summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/vc4
Commit message (Collapse)AuthorAgeFilesLines
* gallium: set pipe_context uploaders in drivers (v3)Marek Olšák2017-02-141-5/+6
| | | | | | | | | | | | | | | Notes: - make sure the default size is large enough to handle all state trackers - pipe wrappers don't receive transfer calls from stream_uploader, because pipe_context::stream_uploader points directly to the underlying driver's stream_uploader (to keep it simple for now) v2: add error handling to nv50, nvc0, noop v3: set const_uploader Reviewed-by: Nicolai Hähnle <[email protected]> Tested-by: Edmondo Tommasina <[email protected]> (v1) Tested-by: Charmaine Lee <[email protected]>
* vc4: Enable glSampleMask() even when !rasterizer->multisample.Eric Anholt2017-02-101-2/+1
| | | | | | | | gallium's blitter expects that it can set the sample mask even when the rasterizer doesn't have the flag on. Between this and the previous test, 10 new ext_framebuffer_multisample tests start passing.
* vc4: Respect glSampleMask() even when we're not writing color.Eric Anholt2017-02-101-3/+13
| | | | | | gallium's quad-based blitter for copying MSAA depth textures expects to be able to do 4 passes updating a sample at a time using glSampleMask, and there's no color buffer bound when it's doing that.
* vc4: Use the nir_builder helper for loading sample mask.Eric Anholt2017-02-101-10/+1
|
* vc4: Use accurate 1/w in coordinate shader as well as vert shader.Eric Anholt2017-02-101-1/+1
| | | | | We probably shouldn't be emitting different scaled viewport coordinates between vertex and coord.
* vc4: Drop VS inputs to 8.Eric Anholt2017-02-101-4/+1
| | | | | | | In the hardware we only get to declare 8 vertex elements (GLES2's minimum), so we should be exposing that number here. Fixes an assertion failure in piglit texrect-many, at the expense of various GL 2.0-ish minmax tests now complaining that our count is too low.
* vc4: Avoid emitting small immediates for UBO indirect load address guards.Eric Anholt2017-02-105-4/+20
| | | | | | | | | | | | The kernel will reject our shader if we emit one here, and having 4, 8, or 12 as the top end of our UBO clamp rare is enough that it's not worth making the kernel let us. Fixes piglit fs-const-array-of-struct and fs-const-array-of-struct-of-array since recent GLSL linking changes made us get this as an indirect load of a uniform, instead of a tempoary. Cc: "13.0 17.0" <[email protected]>
* gallium: add separate PIPE_CAP_INT64_DIVMODIlia Mirkin2017-02-091-0/+1
| | | | | | | | | | | Nouveau does not currently have logic to implement this as a library function. Even though such a library could be written, there's no big advantage to do it that way for now given that int64 is a very uncommon use-case. Allow a driver to expose INT64 without supporting division and modulo operations. Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Nicolai Hähnle <[email protected]>
* gallium: turn PIPE_SHADER_CAP_DOUBLES into a screen capabilityNicolai Hähnle2017-02-021-1/+1
| | | | | | | | | | | | | | | | | | | Make the cap consistent with PIPE_CAP_INT64. Aside from the hypothetical case of using draw for vertex shaders (and actually caring about doubles...), every implementation supports doubles either nowhere or everywhere. Also, st/mesa didn't even check the cap correctly in all supported shader stages. While at it, add a missing LLVM version check for 64-bit integers in radeonsi. This is conservative: judging by the log, LLVM 3.8 might be sufficient, but there are probably bugs that have been fixed since then. v2: fix clover (Marek) Reviewed-by: Marek Olšák <[email protected]>
* vc4: Enable Neon on arm android buildsRob Herring2017-01-311-0/+2
| | | | | Signed-off-by: Rob Herring <[email protected]> Reviewed-by: Eric Anholt <[email protected]>
* vc4: fix arm64 build with NeonRob Herring2017-01-311-1/+1
| | | | | | | | The addition of Neon assembly breaks on arm64 builds because the assembly syntax is different. For now, restrict Neon to ARMv7 builds. Signed-off-by: Rob Herring <[email protected]> Reviewed-by: Eric Anholt <[email protected]>
* vc4: Make Neon inline assembly clang compatibleRob Herring2017-01-311-35/+35
| | | | | | | | | | | | clang throws an error on "%r2" and similar. I couldn't find any documentation on what "%r?" is supposed to mean and I've never seen any use like that as far as I remember. The parameter is supposed to be cpu_stride and just %2/%3 should be sufficient. There's no need for trailing ";" either, so remove those, too. Signed-off-by: Rob Herring <[email protected]> Reviewed-by: Eric Anholt <[email protected]>
* vc4: Coalesce into TLB writes as well as VPM/tex.Eric Anholt2017-01-281-1/+5
| | | | | | | | This generally cuts an instruction when blending is enabled and we thus have a single instruction generating the color value. total instructions in shared programs: 91759 -> 91634 (-0.14%) instructions in affected programs: 5338 -> 5213 (-2.34%)
* vc4: Avoid an extra temporary and mov in ffloor/ffract/fceil.Eric Anholt2017-01-281-13/+18
| | | | | | | | | | shader-db results: total instructions in shared programs: 92611 -> 91764 (-0.91%) instructions in affected programs: 27417 -> 26570 (-3.09%) The star is one shader in glmark2's terrain (drops 16% of its instructions), but there are also wins in mupen64plus and glb2.7.
* vc4: Flip the switch to run the GLSL compiler optimization loop once.Eric Anholt2017-01-281-1/+1
| | | | | | | | | | | | | This has almost no effect on shader-db: total instructions in shared programs: 92572 -> 92611 (0.04%) instructions in affected programs: 4486 -> 4525 (0.87%) Looking at 2 of the 7 different shaders that were hurt (all of which were in mupen64), they all appear to be just differences in order of instructions at the NIR level. The advantage is that this should significantly reduce time in the compiler.
* gallium: Add integer 64 capabilityDave Airlie2017-01-271-0/+1
| | | | | | | | | v1.1: move to using a normal CAP. (Marek) v2: fill in the cap everywhere Signed-off-by: Dave Airlie <[email protected]> Reviewed-by: Marek Olšák <[email protected]>
* vc4: Use NEON to speed up utile stores on Pi2+.cros-mesa-17.1.0-r2-vanillacros-mesa-17.1.0-r1-vanillachadv/cros-mesa-17.1.0-r2-vanillachadv/cros-mesa-17.1.0-r1-vanillaEric Anholt2017-01-261-5/+50
| | | | Improves 1024x1024 TexSubImage2D by 41.2371% +/- 3.52799% (n=10).
* vc4: Use NEON to speed up utile loads on Pi2.Eric Anholt2017-01-263-18/+115
| | | | | | | | | | | | | | | | | | | We had a lot of memcpy call overhead because gpu_stride wasn't being inlined. But if you split out the stride==8 and stride==16 cases like this code does while still using memcpy, you'd no longer have glibc's NEON memcpy applied at which point we'd be doing 16 uncached reads instead of 64/(NEON memcpy granularity), for about a 30% performance hit. By hand writing the assembly, we can get a whole cacheline loaded at a time. Unfortunately, NEON intrinsics turned out to be unusable -- they didn't have the vldm instruction available. Note that, for now, the NEON code is only enabled when building for ARMv7 (Pi 2+). We may want to do runtime detection for the Raspbian case, in the future. Improves 1024x1024 GetTexImage by 208.256% +/- 7.07029% (n=10).
* vc4: Move LT tiling code to a separate file.Eric Anholt2017-01-264-80/+122
| | | | This paves the way for building it twice, with NEON assembly or not.
* vc4: Use unreachable() in an unreachable codepath for tiling.Eric Anholt2017-01-261-4/+2
|
* gallium: add PIPE_CAP_TGSI_MUL_ZERO_WINSIlia Mirkin2017-01-231-0/+1
| | | | | | Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Nicolai Hähnle <[email protected]> Reviewed-by: Axel Davy <[email protected]>
* gallium: add PIPE_CAP_TGSI_FS_FBFETCHIlia Mirkin2017-01-161-0/+1
| | | | | Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Nicolai Hähnle <[email protected]>
* vc4: Rewrite T image handling based on calling the LT handler.Eric Anholt2017-01-051-34/+75
| | | | | | | | | | | | | The T images are composed of effectively swizzled-around blocks of LT (4x4 utile) images, so we can reduce the t_utile_address() calls by 16x by calling into the simpler LT loop. This also adds support for calling down with non-utile-aligned coordinates, which will be part of lifting the utile alignment requirement on our callers and avoiding the RMW on non-utile-aligned stores. Improves 1024x1024 TexSubImage by 2.55014% +/- 1.18584% (n=46) Improves 1024x1024 GetTexImage by 2.242% +/- 0.880954% (n=32)
* vc4: Move the utile_width/height functions to header inlines.Eric Anholt2017-01-052-37/+36
| | | | | | I want these inlined in the callers, particularly with the tiling changes coming up, but we're not building with lto so some caller would suffer.
* vc4: Make the load/store utile functions static.Eric Anholt2017-01-052-4/+2
| | | | | They don't have any other callers outside of this file, and I'm hoping they get inlined soon.
* vc4: Simplify the load/store utile functions.Eric Anholt2017-01-051-10/+22
| | | | | | | | | They now have less of a dependency on the cpp, and don't have to do a divide. Hacking up mesa-demos teximage to do only one subtest and not draw points, I saw 1024x1024 glTexSubImage2D() improve by 4.86939% +/- 1.40408% (n=30) and glGetTexImage() by 2.18978% +/- 0.140268% (n=5).
* vc4: Reuse a list function to simplify bufmgr code.Eric Anholt2017-01-051-11/+2
|
* vc4: Flush the job early if we're referencing too many BOs.Eric Anholt2017-01-053-0/+16
| | | | | | | | | | | If we get up toward 256MB (or whatever the CMA area size is), VC4_GEM_CREATE will start throwing errors. Even if we don't trigger that, when we flush the kernel's BO allocation for the CLs or bin memory may end up throwing an error, at which point our job won't get rendered at all. Just flush early (half of maximum CMA size) so that hopefully we never get to that point.
* gallium: add PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELYMarek Olšák2017-01-051-0/+1
| | | | | | Drivers with good compilers don't need aggressive optimizations before TGSI. Reviewed-by: Eric Anholt <[email protected]>
* nir: Rename convert_to_ssa lower_regs_to_ssaJason Ekstrand2016-12-291-1/+1
| | | | This matches the naming of nir_lower_vars_to_ssa, the other to-SSA pass.
* vc4: Rework scheduling of thread switch to cut one more NOP.Eric Anholt2016-12-291-46/+75
| | | | | | | | | | | | | | Jonas's patch got us most of the benefit of scheduling instructions into the delay slots of thread switch, but if there had been nothing to pair the thrsw with, it would move the thrsw up and leave a NOP where the thrsw was. Instead, don't pair anything with thrsw through the normal scheduling path, and have a separate helper function that inserts the thrsw earlier if possible and inserts any necessary NOPs. total instructions in shared programs: 93027 -> 92643 (-0.41%) instructions in affected programs: 14952 -> 14568 (-2.57%)
* vc4: Fill thread switching delay slotsJonas Pfeil2016-12-291-7/+38
| | | | | | | | | | | | | | | Scan for instructions without a signal set in front of the switching instruction and move the signal up there. shader-db results: total instructions in shared programs: 94494 -> 93027 (-1.55%) instructions in affected programs: 23545 -> 22078 (-6.23%) v2: Fix re-emitting of the instruction in the loop trying to emit NOPs, drop a scheduling change from branch delay slots. (by anholt) Signed-off-by: Jonas Pfeil <[email protected]>
* vc4: Enable NIR-based loop unrolling.Eric Anholt2016-12-291-0/+5
| | | | | This successfully unrolls a new shader in GLB2.7, which also gets that shader to successfully compile in multithreaded mode.
* treewide: s/comparitor/comparator/Ilia Mirkin2016-12-121-1/+1
| | | | | | | | | | git grep -l comparitor | xargs sed -i 's/comparitor/comparator/g' Just happened to notice this in a patch that was sent and included one of the tokens in question. Signed-off-by: Ilia Mirkin <[email protected]> Acked-by: Nicolai Hähnle <[email protected]>
* gallium: support for native fence fd'sRob Clark2016-12-011-0/+1
| | | | | | | This enables gallium support for EGL_ANDROID_native_fence_sync, for drivers which support PIPE_CAP_NATIVE_FENCE_FD. Signed-off-by: Rob Clark <[email protected]>
* vc4: Avoid false scheduling dependencies for LOAD_IMMs.Eric Anholt2016-11-302-0/+9
| | | | | | | | | | Noticed in shaders with branching, where we ended up scheduling delay slots near the start of a block for the uniforms reset setup. total instructions in shared programs: 93970 -> 93951 (-0.02%) instructions in affected programs: 3117 -> 3098 (-0.61%) 3DMMES performance +0.423087% +/- 0.133521% (n=9,10)
* vc4: Try to schedule QIR instructions between writing to and reading math.Eric Anholt2016-11-301-0/+22
| | | | | | | | | This helps us get the delay slots between SFU writes and reads filled. total instructions in shared programs: 94494 -> 93970 (-0.55%) instructions in affected programs: 59206 -> 58682 (-0.89%) 3DMMES performance +1.89967% +/- 0.157611% (n=10,9)
* vc4: Improve interleaving of texture coordinates vs results.Eric Anholt2016-11-301-3/+3
| | | | | | | | | | | | | | | | | | | | The latency_between was trying to handle the delay between the coordinate write ("before") and the corresponding sample read ("after"), but we were handing in the two instructions swapped. This meant that we tried to fit things between a tex_s and its *preceding* tex_result. This made us only interleave normal texture coordinates by accident, and pessimized UBO reads by pushing the tex_result collection earlier until there was nothing but it (and then its preceding coordinate setup) left. In addition to latency reduction, things end up packing better (probably due to reduced live ranges of the texture results): total instructions in shared programs: 98121 -> 94775 (-3.41%) instructions in affected programs: 91196 -> 87850 (-3.67%) 3DMMES performance +1.15569% +/- 0.124714% (n=8,10)
* vc4: Fix stray "." on no-op MUL packs.Eric Anholt2016-11-301-6/+6
| | | | | This happened when the PM bit was set for R4 unpacks, where the MUL pack was NOP.
* vc4: Allow merging instructions with SF set where the other writes NOP.Eric Anholt2016-11-301-0/+1
| | | | | | | | | | | | | I'm not sure how I managed to write the SF merge code (7d8b79f398f18ed7bb48a74b1b82950e2f08abad) without allowing merges with NOPs. *Everything* we try to merge with will have a NOP on one or the other side of the instruction, and that's why that commit showed no benefit. total instructions in shared programs: 99347 -> 95128 (-4.25%) instructions in affected programs: 91906 -> 87687 (-4.59%) 3DMMES performance +2.57105% +/- 0.135276% (n=6,8)
* vc4: In a loop break/continue, jump if everyone has taken the path.Eric Anholt2016-11-301-10/+17
| | | | | | | | | | | | | | | | | This should be a win for most loops, which tend to have uniform control flow. More importantly, it exposes important information to live variables: that the break/continue here means that our jump target may have access to values that were live on our input. Previously, we were just setting the exec mask and letting control flow fall through, so an intervening def between the break and the end of the loop would appear to live variables as if it screened off the variable, when it didn't actually. Fixes a regression in glsl-vs-loop-redundant-condition.shader_test when a perturbing of register allocation caused a live variable to get stomped. Cc: 13.0 <[email protected]>
* gallium: add PIPE_CAP_TGSI_CAN_READ_OUTPUTSNicolai Hähnle2016-11-301-0/+1
| | | | | | | | | | | Drivers that support this benefit by saving one lowering pass in the GLSL-to-TGSI conversion. radeonsi already supports this because all outputs are stored in temporary variables before the export (except for TCS outputs, which have always been readable in TGSI anyway due to their special semantics). Reviewed-by: Marek Olšák <[email protected]>
* vc4: Add a note for the future about texture latency calculation.Eric Anholt2016-11-291-0/+20
| | | | | | | Debugging a shader-db reported cycle count regression from the tex coalescing, I eventually figured out that the texture latencies were totally bogus. Really fixing it will probably involve mirroring vc4_qir_schedule.c's texture fifo management here.
* vc4: Add support for coalescing ALU ops into tex_[srtb] MOVs.Eric Anholt2016-11-294-29/+37
| | | | | | | | | | | This isn't as complete as I would like (can't merge interpolation because of the implicit r5 dependency, doesn't work with control flow), but this was cheap and easy. Improves 3DMMES Taiji performance by 1.15353% +/- 0.299896% (n=29, 16) total instructions in shared programs: 99810 -> 99059 (-0.75%) instructions in affected programs: 10705 -> 9954 (-7.02%)
* vc4: Restructure VPM write optimization into two passes.Eric Anholt2016-11-291-18/+10
| | | | | For texturing, there won't be a fixed limit on how many writes there are, so we need to compute uses up front.
* vc4: Make qir_for_each_inst_inorder() safe against removal.Eric Anholt2016-11-291-1/+1
| | | | | The dead code elimination wants it to be safe, and I actually got segfaults due to it being unsafe with the new coalescing pass.
* vc4: Split optimizing VPM writes from VPM reads.Eric Anholt2016-11-295-51/+110
| | | | | | The VPM write logic will be basically the same as the texture coordinate write logic we need, and it's not really related to the VPM read logic other than the reuse of the use_count array.
* vc4: Restructure texture insts as ALU ops with tex_[strb] as the dst.Eric Anholt2016-11-299-89/+194
| | | | | For now we're still just generating MOVs, but this will let us fold into other ops in the future. No difference on shader-db.
* vc4: Refactor qir_get_op_nsrc(enum qop) to qir_get_nsrc(struct qinst *).Eric Anholt2016-11-2917-36/+34
| | | | | | Every caller was dereffing the qinst, and this will let us make the number of sources vary depending on the destination of the qinst so that we can have general ALU ops that store to tex_[strb] and get an implicit uniform.
* vc4: Replace the qinst src[] with a fixed-size array.Eric Anholt2016-11-293-4/+2
| | | | | | This may have made a tiny bit of sense when we had one 4-arg inst per shader, but if we only ever put 2 things in, having a pointer to 2 things almost every instruction is pointless indirection.
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-28 05:19:19 +0000