summaryrefslogtreecommitdiffstats
path: root/src/intel/compiler
Commit message (Collapse)AuthorAgeFilesLines
* intel/compiler: Explicitly cast register type in switchIan Romanick2018-04-061-1/+1
| | | | | | | | | | | | | | | | | | brw_reg::type is "enum brw_reg_type type:4". For whatever reason, GCC is treating this as an int instead of an enum. As a result, it doesn't detect missing switch cases and it doesn't detect that flow can get out of the switch. This silences the warning: src/intel/compiler/brw_reg.h: In function ‘bool brw_regs_negative_equal(const brw_reg*, const brw_reg*)’: src/intel/compiler/brw_reg.h:305:1: warning: control reaches end of non-void function [-Wreturn-type] } ^ Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Matt Turner <[email protected]> Reviewed-by: Brian Paul <[email protected]>
* intel: compiler: silence compiler warningLionel Landwerlin2018-04-041-0/+1
| | | | | | | | | | ../src/intel/compiler/brw_reg.h: In function ‘bool brw_regs_negative_equal(const brw_reg*, const brw_reg*)’: ../src/intel/compiler/brw_reg.h:305:1: warning: control reaches end of non-void function [-Wreturn-type] Introduced by 8f83eea71e233 ("i965: Add negative_equals methods"). Signed-off-by: Lionel Landwerlin <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]>
* nir+drivers: add helpers to get # of src/dest componentsRob Clark2018-04-031-6/+5
| | | | | | | | | Add helpers to get the number of src/dest components for an intrinsic, and update spots that were open-coding this logic to use the helpers instead. Signed-off-by: Rob Clark <[email protected]> Reviewed-by: Eric Anholt <[email protected]>
* intel/vec4: Set channel_sizes for MOV_INDIRECT sourcesJason Ekstrand2018-03-301-1/+4
| | | | | | | | | | | Otherwise, any indirect push constant access results in an assertion failure when we start digging through the channel_sizes array. This fixes dEQP-VK.pipeline.push_constant.graphics_pipeline.dynamic_index_vert on Haswell. It should be a harmless no-op for GL since indirect push constants aren't used there. Reviewed-by: Kenneth Graunke <[email protected]> Fixes: e69e5c7006d "i965/vec4: load dvec3/4 uniforms first in the..."
* util: Add and use util_is_power_of_two_nonzeroIan Romanick2018-03-291-2/+2
| | | | | Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Eduardo Lima Mitev <[email protected]>
* util: Move util_is_power_of_two to bitscan.h and rename to ↵Ian Romanick2018-03-291-2/+2
| | | | | | | | | | | util_is_power_of_two_or_zero The new name make the zero-input behavior more obvious. The next patch adds a new function with different zero-input behavior. Signed-off-by: Ian Romanick <[email protected]> Suggested-by: Matt Turner <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]>
* intel/fs: Don't emit a des copy for image ops with has_dest == falseJason Ekstrand2018-03-271-3/+6
| | | | | | | | | | This was causing us to walk dest_components times over a thing with no destination. This happened to work because all of the image intrinsics without a destination also happened to have dest_components == 0. We shouldn't be reading dest_components if has_dest == false. Reviewed-by: Matt Turner <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* i965/vec4: Fix null destination register in 3-source instructionsIan Romanick2018-03-262-0/+27
| | | | | | | | | | | | | | | | | | | | | | | A recent commit (see below) triggered some cases where conditional modifier propagation and dead code elimination would cause a MAD instruction like the following to be generated: mad.l.f0 null, ... Matt pointed out that fs_visitor::fixup_3src_null_dest() fixes cases like this in the scalar backend. This commit basically ports that code to the vec4 backend. NOTE: I have sent a couple tests to the piglit list that reproduce this bug *without* the commit mentioned below. This commit fixes those tests. Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Matt Turner <[email protected]> Tested-by: Tapani Pälli <[email protected]> Cc: [email protected] Fixes: ee63933a7 ("nir: Distribute binary operations with constants into bcsel") Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105704
* i965/vec4: Propagate conditional modifiers from compares to addsIan Romanick2018-03-261-5/+65
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | No changes on Broadwell or later as those platforms do not use the vec4 backend. Ivy Bridge and Haswell had similar results. (Ivy Bridge shown) total instructions in shared programs: 11682119 -> 11681056 (<.01%) instructions in affected programs: 150403 -> 149340 (-0.71%) helped: 950 HURT: 0 helped stats (abs) min: 1 max: 16 x̄: 1.12 x̃: 1 helped stats (rel) min: 0.23% max: 2.78% x̄: 0.82% x̃: 0.71% 95% mean confidence interval for instructions value: -1.19 -1.04 95% mean confidence interval for instructions %-change: -0.84% -0.79% Instructions are helped. total cycles in shared programs: 257495842 -> 257495238 (<.01%) cycles in affected programs: 270302 -> 269698 (-0.22%) helped: 271 HURT: 13 helped stats (abs) min: 2 max: 14 x̄: 2.42 x̃: 2 helped stats (rel) min: 0.06% max: 1.13% x̄: 0.32% x̃: 0.28% HURT stats (abs) min: 2 max: 12 x̄: 4.00 x̃: 4 HURT stats (rel) min: 0.15% max: 1.18% x̄: 0.30% x̃: 0.26% 95% mean confidence interval for cycles value: -2.41 -1.84 95% mean confidence interval for cycles %-change: -0.31% -0.26% Cycles are helped. Sandy Bridge total instructions in shared programs: 10430493 -> 10429727 (<.01%) instructions in affected programs: 120860 -> 120094 (-0.63%) helped: 766 HURT: 0 helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1 helped stats (rel) min: 0.30% max: 2.70% x̄: 0.78% x̃: 0.73% 95% mean confidence interval for instructions value: -1.00 -1.00 95% mean confidence interval for instructions %-change: -0.80% -0.75% Instructions are helped. total cycles in shared programs: 146138718 -> 146138446 (<.01%) cycles in affected programs: 244114 -> 243842 (-0.11%) helped: 132 HURT: 0 helped stats (abs) min: 2 max: 4 x̄: 2.06 x̃: 2 helped stats (rel) min: 0.03% max: 0.43% x̄: 0.16% x̃: 0.19% 95% mean confidence interval for cycles value: -2.12 -2.00 95% mean confidence interval for cycles %-change: -0.18% -0.15% Cycles are helped. GM45 and Iron Lake had identical results. (Iron Lake shown) total instructions in shared programs: 7780251 -> 7780248 (<.01%) instructions in affected programs: 175 -> 172 (-1.71%) helped: 3 HURT: 0 helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1 helped stats (rel) min: 1.49% max: 2.44% x̄: 1.81% x̃: 1.49% total cycles in shared programs: 177851584 -> 177851578 (<.01%) cycles in affected programs: 9796 -> 9790 (-0.06%) helped: 3 HURT: 0 helped stats (abs) min: 2 max: 2 x̄: 2.00 x̃: 2 helped stats (rel) min: 0.05% max: 0.08% x̄: 0.06% x̃: 0.05% Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* i965/vec4: Allow cmod propagation when src0 is a uniform or shader inputIan Romanick2018-03-261-1/+2
| | | | | | | | | | No shader-db changes. This source must have been written by a previous instruction, so it cannot be a uniform or a shader input. However, this change allows the next commit to help more shaders. Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* i965/fs: Propagate conditional modifiers from compares to addsIan Romanick2018-03-262-5/+400
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The math inside the add and the cmp in this instruction sequence is the same. We can utilize this to eliminate the compare. add(8) g5<1>F g2<8,8,1>F g64.5<0,1,0>F { align1 1Q compacted }; cmp.z.f0(8) null<1>F g2<8,8,1>F -g64.5<0,1,0>F { align1 1Q switch }; (-f0) sel(8) g8<1>F (abs)g5<8,8,1>F 3e-37F { align1 1Q }; This is reduced to: add.z.f0(8) g5<1>F g2<8,8,1>F g64.5<0,1,0>F { align1 1Q compacted }; (-f0) sel(8) g8<1>F (abs)g5<8,8,1>F 3e-37F { align1 1Q }; This optimization pass could do even better. The nature of converting vectorized code from the GLSL front end to scalar code in NIR results in sequences like: add(8) g7<1>F g4<8,8,1>F g64.5<0,1,0>F { align1 1Q compacted }; add(8) g6<1>F g3<8,8,1>F g64.5<0,1,0>F { align1 1Q compacted }; add(8) g5<1>F g2<8,8,1>F g64.5<0,1,0>F { align1 1Q compacted }; cmp.z.f0(8) null<1>F g2<8,8,1>F -g64.5<0,1,0>F { align1 1Q switch }; (-f0) sel(8) g8<1>F (abs)g5<8,8,1>F 3e-37F { align1 1Q }; cmp.z.f0(8) null<1>F g3<8,8,1>F -g64.5<0,1,0>F { align1 1Q switch }; (-f0) sel(8) g10<1>F (abs)g6<8,8,1>F 3e-37F { align1 1Q }; cmp.z.f0(8) null<1>F g4<8,8,1>F -g64.5<0,1,0>F { align1 1Q switch }; (-f0) sel(8) g12<1>F (abs)g7<8,8,1>F 3e-37F { align1 1Q }; In this sequence, only the first cmp.z is removed. With different scheduling, all 3 could get removed. Skylake total instructions in shared programs: 14407009 -> 14400173 (-0.05%) instructions in affected programs: 1307274 -> 1300438 (-0.52%) helped: 4880 HURT: 0 helped stats (abs) min: 1 max: 33 x̄: 1.40 x̃: 1 helped stats (rel) min: 0.03% max: 8.70% x̄: 0.70% x̃: 0.52% 95% mean confidence interval for instructions value: -1.45 -1.35 95% mean confidence interval for instructions %-change: -0.72% -0.69% Instructions are helped. total cycles in shared programs: 532943169 -> 532923528 (<.01%) cycles in affected programs: 14065798 -> 14046157 (-0.14%) helped: 2703 HURT: 339 helped stats (abs) min: 1 max: 1062 x̄: 12.27 x̃: 2 helped stats (rel) min: <.01% max: 28.72% x̄: 0.38% x̃: 0.21% HURT stats (abs) min: 1 max: 739 x̄: 39.86 x̃: 12 HURT stats (rel) min: 0.02% max: 27.69% x̄: 1.38% x̃: 0.41% 95% mean confidence interval for cycles value: -8.66 -4.26 95% mean confidence interval for cycles %-change: -0.24% -0.14% Cycles are helped. LOST: 0 GAINED: 1 Broadwell total instructions in shared programs: 14719636 -> 14712949 (-0.05%) instructions in affected programs: 1288188 -> 1281501 (-0.52%) helped: 4845 HURT: 0 helped stats (abs) min: 1 max: 33 x̄: 1.38 x̃: 1 helped stats (rel) min: 0.03% max: 8.00% x̄: 0.70% x̃: 0.52% 95% mean confidence interval for instructions value: -1.43 -1.33 95% mean confidence interval for instructions %-change: -0.72% -0.68% Instructions are helped. total cycles in shared programs: 559599253 -> 559581699 (<.01%) cycles in affected programs: 13315565 -> 13298011 (-0.13%) helped: 2600 HURT: 269 helped stats (abs) min: 1 max: 2128 x̄: 12.24 x̃: 2 helped stats (rel) min: <.01% max: 23.95% x̄: 0.41% x̃: 0.20% HURT stats (abs) min: 1 max: 790 x̄: 53.07 x̃: 20 HURT stats (rel) min: 0.02% max: 15.96% x̄: 1.55% x̃: 0.75% 95% mean confidence interval for cycles value: -8.47 -3.77 95% mean confidence interval for cycles %-change: -0.27% -0.18% Cycles are helped. LOST: 0 GAINED: 8 Haswell total instructions in shared programs: 12978609 -> 12973483 (-0.04%) instructions in affected programs: 932921 -> 927795 (-0.55%) helped: 3480 HURT: 0 helped stats (abs) min: 1 max: 33 x̄: 1.47 x̃: 1 helped stats (rel) min: 0.03% max: 7.84% x̄: 0.78% x̃: 0.58% 95% mean confidence interval for instructions value: -1.53 -1.42 95% mean confidence interval for instructions %-change: -0.80% -0.75% Instructions are helped. total cycles in shared programs: 410270788 -> 410250531 (<.01%) cycles in affected programs: 10986161 -> 10965904 (-0.18%) helped: 2087 HURT: 254 helped stats (abs) min: 1 max: 2672 x̄: 14.63 x̃: 4 helped stats (rel) min: <.01% max: 39.61% x̄: 0.42% x̃: 0.21% HURT stats (abs) min: 1 max: 519 x̄: 40.49 x̃: 16 HURT stats (rel) min: 0.01% max: 12.83% x̄: 1.20% x̃: 0.47% 95% mean confidence interval for cycles value: -12.82 -4.49 95% mean confidence interval for cycles %-change: -0.31% -0.18% Cycles are helped. LOST: 0 GAINED: 5 Ivy Bridge total instructions in shared programs: 11686082 -> 11681548 (-0.04%) instructions in affected programs: 937696 -> 933162 (-0.48%) helped: 3150 HURT: 0 helped stats (abs) min: 1 max: 33 x̄: 1.44 x̃: 1 helped stats (rel) min: 0.03% max: 7.84% x̄: 0.69% x̃: 0.49% 95% mean confidence interval for instructions value: -1.49 -1.38 95% mean confidence interval for instructions %-change: -0.71% -0.67% Instructions are helped. total cycles in shared programs: 257514962 -> 257492471 (<.01%) cycles in affected programs: 11524149 -> 11501658 (-0.20%) helped: 1970 HURT: 239 helped stats (abs) min: 1 max: 3525 x̄: 17.48 x̃: 3 helped stats (rel) min: <.01% max: 49.60% x̄: 0.46% x̃: 0.17% HURT stats (abs) min: 1 max: 1358 x̄: 50.00 x̃: 15 HURT stats (rel) min: 0.02% max: 59.88% x̄: 1.84% x̃: 0.65% 95% mean confidence interval for cycles value: -17.01 -3.35 95% mean confidence interval for cycles %-change: -0.33% -0.08% Cycles are helped. LOST: 9 GAINED: 1 Sandy Bridge total instructions in shared programs: 10432841 -> 10429893 (-0.03%) instructions in affected programs: 685071 -> 682123 (-0.43%) helped: 2453 HURT: 0 helped stats (abs) min: 1 max: 9 x̄: 1.20 x̃: 1 helped stats (rel) min: 0.02% max: 7.55% x̄: 0.64% x̃: 0.46% 95% mean confidence interval for instructions value: -1.23 -1.17 95% mean confidence interval for instructions %-change: -0.67% -0.62% Instructions are helped. total cycles in shared programs: 146133660 -> 146134195 (<.01%) cycles in affected programs: 3991634 -> 3992169 (0.01%) helped: 1237 HURT: 153 helped stats (abs) min: 1 max: 2853 x̄: 6.93 x̃: 2 helped stats (rel) min: <.01% max: 29.00% x̄: 0.24% x̃: 0.14% HURT stats (abs) min: 1 max: 1740 x̄: 59.56 x̃: 12 HURT stats (rel) min: 0.03% max: 78.98% x̄: 1.96% x̃: 0.42% 95% mean confidence interval for cycles value: -5.13 5.90 95% mean confidence interval for cycles %-change: -0.17% 0.16% Inconclusive result (value mean confidence interval includes 0). LOST: 0 GAINED: 1 GM45 and Iron Lake had similar results (GM45 shown): total instructions in shared programs: 4800332 -> 4798380 (-0.04%) instructions in affected programs: 565995 -> 564043 (-0.34%) helped: 1451 HURT: 0 helped stats (abs) min: 1 max: 20 x̄: 1.35 x̃: 1 helped stats (rel) min: 0.05% max: 5.26% x̄: 0.47% x̃: 0.31% 95% mean confidence interval for instructions value: -1.40 -1.29 95% mean confidence interval for instructions %-change: -0.50% -0.45% Instructions are helped. total cycles in shared programs: 122032318 -> 122027798 (<.01%) cycles in affected programs: 8334868 -> 8330348 (-0.05%) helped: 1029 HURT: 1 helped stats (abs) min: 2 max: 40 x̄: 4.43 x̃: 2 helped stats (rel) min: <.01% max: 1.83% x̄: 0.09% x̃: 0.04% HURT stats (abs) min: 38 max: 38 x̄: 38.00 x̃: 38 HURT stats (rel) min: 0.25% max: 0.25% x̄: 0.25% x̃: 0.25% 95% mean confidence interval for cycles value: -4.70 -4.08 95% mean confidence interval for cycles %-change: -0.09% -0.08% Cycles are helped. Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* i965/fs: Allow cmod propagation when src0 is a uniform or shader inputIan Romanick2018-03-261-1/+2
| | | | | | | | | | No shader-db changes. This source must have been written by a previous instruction, so it cannot be a uniform or a shader input. However, this change allows the next commit to help about 900 more shaders. Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* i965: Add negative_equals methodsIan Romanick2018-03-267-0/+72
| | | | | | | | | | | | | | | | | | | | | | This method is similar to the existing ::equals methods. Instead of testing that two src_regs are equal to each other, it tests that one is the negation of the other. v2: Simplify various checks based on suggestions from Matt. Use src_reg::type instead of fixed_hw_reg.type in a check. Also suggested by Matt. v3: Rebase on 3 years. Fix some problems with negative_equals with VF constants. Add fs_reg::negative_equals. v4: Replace the existing default case with BRW_REGISTER_TYPE_UB, BRW_REGISTER_TYPE_B, and BRW_REGISTER_TYPE_NF. Suggested by Matt. Expand the FINISHME comment to better explain why it isn't already finished. Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]> [v3] Reviewed-by: Matt Turner <[email protected]>
* nir: Rename image intrinsics to image_varJason Ekstrand2018-03-231-23/+23
| | | | | | | | | | | Generated with git grep -l nir_intrinsic_image | xargs \ sed -i 's/nir_intrinsic_image/nir_intrinsic_image_var/g' and some manual fixing in nir_intrinsics.h Reviewed-by: Timothy Arceri <[email protected]>
* intel/compiler: Readd ICL to test_eu_validate.cppMatt Turner2018-03-221-0/+1
| | | | Now that the PCI IDs are upstream, this can be readded.
* intel/compiler: Skip 64-bit type tests when types not availableMatt Turner2018-03-221-0/+19
| | | | Reviewed-by: Kenneth Graunke <[email protected]>
* intel/compiler/icl: Clear "null render target" bit in extended message ↵Jason Ekstrand2018-03-222-0/+6
| | | | | | | | | descriptor Otherwise all our render target writes go no where. Reviewed-by: Matt Turner <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* intel/compiler/icl: Update the assert in brw_stage_has_packed_dispatch()Anuj Phogat2018-03-221-1/+1
| | | | | | | | Rafael ran piglit with the test code enabled and saw no additional GPU hangs. Reviewed-by: Matt Turner <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* i965: Silence compiler warning about promoted_constants.Eric Anholt2018-03-161-1/+1
| | | | | | | | We only have a cfg != NULL if we went through one of the paths that set it, but my compiler doesn't figure that out. Reviewed-by: Lionel Landwerlin <[email protected]> Fixes: 6411defdcd6f ("intel/cs: Re-run final NIR optimizations for each SIMD size")
* intel/compiler: Use gen_get_device_info() in test_eu_validateMatt Turner2018-03-162-39/+18
| | | | | | | | | | | | | Previously the unit test filled out a minimal devinfo struct. A previous patch caused the test to begin assert failing because the devinfo was not complete. Avoid this by using the real mechanism to create devinfo. Note that we have to drop icl from the table, since we now rely on the name -> PCI ID translation done by gen_device_name_to_pci_device_id(), and ICL's PCI IDs are not upstream yet. Fixes: f89e735719a6 ("intel/compiler: Check for unsupported register sizes.") Reviewed-by: Rafael Antognolli <[email protected]>
* intel/compiler: Check for unsupported register sizes.Rafael Antognolli2018-03-161-0/+3
| | | | | | | | | Make sure we don't emit 64 bit types if the hardware doesn't support them. Signed-off-by: Rafael Antognolli <[email protected]> Suggested-by: Kenneth Graunke <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* compiler: int8/uint8 supportKarol Herbst2018-03-143-0/+9
| | | | | | | | | | OpenCL kernels also have int8/uint8. v2: remove changes in nir_search as Jason posted a patch for that Reviewed-by: Jason Ekstrand <[email protected]> Signed-off-by: Rob Clark <[email protected]> Signed-off-by: Karol Herbst <[email protected]>
* i965/vec4: Allow CSE on subset VF constant loadsIan Romanick2018-03-081-0/+19
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | v2: Rewrite the code that generates the VF mask. Suggested by Ken. No changes on other platforms. Haswell, Ivy Bridge, and Sandy Bridge had similar results. (Haswell shown) total instructions in shared programs: 13059891 -> 13059884 (<.01%) instructions in affected programs: 431 -> 424 (-1.62%) helped: 7 HURT: 0 helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1 helped stats (rel) min: 1.19% max: 5.26% x̄: 2.05% x̃: 1.49% 95% mean confidence interval for instructions value: -1.00 -1.00 95% mean confidence interval for instructions %-change: -3.39% -0.71% Instructions are helped. total cycles in shared programs: 409260032 -> 409260018 (<.01%) cycles in affected programs: 4228 -> 4214 (-0.33%) helped: 7 HURT: 0 helped stats (abs) min: 2 max: 2 x̄: 2.00 x̃: 2 helped stats (rel) min: 0.28% max: 2.04% x̄: 0.54% x̃: 0.28% 95% mean confidence interval for cycles value: -2.00 -2.00 95% mean confidence interval for cycles %-change: -1.15% 0.07% Inconclusive result (%-change mean confidence interval includes 0). Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* i965/vec4: Relax writemask condition in CSEIan Romanick2018-03-081-1/+9
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | If the previously seen instruction generates more fields than the new instruction, still allow CSE to happen. This doesn't do much, but it also enables a couple more shaders in the next patch. It helped quite a bit in another change series that I have (at least for now) abandoned. v2: Add some extra comentary about the parameters to instructions_match. Suggested by Ken. No changes on Skylake, Broadwell, Iron Lake or GM45. Ivy Bridge and Haswell had similar results. (Ivy Bridge shown) total instructions in shared programs: 11780295 -> 11780294 (<.01%) instructions in affected programs: 302 -> 301 (-0.33%) helped: 1 HURT: 0 total cycles in shared programs: 257308315 -> 257308313 (<.01%) cycles in affected programs: 2074 -> 2072 (-0.10%) helped: 1 HURT: 0 Sandy Bridge total instructions in shared programs: 10506687 -> 10506686 (<.01%) instructions in affected programs: 335 -> 334 (-0.30%) helped: 1 HURT: 0 Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* i965/fs: Merge CMP and SEL into CSEL on Gen8+Ian Romanick2018-03-082-0/+107
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | v2: Fix several problems handling inverted predicates. Add a much bigger comment around the BRW_CONDITIONAL_NZ case. v3: Allow uniforms and shader inputs as sources for the original SEL and CMP instructions. This enables a LOT more shaders to receive CSEL merging (5816 vs 8564 on SKL). v4: Report progress. Broadwell and Skylake had similar results. (Broadwell shown) helped: 8527 HURT: 0 helped stats (abs) min: 1 max: 27 x̄: 2.44 x̃: 1 helped stats (rel) min: 0.03% max: 17.80% x̄: 1.12% x̃: 0.70% 95% mean confidence interval for instructions value: -2.51 -2.36 95% mean confidence interval for instructions %-change: -1.15% -1.10% Instructions are helped. total cycles in shared programs: 559442317 -> 558288357 (-0.21%) cycles in affected programs: 372699860 -> 371545900 (-0.31%) helped: 6748 HURT: 1450 helped stats (abs) min: 1 max: 32000 x̄: 182.41 x̃: 12 helped stats (rel) min: <.01% max: 66.08% x̄: 3.42% x̃: 0.70% HURT stats (abs) min: 1 max: 2538 x̄: 53.08 x̃: 14 HURT stats (rel) min: <.01% max: 96.72% x̄: 3.32% x̃: 0.90% 95% mean confidence interval for cycles value: -179.01 -102.51 95% mean confidence interval for cycles %-change: -2.37% -2.08% Cycles are helped. LOST: 0 GAINED: 6 No changes on earlier platforms. Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Samuel Iglesias Gonsálvez <[email protected]> [v1] Reviewed-by: Kenneth Graunke <[email protected]> [v3] Reviewed-by: Matt Turner <[email protected]>
* i965/fs: Add infrastructure for generating CSEL instructions.Kenneth Graunke2018-03-088-1/+34
| | | | | | | | | | | | | | | | v2 (idr): Don't allow CSEL with a non-float src2. v3 (idr): Add CSEL to fs_inst::flags_written. Suggested by Matt. v4 (idr): Only set BRW_ALIGN_16 on Gen < 10 (suggested by Matt). Don't reset the access mode afterwards (suggested by Samuel and Matt). Add support for CSEL not modifying the flags to more places (requested by Matt). Signed-off-by: Kenneth Graunke <[email protected]> Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Samuel Iglesias Gonsálvez <[email protected]> [v3] Reviewed-by: Matt Turner <[email protected]>
* anv: Add support for SPIR-V 1.3 subgroup operationsJason Ekstrand2018-03-072-2/+9
| | | | | | | This requires us to bump the subgroup size to 32 for all shader stages because Vulkan requires that to be a physical device query. Reviewed-by: Iago Toral Quiroga <[email protected]>
* intel/fs: Add support for subgroup quad operationsJason Ekstrand2018-03-075-0/+124
| | | | | | | | NIR has code to lower these away for us but we can do significantly better in many cases with register regioning and SIMD4x2. Acked-by: Lionel Landwerlin <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]>
* intel/fs: Implement reduce and scan opeprationsJason Ekstrand2018-03-072-0/+162
| | | | | Acked-by: Lionel Landwerlin <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]>
* intel/fs: Add a helper for emitting scan operationsJason Ekstrand2018-03-072-0/+148
| | | | | | | | | | | | | This commit adds a helper to the builder for emitting "scan" operations. Given a binary operation #, a scan takes the vector [a0, a1, ..., aN] and returns the vector [a0, a0 # a1, ..., a0 # a1 # ... # aN] where each channel contains the combination of all previous channels. The sequence of instructions to perform the scan is fairly optimal; a 16-wide scan on a 32-bit type is only 6 instructions. The subgroup scan and reduction operations will be implemented in terms of this. Acked-by: Lionel Landwerlin <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]>
* intel/fs: Add a couple of simple helper opcodesJason Ekstrand2018-03-074-0/+76
| | | | | Acked-by: Lionel Landwerlin <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]>
* i965/fs: Add support for nir_intrinsic_shuffleJason Ekstrand2018-03-077-0/+150
| | | | Reviewed-by: Iago Toral Quiroga <[email protected]>
* i965/fs: Support nir_intrinsic_vote_feqJason Ekstrand2018-03-071-0/+6
| | | | | Reviewed-by: Lionel Landwerlin <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]>
* nir: Generalize nir_intrinsic_vote_eqJason Ekstrand2018-03-071-1/+1
| | | | | | | | | | | | | The SPIR-V extension wants us to be able to do an AllEqual on any vector or scalar type. This has two implications: 1) We need to be able to handle vectors so we switch the vote_eq intrinsics to be vectorized intrinsics. 2) We need to handle floats which have different behavior with respect to +-0, NaN, etc. than the integer variant so we need two variants. Reviewed-by: Lionel Landwerlin <[email protected]>
* i965/fs: Implement basic SPIR-V subgroup intrinsicsJason Ekstrand2018-03-072-0/+26
| | | | | Reviewed-by: Samuel Iglesias Gonsálvez <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]>
* anv: Trivially implement VK_KHR_device_groupJason Ekstrand2018-03-071-0/+1
| | | | Reviewed-by: Iago Toral Quiroga <[email protected]>
* anv: Implement vkCmdDispatchBaseJason Ekstrand2018-03-071-0/+3
| | | | | | | | | | This is part of the device groups extension/feature but it's a decent chunk of work in its own right so it's worth breaking into its own patch. The mechanism we use is fairly straightforward: we just push the base work group id into the shader and add it to the work group id we get from dispatch. Reviewed-by: Samuel Iglesias Gonsálvez <[email protected]>
* intel: Split gen_device_info out into libintel_devJordan Justen2018-03-053-3/+3
| | | | | | | | | | | | Split out the device info so isl doesn't depend on intel/common. Now it will depend on the new intel/dev device info lib. This will allow the decoder in intel/common to use isl, allowing us to apply Ken's patch that removes the genxml duplication of surface formats. Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Samuel Iglesias Gonsálvez <[email protected]>
* i965: Silence warnings about mixing enum and non-enum in conditionalIan Romanick2018-03-021-1/+1
| | | | | | | | | | | | | | | | | | | | Reduces my build from 6451 warnings to 6301 warnings by silencing 150 instances of ../../SOURCE/master/src/intel/compiler/brw_inst.h: In function ‘brw_reg_type brw_inst_src1_type(const gen_device_info*, const brw_inst*)’: ../../SOURCE/master/src/intel/compiler/brw_inst.h:802:55: warning: enumeral and non-enumeral type in conditional expression [-Wextra] unsigned file = __builtin_strcmp("dst", #reg) == 0 ? \ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~ BRW_GENERAL_REGISTER_FILE : \ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ brw_inst_##reg##_reg_file(devinfo, inst); \ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ../../SOURCE/master/src/intel/compiler/brw_inst.h:811:1: note: in expansion of macro ‘REG_TYPE’ REG_TYPE(src1) ^~~~~~~~ Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Samuel Iglesias Gonsálvez <[email protected]>
* intel/compiler: Silence unused parameter warnings in release buildsIan Romanick2018-03-022-6/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Reduces my build from 7005 warnings to 6451 warnings by silencing 554 instances of In file included from ../../SOURCE/master/src/intel/compiler/brw_disasm.c:28:0: ../../SOURCE/master/src/intel/compiler/brw_inst.h: In function ‘brw_inst_3src_a1_src0_imm’: ../../SOURCE/master/src/intel/compiler/brw_inst.h:346:57: warning: unused parameter ‘devinfo’ [-Wunused-parameter] brw_inst_3src_a1_src0_imm(const struct gen_device_info *devinfo, ^~~~~~~ ../../SOURCE/master/src/intel/compiler/brw_inst.h: In function ‘brw_inst_3src_a1_src2_imm’: ../../SOURCE/master/src/intel/compiler/brw_inst.h:354:57: warning: unused parameter ‘devinfo’ [-Wunused-parameter] brw_inst_3src_a1_src2_imm(const struct gen_device_info *devinfo, ^~~~~~~ ../../SOURCE/master/src/intel/compiler/brw_inst.h: In function ‘brw_inst_set_3src_a1_src0_imm’: ../../SOURCE/master/src/intel/compiler/brw_inst.h:362:61: warning: unused parameter ‘devinfo’ [-Wunused-parameter] brw_inst_set_3src_a1_src0_imm(const struct gen_device_info *devinfo, ^~~~~~~ ../../SOURCE/master/src/intel/compiler/brw_inst.h: In function ‘brw_inst_set_3src_a1_src2_imm’: ../../SOURCE/master/src/intel/compiler/brw_inst.h:370:61: warning: unused parameter ‘devinfo’ [-Wunused-parameter] brw_inst_set_3src_a1_src2_imm(const struct gen_device_info *devinfo, ^~~~~~~ ../../SOURCE/master/src/intel/compiler/brw_inst.h: In function ‘brw_inst_imm_uq’: ../../SOURCE/master/src/intel/compiler/brw_inst.h:703:47: warning: unused parameter ‘devinfo’ [-Wunused-parameter] brw_inst_imm_uq(const struct gen_device_info *devinfo, const brw_inst *insn) ^~~~~~~ In file included from ../../SOURCE/master/src/intel/compiler/brw_shader.h:29:0, from ../../SOURCE/master/src/intel/compiler/brw_disasm.c:29: ../../SOURCE/master/src/intel/compiler/brw_compiler.h: In function ‘brw_stage_has_packed_dispatch’: ../../SOURCE/master/src/intel/compiler/brw_compiler.h:1277:61: warning: unused parameter ‘devinfo’ [-Wunused-parameter] brw_stage_has_packed_dispatch(const struct gen_device_info *devinfo, ^~~~~~~ ../../SOURCE/master/src/intel/compiler/brw_disasm.c: In function ‘src_ia1’: ../../SOURCE/master/src/intel/compiler/brw_disasm.c:849:18: warning: unused parameter ‘_reg_file’ [-Wunused-parameter] unsigned _reg_file, ^~~~~~~~~ Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Samuel Iglesias Gonsálvez <[email protected]>
* intel: Drop program size pointer from vec4/fs assembly getters.Kenneth Graunke2018-03-029-25/+17
| | | | | | | | | These days, we're just passing a pointer to a prog_data field, which we already have access to. We can just use it directly. (In the past, it was a pointer to a separate value.) Reviewed-by: Iago Toral Quiroga <[email protected]>
* intel/compiler: Memory fence commit must always be enabled for gen10+Anuj Phogat2018-03-021-1/+3
| | | | | | | | | | | | Commit bit in the message descriptor (Bit 13) must be always set to true in CNL+ for memory fence messages. It also fixes a piglit GPU hang on cnl+ in simulation environment. Piglit test: arb_shader_image_load_store-shader-mem-barrier See HSD ES # 1404612949 Signed-off-by: Anuj Phogat <[email protected]> Cc: [email protected] Reviewed-by: Francisco Jerez <[email protected]>
* Revert "i965/fs: Predicate byte scattered writes if needed"Francisco Jerez2018-03-021-14/+1
| | | | | | | | | | | | | This reverts commit a4031bdfa927fb4c3c5d0bdadc70634f3c1a5eac. It's redundant with the sample mask predication done at this point by the common logical send lowering infrastructure, and rather buggy because it wasn't applying the correct sample mask in shaders using discard, since the dispatch mask returned by FS_OPCODE_MOV_DISPATCH_TO_FLAGS doesn't reflect samples discarded by the shader, so it could have led to data corruption in fragment shader invocations that execute discard based on a non-dynamically uniform condition. Reviewed-by: Kenneth Graunke <[email protected]>
* intel/fs: Handle surface opcode sample masks via predication.Francisco Jerez2018-03-021-1/+42
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The main motivation is to enable HDC surface opcodes on ICL which no longer allows the sample mask to be provided in a message header, but this is enabled all the way back to IVB when possible because it decreases the instruction count of some shaders using HDC messages significantly, e.g. one of the SynMark2 CSDof compute shaders decreases instruction count by about 40% due to the removal of header setup boilerplate which in turn makes a number of send message payloads more easily CSE-able. Shader-db results on SKL: total instructions in shared programs: 15325319 -> 15314384 (-0.07%) instructions in affected programs: 311532 -> 300597 (-3.51%) helped: 491 HURT: 1 Shader-db results on BDW where the optimization needs to be disabled in some cases due to hardware restrictions: total instructions in shared programs: 15604794 -> 15598028 (-0.04%) instructions in affected programs: 220863 -> 214097 (-3.06%) helped: 351 HURT: 0 The FPS of SynMark2 CSDof improves by 5.09% ±0.36% (n=10) on my SKL laptop with this change. According to Eero this improves performance of the same test by 9% on BYT and by 7-8% on BXT J4205 and on SKL GT2 desktop. Reviewed-by: Kenneth Graunke <[email protected]> Tested-By: Eero Tamminen <[email protected]>
* intel/eu: Plumb header present bit to codegen helpers for HDC messages.Francisco Jerez2018-03-024-29/+50
| | | | | | | | | | This makes sure that the header-present bit of the message descriptor is in sync with the IR instruction fields, which gives the optimizer more control to avoid the overhead of setting up a message header when it's possible to do so. Reviewed-by: Jordan Justen <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* intel/ir: Allow arbitrary scratch flag registers for ↵Francisco Jerez2018-03-023-4/+6
| | | | | | | | | | | | | | SHADER_OPCODE_FIND_LIVE_CHANNEL. This shouldn't cause any functional change at this point, it changes SHADER_OPCODE_FIND_LIVE_CHANNEL to use the flag register specified at the IR level instead of the hard-coded f1.0, now that it can be represented in backend_instruction::flag_subreg. This will be necessary for scheduling to behave correctly once more things start making use of f1.0. Reviewed-by: Jordan Justen <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* intel/ir: Allow representing additional flag subregisters in the IR.Francisco Jerez2018-03-027-14/+24
| | | | | | | | | This allows representing conditional mods and predicates on f1.0-f1.1 at the IR level by adding an extra bit to the flag_subreg backend_instruction field. Reviewed-by: Jordan Justen <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* intel/fs: Set up sampler message headers in the visitor on gen7+Jason Ekstrand2018-03-012-22/+39
| | | | | | | | | | | | | | This gives the scheduler visibility into the headers which should improve scheduling. More importantly, however, it lets the scheduler know that the header gets written. As-is, the scheduler thinks that a texture instruction only reads it's payload and is unaware that it may write to the first register so it may reorder it with respect to a read from that register. This is causing issues in a couple of Dota 2 vertex shaders. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104923 Cc: [email protected] Reviewed-by: Francisco Jerez <[email protected]>
* spirv/i965/anv: Relax push constant offset assertions being 32-bit alignedJose Maria Casanova Crespo2018-02-281-5/+10
| | | | | | | | | | | | | | | | The introduction of 16-bit types with VK_KHR_16bit_storages implies that push constant offsets could be multiple of 2-bytes. Some assertions are updated so offsets should be just multiple of size of the base type but in some cases we can not assume it as doubles aren't aligned to 8 bytes in some cases. For 16-bit types, the push constant offset takes into account the internal offset in the 32-bit uniform bucket adding 2-bytes when we access not 32-bit aligned elements. In all 32-bit aligned cases it just becomes 0. v2: Assert offsets to be aligned to the dest type size. (Jason Ekstrand) Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Support 16-bit store_ssbo with VK_KHR_relaxed_block_layoutJose Maria Casanova Crespo2018-02-281-7/+15
| | | | | | | | | | | | | | | | Restrict the use of untyped_surface_write with 16-bit pairs in ssbo to the cases where we can guarantee that offset is multiple of 4. Taking into account that VK_KHR_relaxed_block_layout is available in ANV we can only guarantee that when we have a constant offset that is multiple of 4. For non constant offsets we will always use byte_scattered_write. v2: (Jason Ekstrand) - Assert offset_reg to be multiple of 4 if it is immediate. Reviewed-by: Jason Ekstrand <[email protected]>
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-29 11:22:17 +0000