | | | | | | | | | | | | | | | | | Skylake changes the representation of shared local memory size: Size | 0 kB | 1 kB | 2 kB | 4 kB | 8 kB | 16 kB | 32 kB | 64 kB | ------------------------------------------------------------------- Gen7-8 | 0 | none | none | 1 | 2 | 4 | 8 | 16 | ------------------------------------------------------------------- Gen9+ | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | The old formula would substantially underallocate the amount of space. This fixes GPU hangs on Skylake when running with full thread counts. v2: Fix the Vulkan driver too, use a helper function, and fix the table in the comments and commit message. Cc: "12.0" <[email protected]> Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* nv50: reinstate dedicated constbuf push pathIlia Mirkin2016-06-115-29/+50
| | | | | | | | | | | | | | | | | This was disabled due to occasionally incorrect behavior when trying to upload data. It later became apparent that nvc0 also had a similar but slightly different issue, which was resolved in commit e50c01d5. This takes the same logic as nvc0 and applies it to nv50 (which has somewhat different interfaces). Unfortunately I did not note down precisely what was broken with UBOs when removing the support from nv50, but I've tested a bunch of local traces, and none of them appear to regress. This should hopefully improve performance when UBOs are used, but this was not directly verified. Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Samuel Pitoiset <[email protected]>
* nv50: enable indirect addressing of fragment shader inputsIlia Mirkin2016-06-112-1/+2
| | | | Signed-off-by: Ilia Mirkin <[email protected]>
* mesa: add drawbuffer argument to ClearNamedFramebufferfiIlia Mirkin2016-06-103-3/+4
| | | | | | | | | | | This was fixed in revision 47 of the ARB_dsa spec in Oct 22, 2015. Since it's horrible to have differing APIs across library versions, we should attempt to minimize the impact by backporting it as far as possible and hope no one notices. Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Nicolai Hähnle <[email protected]> Cc: "11.2 12.0" <[email protected]>
* GL: update glcorearb.h to svn 32433Ilia Mirkin2016-06-101-38/+70
| | | | | | | | This brings in the fixed glClearNamedFramebufferfi definition, as well as a lot of GLsizei -> GLsizeiptr changes. Signed-off-by: Ilia Mirkin <[email protected]> Cc: "11.2 12.0" <[email protected]>
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-26 08:16:03 +0000