summaryrefslogtreecommitdiffstats
path: root/src
Commit message (Collapse)AuthorAgeFilesLines
* nir: Rely on the fact that bcsel takes a well formed boolean.Kenneth Graunke2016-08-191-3/+3
| | | | | | | | | | | | | | | According to Connor, it's safe to assume that the first operand of bcsel, as well as the operand of b2f and b2i, must be well formed booleans. https://lists.freedesktop.org/archives/mesa-dev/2016-August/125658.html With the previous improvements to a@bool handling, this now has no change in shader-db instruction counts on Broadwell. Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* i965/sched: Change the scheduling heuristics to favor early program termination.Francisco Jerez2016-08-181-3/+16
| | | | | | | | | | | | | | | | | | | | | | | | | | | | This uses the unblocked time of the exit assigned to each available node to attempt to unblock exit nodes as early as possible, potentially reducing the runtime of the shader when an exit branch is taken. There is a natural trade-off between terminating the program as early as possible and reducing the worst-case latency of the program as a whole (since this will typically move exit-unblocking nodes closer to its dependencies potentially causing additional stalls of the execution pipeline), but in practice the bandwidth and ALU cycle savings from terminating the program earlier tend to outweigh the slight increase in worst-case program execution latency, so it makes sense to prefer nodes likely to unblock an earlier exit regardless of the latency benefits of other available nodes. I haven't observed any benchmark regressions from this change after testing on VLV, HSW, BDW, BSW and SKL. The FPS of the GfxBench Manhattan benchmark increases by 10%-20% and the FPS of Unigine Valley improves by roughly 5% depending on the platform and settings. The change to the register pressure-sensitive heuristic is rather conservative and gives precedence to the existing heuristic in order to avoid increasing register pressure and causing spill count and SIMD width regressions in shader-db. It may make sense to revisit this with additional performance data. Reviewed-by: Jason Ekstrand <[email protected]>
* i965/sched: Assign a preferred exit node to each node of the dependency graph.Francisco Jerez2016-08-181-0/+59
| | | | | | | | | | | | | This adds a bit of metadata to schedule_node that will be used to compare available nodes in the scheduling heuristic code based on which of them unblocks the earliest successor exit node. Note that assigning exit nodes wouldn't be necessary in a bottom-up scheduler because we could achieve the same effect by scheduling the exit nodes themselves appropriately. No shader-db changes. Reviewed-by: Jason Ekstrand <[email protected]>
* i965/sched: Calculate the critical path of scheduling nodes non-recursively.Francisco Jerez2016-08-181-13/+12
| | | | | | | | | | | | | | | | | | | The critical path of each node is calculated by induction based on the critical paths of its children, which can be done in a post-order depth-first traversal of the dependency graph. The current code implements graph traversal by iterating over all nodes of the graph and then recursing into its children -- But it turns out that recursion is unnecessary because the lexical order of instructions in the block is already a good enough reverse post-order of the dependency graph (if it weren't a reverse post-order some instruction would have been located before one of its dependencies in the original ordering of the basic block, which is impossible), so we just need to walk the instruction list in reverse to achieve the same result more efficiently. No shader-db changes. Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Switch to per-subspan discard jumps.Francisco Jerez2016-08-181-3/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | ANY4H is more efficient than ANY8H and ANY16H because it makes sure that whenever a whole subspan hits a discard statement it gets disabled by the EU until the end of the program, regardless of whether the discard condition is uniform across all channels of the SIMD8-16 thread. OTOH ANY8H/ANY16H would cause the rest of the program to be executed for *all* channels if only one of the channels hadn't taken the discard branch, potentially increasing the bandwidth and ALU usage of the program unnecessarily. This change increases the FPS by over 3x of a simple micro-benchmark that discards a bunch of fragments and then does a single costly texturing operation. I've just re-verified the FPS change on HSW and SKL, but I expect all platforms from Gen6 up to get a similar benefit. Note that we could potentially be more aggressive and use the NORMAL predicate to discard individual channels, but that would need to happen post-scheduling because the scheduler currently doesn't care to reorder HALT instructions with respect to other instructions, and the NORMAL predicate would cause the results of subsequent derivative computations to become undefined -- If the scheduler didn't reorder HALT instructions it would actually be safe to switch to NORMAL because the behavior of derivative computations after a non-uniform discard statement is undefined by the GLSL spec, but that would make the optimization implemented by one of the following commits somewhat more difficult. Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Drop bogus writemasking disable bit from HALT instructions.Francisco Jerez2016-08-181-4/+0
| | | | | | | | | | This may have been the reason people ran into problems with non-uniform HALT instructions and ended up using the inefficient ANY16H/ANY8H predicates instead of ANY4H or NORMAL in order to prevent non-uniform discard. The HALT instruction is able to handle non-uniform execution masks just fine. Reviewed-by: Jason Ekstrand <[email protected]>
* mesa: avoid valgrind warning due to opaque only being set sometimesIlia Mirkin2016-08-181-2/+2
| | | | | | | | | | | Valgrind complains with a "Conditional jump or move depends on uninitialised value(s)" warning due to opaque being conditionally initialized. However in the punchthrough_alpha == true case, it is always initialized, so just flip the condition around to silence the warning. Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Chris Forbes <[email protected]>
* vbo: remove unnecessary max_basevertex computationIlia Mirkin2016-08-181-8/+0
| | | | | | | | The max basevertex is already computed and added into max_index by the caller, _tnl_draw_prims. Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Marek Olšák <[email protected]>
* vbo: add basevertex when looking up elements for vbo splittingIlia Mirkin2016-08-181-1/+1
| | | | | | | Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97351 Signed-off-by: Ilia Mirkin <[email protected]> Cc: [email protected] Reviewed-by: Marek Olšák <[email protected]>
* radeonsi: initialize and finalize the LLVM function pass managerMarek Olšák2016-08-181-0/+2
| | | | Reviewed-by: Tom Stellard <[email protected]>
* isl: automake: use VISIBILITY_CFLAGS to restrict symbol visibilityEmil Velikov2016-08-181-8/+10
| | | | | | | | v2: Add VISIBILITY_CFLAGS to AM_CFLAGS (Ken) Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]> (v1) Signed-off-by: Emil Velikov <[email protected]>
* anv: remove dummy VK_DEBUG_MARKER_EXT entry pointsmil Velikov2016-08-181-20/+0
| | | | | | | | | | | | The vkCmdDbgMarker{Begin,End} symbols are exported, yet the json does no advertise that the driver supports the extension. Furthermore the functions are empty stubs. Remove those until we get a proper implementation and json notation. Cc: "12.0" <[email protected]> Cc: Jason Ekstrand <[email protected]> Signed-off-by: Emil Velikov <[email protected]>
* anv: do not export the Vulkan APIEmil Velikov2016-08-181-11/+0
| | | | | | | | | | | | | | | | | | | | | | | | | With version 1 of the Loader interface there is an internal/private symbol (vk_icdGetInstanceProcAddr) which is used to retrieve all the API from the Vulkan entrypoints from the ICD. Implying that exposing the Vulkan API is not recommended. Version 2 goes a step further explicitly forbiding the ICD from exposing Vulkan symbols (and adding a negotiation API) As a reference: - Nvidia 367.35 Missing negotiation API - version 1. Exposes only vk_icdGetInstanceProcAddr. - AMD 16.30.3.306809 Have negotiation API - version 2, Exposes vk_icdGetInstanceProcAddr. Exposes a couple of Vulkan entry points - seems to be in violation with the spec. Cc: "12.0" <[email protected]> Cc: Christian König <[email protected]> Signed-off-by: Emil Velikov <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* anv: automake: build with -BsymbolicEmil Velikov2016-08-181-0/+1
| | | | | | | | | Explicitly suggested in the Loader interface version 2 section, but it's good idea either way. It essentially, ensures that our symbols are not interposed. Cc: "12.0" <[email protected]> Cc: Jason Ekstrand <[email protected]> Signed-off-by: Emil Velikov <[email protected]>
* anv: automake: use VISIBILITY_CFLAGS to restrict symbol visibilityEmil Velikov2016-08-182-3/+7
| | | | | | | | | | | Hide the internal symbols and annotate the vk_icdGetInstanceProcAddr as public since the loader needs it (since v1 of the loader interface). v2: Add VISIBILITY_CFLAGS to AM_CFLAGS (Ken) Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]> (v1) Signed-off-by: Emil Velikov <[email protected]>
* anv: remove internal 'validate' layerEmil Velikov2016-08-182-110/+11
| | | | | | | | | | | Presently the layer has only a single entry point. As mentioned by Jason the function does not validate anything that isn't checked elsewhere, thus we can drop the whole thing. Cc: "12.0" <[email protected]> Cc: Jason Ekstrand <[email protected]> Suggested-by: Jason Ekstrand <[email protected]> Signed-off-by: Emil Velikov <[email protected]>
* nir/search: Extend 'a@bool' to handle a couple of system values.Kenneth Graunke2016-08-181-0/+7
| | | | | | | | | | | | | | load_front_face and load_helper_invocation produce booleans. On Broadwell: total instructions in shared programs: 11638956 -> 11638011 (-0.01%) instructions in affected programs: 115093 -> 114148 (-0.82%) helped: 628 HURT: 14 Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* nir/search: Fold src_is_bool()/alu_instr_is_bool() into src_is_type().Kenneth Graunke2016-08-181-31/+19
| | | | | | | | | | | | I don't want src_is_bool() and src_is_type(x, nir_type_bool) to behave differently. Having the logic spread out over three functions makes it harder to decide where to put new logic, as well. So, combine them all. It's a bit simpler because there's now only one recursive function rather than a pair of mutually recursive functions. Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* nir/search: Introduce a src_is_type() helper for 'a@type' handling.Kenneth Graunke2016-08-181-13/+29
| | | | | | | | | | | | Currently, 'a@type' can only match if 'a' is produced by an ALU instruction. This is rather limited - there are other cases we can easily detect which we should handle. Extending the code in-place would be fairly messy, so we introduce a new src_is_type() helper. Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Fix barrier count shift in scalar TCS backend.Kenneth Graunke2016-08-181-1/+1
| | | | | | | | | | | | The "Barrier Count" field goes in 14:9 of m0.2. The vec4 backend correctly shifts by 9, but the scalar backend only shifted by 8. It's not like this changed - I think I just made a typo when writing the original scalar TCS backend code. Cc: [email protected] Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]>
* i965: Fix execution size of scalar TCS barrier setup code.Kenneth Graunke2016-08-181-6/+6
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Previously, the scalar TCS backend was generating: mov(8) g17<1>UD 0x00000000UD { align1 WE_all 1Q compacted }; and(8) g17.2<1>UD g0.2<0,1,0>UD 0x0001e000UD { align1 WE_all 1Q }; shl(8) g17.2<1>UD g17.2<8,8,1>UD 0x0000000bUD { align1 WE_all 1Q }; or(8) g17.2<1>UD g17.2<8,8,1>UD 0x00008200UD { align1 WE_all 1Q }; send(8) null<1>UW g17<8,8,1>UD gateway (barrier msg) mlen 1 rlen 0 { align1 WE_all 1Q }; This is rubbish - g17.2<8,8,1>UD spans two registers, and is an illegal region. Not to mention it clobbers 8 channels of data when we only wanted to touch m0.2. Instead, we want: mov(8) g17<1>UD 0x00000000UD { align1 WE_all 1Q compacted }; and(1) g17.2<1>UD g0.2<0,1,0>UD 0x0001e000UD { align1 WE_all }; shl(1) g17.2<1>UD g17.2<0,1,0>UD 0x0000000bUD { align1 WE_all }; or(1) g17.2<1>UD g17.2<0,1,0>UD 0x00008200UD { align1 WE_all }; send(8) null<1>UW g17<8,8,1>UD gateway (barrier msg) mlen 1 rlen 0 { align1 WE_all 1Q }; Using component() accomplishes this. Fixes GL44-CTS.tessellation_shader.tessellation_shader_tc_barriers. barrier_guarded_read_write_calls on Skylake. Probably fixes other barrier issues on Gen8+. v2: Use a group(1, 0) builder so inst->exec_size is set correctly (thanks to Francisco Jerez for catching that it was incorrect). Cc: [email protected] Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]> [v1] Reviewed-by: Francisco Jerez <[email protected]>
* i965: Implement the WaPreventHSTessLevelsInterference workaround.Kenneth Graunke2016-08-186-3/+175
| | | | | | | | | | | Fixes several GL44-CTS.tessellation_shader (and GL45 and ES31) subcases: - vertex_spacing - tessellation_shader_point_mode.points_verification - tessellation_shader_quads_tessellation.inner_tessellation_level_rounding Cc: [email protected] Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]>
* nir/builder: Add bany_inequal and bany helpers.Kenneth Graunke2016-08-181-0/+19
| | | | | | | | | | The first simply picks the bany_inequal[234] opcodes based on the SSA def's number of components. The latter implicitly compares with zero to achieve the same semantics of GLSL's any(). Cc: [email protected] Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Alejandro Piñeiro <[email protected]>
* mesa: Fix uf10_to_f32() scale factor in the E == 0 and M != 0 case.Kenneth Graunke2016-08-171-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | GL_EXT_packed_float, 2.1.B Unsigned 10-Bit Floating-Point Numbers: 0.0, if E == 0 and M == 0, 2^-14 * (M / 32), if E == 0 and M != 0, 2^(E-15) * (1 + M/32), if 0 < E < 31, INF, if E == 31 and M == 0, or NaN, if E == 31 and M != 0, In the second case (E == 0 and M != 0), we were multiplying the mantissa by 2^-20, when we should have been multiplying by 2^-19 (which is 2^(-14 + -5), or 2^-14 * 2^-5, or 2^-14 / 32). The previous section defines the formula for 11-bit numbers, which is: 2^-14 * (M / 64), if E == 0 and M != 0, In other words, we had accidentally copy and pasted the 11-bit code to the 10-bit case, and neglected to change the exponent. Fixes dEQP-GLES3.functional.pbo.renderbuffer.r11f_g11f_b10f_triangles when run with surface dimensions of 1536x1152 or 1920x1080. Cc: [email protected] References: https://code.google.com/p/chrome-os-partner/issues/detail?id=56244 Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Stephane Marchesin <[email protected]> Reviewed-by: Antia Puentes <[email protected]>
* swr: [rasterizer core] only use Viewport/Scissors during SwrDraw* operationsTim Rowley2016-08-1712-415/+400
| | | | | | | | | | | Add explicit rects for: - SwrClearRenderTarget - SwrDiscardRect - SwrInvalidateTiles - SwrStoreTiles Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer common] reorder SWR_FORMAT_INFOTim Rowley2016-08-172-825/+1433
| | | | Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer core] make dirtytile list point directly to macrotilequeuesTim Rowley2016-08-173-14/+15
| | | | | | Speeds up high geometry HPC workloads. Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer core] portability - remove use of INT64Tim Rowley2016-08-171-2/+2
| | | | Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer core] viewport transform disabled fixTim Rowley2016-08-171-4/+11
| | | | | | | When viewport transform is disabled (ie. screen space coords are passed in directly), the W component should be interpreted as RHW. Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer core] clamp scissor rects to current tile rectTim Rowley2016-08-171-0/+18
| | | | Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer core] align stats structuresTim Rowley2016-08-171-2/+2
| | | | Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer core] use AVX2 permute to simplify PaTriListTim Rowley2016-08-171-1/+35
| | | | Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer core] move some global variables to SWR_CONTEXTTim Rowley2016-08-172-9/+9
| | | | Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer core] change scale on VP matrix element gathersTim Rowley2016-08-171-6/+6
| | | | | | | Was 1, which led to pulling denorms for non-zero indices. Changed to sizeof(float). Signed-off-by: Tim Rowley <[email protected]>
* swr: [rasterizer] implementing native AVX-512 simd16 intrinsicsTim Rowley2016-08-172-84/+265
| | | | Signed-off-by: Tim Rowley <[email protected]>
* i965/blorp: Use nir_alu_type for the texture data typeJason Ekstrand2016-08-172-27/+16
| | | | | | This lets us remove the brw_reg.h include Reviewed-by: Topi Pohjolainen <[email protected]>
* i965: brw_blorp_blit.cpp -> blorp_blit.cJason Ekstrand2016-08-172-6/+6
| | | | Reviewed-by: Topi Pohjolainen <[email protected]>
* i965: brw_blorp_clear.cpp -> blorp_clear.cJason Ekstrand2016-08-172-5/+1
| | | | Reviewed-by: Topi Pohjolainen <[email protected]>
* i965: Split brw_blorp.c/h into multiple filesJason Ekstrand2016-08-1711-1373/+1495
| | | | | | | | | | | This mega-commit pulls most of the i965-specific bits of blorp into the brw_blorp.c/h files which now contain nothing but i965 wrappers around "core blorp" calls. The "core blorp" api is moved into blorp.h and the internal blorp data structures are moved into blorp_priv.h. The new file blorp.c is created to house "core blorp" internals which are pulled from the old brw_blorp.c Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/blorp: Factor the guts of blorp_hiz_exec into a helperJason Ekstrand2016-08-171-18/+25
| | | | Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/blorp: Break the guts of do_single_blorp_clear into two helpersJason Ekstrand2016-08-171-64/+111
| | | | | | | | The helpers are completely miptree-unaware and each fairly cleanly do a single thing. This does come at the downside of not doing proper debug reporting on whether or not we're doing replicated clears. Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/meta_util: Convert get_fast_clear_rect to take an isl_surfJason Ekstrand2016-08-173-14/+17
| | | | Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/blorp/clear: Move isl_surf setup higher in the functionJason Ekstrand2016-08-171-10/+10
| | | | Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/blorp: Refactor fast-clear logic a bitJason Ekstrand2016-08-171-12/+13
| | | | | | | | | This pulls the mcs allocation into the if statement where we initially determine that we are doing a fast clear and moves the programming of wm_inputs and figuring out the fast clear rect into it's own if statement. The next commit will put code inbetween the two. Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/blorp/clear: Stop stomping the destination formatJason Ekstrand2016-08-171-3/+1
| | | | | | | The blorp_surface_info_init call above should set the format for us and stomping it later does nothing whatsoever. Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/meta_util: Only modify the input parameters in get_fast_clear_rectJason Ekstrand2016-08-173-13/+4
| | | | | | | | We had another inline copy of brw_meta_get_buffer_rect embedded in get_fast_clear_rect for no good reason. This lets us get rid of the gl_frameuffer parameter to get_fast_clear_rect. Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/blorp: Stop calling brw_meta_get_buffer_rectJason Ekstrand2016-08-173-24/+0
| | | | | | | | We already have an inlined version of the function slightly higher up in do_single_blorp_clear and all calling it does is stomp the values with the same thing. We might as well just get rid of it. Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/blorp: Pull the guts of resolve_color into a miptree-agnostic helperJason Ekstrand2016-08-171-18/+25
| | | | Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/meta_util: Convert get_resolve_rect to use ISLJason Ekstrand2016-08-173-28/+28
| | | | Reviewed-by: Topi Pohjolainen <[email protected]>
* i965/blorp: Make the guts of brw_blorp_blit_miptrees miptree-unawareJason Ekstrand2016-08-172-19/+53
| | | | | | | | | Now that we have the brw_blorp_surf struct, we can start to make bits of blorp completely miptree-unaware. To start things off, we split the guts of brw_blorp_blit_miptrees into a brw_blorp_blit function which knows nothing about miptrees. Reviewed-by: Topi Pohjolainen <[email protected]>
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-26 11:26:47 +0000