aboutsummaryrefslogtreecommitdiffstats
path: root/src/mesa/drivers/dri
Commit message (Collapse)AuthorAgeFilesLines
* i965: Integrate precise trig into configuration infrastructureGurchetan Singh2016-06-074-2/+10
| | | | | | | | | | | | | | With this change, to enable precise SIN and COS instructions on Intel hardware, one can put <option name="precise_trig" value="true"/> in the proper drirc file. V2: Make option name more generic Reviewed-by: Kenneth Graunke <[email protected]> Reviewed-by: Stephane Marchesin <[email protected]>
* i965/gen8: fix cull distance emission for tessellation shaders.Dave Airlie2016-06-071-3/+5
| | | | | | | | | | This fixes some cases of: GL45-CTS.cull_distance.functional on Skylake. Reviewed-by: Chris Forbes <[email protected]> Cc: "12.0" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* mesa: hook up core bits of GL_ARB_shader_group_voteIlia Mirkin2016-06-061-0/+5
| | | | | | Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Ian Romanick <[email protected]> Reviewed-by: Dave Airlie <[email protected]>
* i965/gs/scalar: Fix load input for doublesSamuel Iglesias Gonsálvez2016-06-061-18/+54
| | | | | | Signed-off-by: Samuel Iglesias Gonsálvez <[email protected]> Reviewed-by: Timothy Arceri <[email protected]> Cc: "12.0" <[email protected]>
* i965/fs: fix offset when loading double vector input varyingsSamuel Iglesias Gonsálvez2016-06-061-1/+21
| | | | | | | | | | | | | | | | | When we are not packing a double input varying, we might need to read its data in a non-aligned to 64-bit offset, so we read the wrong data. This is happening when using explicit locations in varyings because Mesa disables packing varying for that case. const_index is in 32-bit size units but offset() is multiplying it by destination type size units. When operating with double input varyings, const_index value could be not aligned to 64 bits. To fix it, we load the double vector as if it was a float based vector with twice the number of components. Signed-off-by: Samuel Iglesias Gonsálvez <[email protected]> Reviewed-by: Timothy Arceri <[email protected]> Cc: "12.0" <[email protected]>
* i965/fs: fix FS_OPCODE_CINTERP for unpacked double input varyingsSamuel Iglesias Gonsálvez2016-06-061-1/+12
| | | | | | | | | | | Data starts at suboffet 3 in 32-bit units (12 bytes), so it is not 64-bit aligned and the current implementation fails to read the data properly. Instead, when there is is a double input varying, read it as vector of floats with twice the number of components. Signed-off-by: Samuel Iglesias Gonsálvez <[email protected]> Reviewed-by: Timothy Arceri <[email protected]> Cc: "12.0" <[email protected]>
* i965: don't use NumLayers for 3D textures.Dave Airlie2016-06-061-1/+1
| | | | | | | | | | | | For 3D textures we shouldn't be using NumLayers, we need to get it from the depth. This fixes: GL45-CTS.geometry_shader.layered_framebuffer.clear_call_support Reviewed-by: Eduardo Lima Mitev <[email protected]> Cc: "12.0" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* i965/ps_state: Use wm_prog_data.has_side_effectsJason Ekstrand2016-06-032-9/+6
| | | | | Signed-off-by: Jason Ekstrand <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* i965/fs Add a wm_prog_data bit for has_side_effectsJason Ekstrand2016-06-032-0/+15
| | | | | | | | | | | This is more accurate than calling _mesa_active_fragment_shader_has_side_effects because it looks at whether or not the SSBOs, images, or atomic buffers are actually written rather than just existing in the program. Signed-off-by: Jason Ekstrand <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]> Cc: "12.0" <[email protected]>
* Revert "i965/fs: Allow scalar source regions on SNB math instructions."Francisco Jerez2016-06-033-7/+14
| | | | | | | | | | | | | | | | | | | | | | | | | | This reverts commit c1107cec44ab030c7fcc97c67baa12df1cc9d7b5. Apparently the hardware spec text I quoted in the commit message was outright lying about scalar source math being supported on SNB, the hardware seems to load 32 contiguous bits of data for each channel regardless of the regioning mode. Fixes regressions in the following CTS tests (which we didn't catch early due to CTS being temporarily disabled in our CI system): es2-cts.gtf.gl.atan.atan_vec3_frag_xvary es2-cts.gtf.gl.cos.cos_vec2_frag_xvary es2-cts.gtf.gl.atan.atan_vec2_frag_xvary es2-cts.gtf.gl.pow.pow_vec2_frag_xvary_yconsthalf es2-cts.gtf.gl.cos.cos_float_frag_xvary es2-cts.gtf.gl.pow.pow_float_frag_xvary_yconsthalf es2-cts.gtf.gl.atan.atan_vec3_frag_xvaryyvary es2-cts.gtf.gl.pow.pow_vec3_frag_xvary_yconsthalf es2-cts.gtf.gl.cos.cos_vec3_frag_xvary es2-cts.gtf.gl.atan.atan_vec2_frag_xvaryyvary Cc: [email protected] Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96346 Reported-by: Mark Janes <[email protected]> Acked-by: Matt Turner <[email protected]>
* i965/vec4: Fix cmod propagation not to propagate non-identity cmod into CMP(N).Francisco Jerez2016-06-031-0/+12
| | | | | | | | | | | | | | | | | | | | | | | | | The conditional mod of these instructions determines the semantics of the comparison itself (rather than being evaluated based on the result of the instruction as is usually the case for most other instructions that allow conditional mods), so it's in general not legal to propagate a conditional mod into a CMP instruction. This prevents cmod propagation from (mis)optimizing: cmp.z.f0 tmp, ... mov.z.f0 null, tmp into: cmp.z.f0 tmp, ... which gives the negation of the flag result of the original sequence. I originally noticed this while working on SIMD32 in the scalar back-end, but the same scenario is likely to be possible in vec4 programs so this commit ports the bugfix with the same name from the scalar back-end to the vec4 cmod propagation pass. Cc: [email protected] Reviewed-by: Jason Ekstrand <[email protected]>
* isl: add support for Android libmesa_isl static libraryMauro Rossi2016-06-021-1/+2
| | | | | | | | | | | | | | | | isl library is needed to build i965, libmesa_isl static library is added to fix related Android building errors. Any attempt to build libmesa_genxml as phony package module failed to deliver gen{7,75,8,9}_pack.h generated headers, needed for libmesa_isl_gen{7,75,8,9} Due to constraints in Android Build System, libmesa_genxml is built as static, at least one source is needed, so dummy.c is autogenerated for this scope, libmesa_genxml dependency is declared using LOCAL_WHOLE_STATIC_LIBRARIES, to avoid building errors due to missing genxml/gen{7,75,8,9}_pack.h headers. Cc: <[email protected]> Reviewed-by: Emil Velikov <[email protected]>
* i965: Add _NEW_POINT to a couple of comments.Kenneth Graunke2016-06-023-3/+3
| | | | | Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]>
* i965/fs: Reindent emit_zip().Francisco Jerez2016-06-021-14/+14
| | | | | Cc: [email protected] Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Skip SIMD lowering destination zipping if possible.Francisco Jerez2016-06-021-0/+55
| | | | | | | | | | | | | | | | | | Skipping the temporary allocation and copy instructions is easy (just return dst), but the conditions used to find out whether the copy can be optimized out safely without breaking the program are rather complex: The destination must be exactly one component of at most the execution width of the lowered instruction, and all source regions of the instruction must be either fully disjoint from the destination or be aligned with it group by group. v2: Don't handle partial source-destination overlap for simplicity (Jason). No instruction count regressions with respect to v1 in either shader-db or the few FP64 shader_runner test-cases with partial overlap I've checked manually. Cc: [email protected] Reviewed-by: Jason Ekstrand <[email protected]>
* blorp: Fix 16x multisample scaled blitsAnuj Phogat2016-06-021-7/+10
| | | | | | | | | Piglit test ext_framebuffer_multisample_blit_scaled-blit-scaled (with added 16x sample support) now passes with this patch. Cc: "12.0" <[email protected]> Signed-off-by: Anuj Phogat <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Add missing types to type_sz().Matt Turner2016-06-021-1/+5
| | | | | | | | Coverity warns in multiple places about the potential for division by zero, caused by this function's default case. Reviewed-by: Francisco Jerez <[email protected]> Reviewed-by: Jordan Justen <[email protected]>
* i965/formatquery: remove INTERNALFORMAT_PREFERRED implementationAlejandro Piñeiro2016-06-021-71/+0
| | | | | | | | | | | | | Right now the implementation only checks if the internalformat is supported or not. But that implementation is wrong, returning unsupported for some internalformats. Additionally, checking if the internalformat is supported or not is already done at mesa/main before calling the driver hook, so this new check is not needed. Acked-by: Eduardo Lima <[email protected]> Acked-by: Antia Puentes <[email protected]> Reviewed-by: Ian Romanick <[email protected]>
* i965/eu: use simd8 when exec_size != EXECUTE_16Alejandro Piñeiro2016-06-021-2/+2
| | | | | | | | | | Among other thigs, fix a gpu hang when using INTEL_DEBUG=shader_time for any shader. Signed-off-by: Jason Ekstrand <[email protected]> Signed-off-by: Alejandro Piñeiro <[email protected]> Reviewed-by: Francisco Jerez <[email protected]>
* i965: Remove old CS local ID handlingJordan Justen2016-06-016-120/+2
| | | | | | | | | | | | | The old method pushed data for each channels uvec3 data of gl_LocalInvocationID. The new method pushes 1 dword of data that is a 'thread local ID' value. Based on that value, we can generate gl_LocalInvocationIndex and gl_LocalInvocationID with some calculations. Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Enable cross-thread constants and compact local IDs for hsw+Jordan Justen2016-06-013-14/+6
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The cross thread constant support appears on Haswell. It allows us to upload a set of uniform data for all threads without duplicating it per thread. One complication is that cross-thread constants are loaded into registers before per-thread constants. Previously, our local IDs were loaded before the uniform data and treated as 'payload' data, even though they were actually pushed into the registers like the other uniform data. Therefore, in this patch we simultaneously enable a newer layout where each thread now uses a single uniform slot for a unique local ID for the thread. This uniform is handled specially to make sure it is added last into the uniform push constant registers. This minimizes our usage of push constant registers, and maximizes our ability to use cross-thread constants for registers. To swap from the old to the new layout, we also need to flip some lowering pass switches to let our driver handle the lowering instead. We also no longer force thread_local_id_index to -1. v4: * Minimize size of patch that switches from the old local ID layout to the new layout (Jason) Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Support new local ID push constant & cross-thread constantsJordan Justen2016-06-012-45/+52
| | | | | | | | | | | | | | | | | The cross thread constant support appears on Haswell. It allows us to upload a set of uniform data for all threads without duplicating it per thread. We also support per-thread data which allows us to store a per-thread ID in one of the uniforms that can be used to calculate the gl_LocalInvocationIndex and gl_LocalInvocationID variables. v4: * Support the old local ID push constant layout as well (Jason) Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Add CS push constant info to brw_cs_prog_dataJordan Justen2016-06-012-0/+73
| | | | | | | | | | | | | | | | | | | | | | | | | | We need information about push constants in a few places for the GL driver, and another couple places for the vulkan driver. When we add support for uploading both a common (cross-thread) set of push constants, combined with the previous per-thread push constant data, things are going to get even more complicated. To simplify things, we add push constant info into the cs prog_data struct. The cross-thread constant support is added as of Haswell. To support it we need to make sure all push constants with uniform values are added to earlier registers. The register that varies per thread and holds the thread invocation's unique local ID needs to be added last. For now we add the code that would calculate cross-thread constatn information for hsw+, but we force it (cross_thread_supported) off until the other parts of the driver support it. v4: * Support older local ID push constant layout as well. (Jason) Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Store number of threads in brw_cs_prog_dataJordan Justen2016-06-013-25/+23
| | | | | | Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Add nir based intrinsic lowering and thread ID uniformJordan Justen2016-06-014-0/+190
| | | | | | | | | | | | | | | | | | | | | | | | | We add a lowering pass for nir intrinsics. This pass can replace nir intrinsics with driver specific nir lower code. We lower the gl_LocalInvocationIndex intrinsic based on a uniform which is loaded with a thread specific ID. We also lower the gl_LocalInvocationID based on gl_LocalInvocationIndex. v2: * Create variable during lowering pass. (Ken) v3: * Don't create a variable, but instead just insert an intrisic call to load a uniform from the allocated location. (Jason) v4: * Don't run this pass if thread_local_id_index < 0 Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Put CS local thread ID uniform in last push registerJordan Justen2016-06-011-1/+25
| | | | | | | | | | | | | | | | | | | | | | | | | | | This thread ID uniform will be used to compute the gl_LocalInvocationIndex and gl_LocalInvocationID values. It is important for this uniform to be added in the last push constant register. fs_visitor::assign_constant_locations is updated to make sure this happens. The reason this is important is that the cross-thread push constant registers are loaded first, and the per-thread push constant registers are loaded after that. (Broadwell adds another push constant upload mechanism which reverses this order, but we are ignoring this for now.) v2: * Add variable in intrinsics lowering pass * Make sure the ID is pushed last in assign_constant_locations, and that we save a spot for the ID in the push constants v3: * Simplify code based with Jason's suggestions. Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Add uniform for a CS thread local base IDJordan Justen2016-06-013-1/+21
| | | | | | | | | | | | v4: * Force thread_local_id_index to -1 for now, and have fs_visitor::setup_cs_payload look at thread_local_id_index. This enables us to more easily cut over from the old local ID layout to the new layout, as suggested by Jason. Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Add nir channel_num system valueJordan Justen2016-06-011-0/+15
| | | | | | | | | v2: * simd16/32 fixes (curro) Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* nir: Make lowering gl_LocalInvocationIndex optionalJordan Justen2016-06-011-1/+2
| | | | | | Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* glsl: Add glsl LowerCsDerivedVariables optionJordan Justen2016-06-011-0/+1
| | | | | | | | | | v2: * Move lower flag to context constants. (Ken) Cc: "12.0" <[email protected]> Signed-off-by: Jordan Justen <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]> (v1) Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Copy the offset when lowering logical pull constant sendsJason Ekstrand2016-06-011-0/+8
| | | | | | | | This fixes 64 Vulkan CTS tests per gen Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96299 Reviewed-by: Francisco Jerez <[email protected]> Cc: "12.0" <[email protected]>
* i965: Fix isoline reads in scalar TES.Kenneth Graunke2016-06-011-1/+1
| | | | | | | | | | | | | Isolines aren't reversed. commit 5b2d8c2273c6f fixed this for the vec4 TES backend, but not the scalar one. Found while debugging GL45-CTS.tessellation_shader. tessellation_control_to_tessellation_evaluation.gl_tessLevel. Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Dave Airlie <[email protected]> Reviewed-by: Iago Toral Quiroga <[email protected]> Cc: [email protected]
* i965: If control_data_header_size_bits is zero, don't do EndPrimitiveIan Romanick2016-06-012-0/+6
| | | | | | | | This can occur when max_vertices=0 is explicitly specified. Signed-off-by: Ian Romanick <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]> Cc: "12.0" <[email protected]>
* i965: Add norbc debug optionTopi Pohjolainen2016-06-013-0/+4
| | | | | | | | | | | | | | This INTEL_DEBUG option disables lossless compression (also known as render buffer compression). v2: (Matt) Use likely(!lossless_compression_disabled) instead of !likely(lossless_compression_disabled) (Grazvydas) Update docs/envvars.html Cc: "12.0" <[email protected]> Signed-off-by: Topi Pohjolainen <[email protected]> Reviewed-by: Matt Turner <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* i965/gen9: Configure rbc buffers as plain for non-rbc tex viewsTopi Pohjolainen2016-06-012-3/+48
| | | | | | | | | | | | | | | | | | | | | | Fixes rendering in Shadow of Mordor with rbc. Application writes RGBA_UNORM texture filling it with values the application wants to later on treat as SRGB_ALPHA. Intel driver enables lossless compression for the buffer by the time of writing. However, the driver fails to make sure the buffer can be sampled as something else later on and unfortunately there is restriction in the hardware for using lossless compression for srgb formats which looks to extend itself to the sampling engine also. Requesting srgb to linear conversion on top of compressed buffer results the color values to be pretty much garbage. Fortunately none of tracked benchmarks showed a regression with this. v2 (Matt): Add missing space Cc: "12.0" <[email protected]> Signed-off-by: Topi Pohjolainen <[email protected]> Reviewed-by: Kenneth Graunke <[email protected]>
* i965: Fix the passthrough TCS for isolines.Kenneth Graunke2016-05-311-7/+12
| | | | | | | | | | | | | | | | We weren't setting up several of the uniform values for the patch header, so we'd crash when uploading push constants. We at least need to initialize them to zero. We also had the isoline parameters reversed, so it would also render incorrectly (if it didn't crash). Fixes a new Piglit test(*) (isoline-no-tcs), as well as crashes in GL44-CTS.tessellation_shader.single.max_patch_vertices. (*) https://lists.freedesktop.org/archives/piglit/2016-May/019866.html Signed-off-by: Kenneth Graunke <[email protected]> Reviewed-by: Dave Airlie <[email protected]> Cc: [email protected]
* i965/xfb: skip components in correct buffer.Dave Airlie2016-06-011-4/+6
| | | | | | | | | | | The driver was adding the skip components but always for buffer 0. This fixes: GL45-CTS.gtf40.GL3Tests.transform_feedback3.transform_feedback3_skip_multiple_buffers Reviewed-by: Kenneth Graunke <[email protected]> Cc: "12.0 11.2" <[email protected]> Signed-off-by: Dave Airlie <[email protected]>
* i965/fs: Allow scalar source regions on SNB math instructions.Francisco Jerez2016-05-313-17/+8
| | | | | | | | | | | | | | | | I haven't found any evidence that this isn't supported by the hardware, in fact according to the SNB hardware spec: "The supported regioning modes for math instructions are align16, align1 with the following restrictions: - Scalar source is supported. [...] - Source and destination offset must be the same, except the case of scalar source." Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]> Reviewed-by: Matt Turner <[email protected]>
* i965/fs: Fix constant combining for instructions that cannot accept source mods.Francisco Jerez2016-05-311-4/+3
| | | | | | | | | | This is the case for SNB math instructions so we need to be careful and insert the literal value of the immediate into the table (rather than its absolute value) if the instruction is unable to invert the sign of the constant on the fly. Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Extend remove_duplicate_mrf_writes() to handle non-VGRF to MRF copies.Francisco Jerez2016-05-311-8/+8
| | | | | Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Fix compute_to_mrf() to coalesce VGRFs initialized by multiple ↵Francisco Jerez2016-05-311-10/+36
| | | | | | | | | | | | | | | | | | | | | | single-GRF writes. Which requires using a bitset instead of a boolean flag to keep track of the GRFs we've seen a generating instruction for already. The search loop continues until all instructions initializing the value of the source VGRF have been found, or it is determined that coalescing is not possible. Fixes a few piglit test cases on Gen4-6 which were regressed by 6956015aa514f2d06d0e4b33bfe6bca83142fbf0 due to the different (yet perfectly valid) ordering in which copy instructions are emitted now by the simd lowering pass, which had the side effect of causing this optimization pass to start corrupting the program in cases where a VGRF-to-MRF copy instruction would be eliminated but only the last instruction writing to the source VGRF region would be rewritten to point to the target MRF. Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Teach compute_to_mrf() about the COMPR4 address transformation.Francisco Jerez2016-05-311-1/+23
| | | | | | | | | This will be required to correctly transform the destination of 8-wide instructions that write a single GRF of a VGRF to MRF copy marked COMPR4. Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Refactor compute_to_mrf() to split search and rewrite into separate ↵Francisco Jerez2016-05-311-10/+25
| | | | | | | | | | | | | | | loops. This will allow compute_to_mrf to handle cases where the source of the VGRF-to-MRF copy is initialized by more than one instruction. In such cases we cannot rewrite the destination of any of the generating instructions until it's known whether the whole VGRF source region can be coalesced into the destination MRF, which will imply continuing the search until all generating instructions have been found or it has been determined that the VGRF and MRF registers cannot be coalesced. Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Fix compute-to-mrf VGRF region coverage condition.Francisco Jerez2016-05-311-3/+6
| | | | | | | | | | | | | | | Compute-to-mrf was checking whether the destination of scan_inst is more than one component (making assumptions about the instruction data type) in order to find out whether the result is being fully copied into the MRF destination, which is rather inaccurate in cases where a single-component instruction is only partially contained in the source region, or when the execution size of the copy and scan_inst instructions differ. Instead check whether the destination region of the instruction is really contained within the bounds of the source region of the copy. Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Simplify and improve accuracy of compute_to_mrf() by using ↵Francisco Jerez2016-05-311-47/+13
| | | | | | | | | | | | | regions_overlap(). Compute-to-mrf was being rather heavy-handed about checking whether instruction source or destination regions interfere with the copy instruction, which could conceivably lead to program miscompilation. Fix it by using regions_overlap() instead of the open-coded and dubiously correct overlap checks. Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Teach regions_overlap() about COMPR4 MRF regions.Francisco Jerez2016-05-311-3/+17
| | | | | Cc: "12.0" <[email protected]> Reviewed-by: Jason Ekstrand <[email protected]>
* i965: Expose GL 4.3 on Gen8+.Francisco Jerez2016-05-292-2/+2
| | | | | | ARB_compute_shader was the last feature missing. Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Skip gen4 pre/post-send dependency workaronds for the first/last block.Francisco Jerez2016-05-291-2/+2
| | | | | | | | | | | | | | | | | | We know that there cannot be any destination dependency race if we reach the beginning or end of the program without having found any other instruction the send could possibly race with. This avoids emitting a pile of useless moves at the beginning or end of the program in the most common case in which the program has a single basic block only. On the original i965 I get the following shader-db results: total instructions in shared programs: 3354165 -> 3215637 (-4.13%) instructions in affected programs: 3183065 -> 3044537 (-4.35%) helped: 13498 HURT: 0 Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Skip SIMD lowering source unzipping for regular scalar regions.Francisco Jerez2016-05-291-2/+21
| | | | Reviewed-by: Jason Ekstrand <[email protected]>
* i965/fs: Factor out region zipping and unzipping from the SIMD lowering pass.Francisco Jerez2016-05-291-60/+84
| | | | | | | | | | Just to make sure we keep the SIMD lowering pass tidy when we introduce additional logic to try to optimize out the copy instructions used to zip and unzip the destination and source regions into multiple packed regions of the lowered instruction width. Shouldn't cause any functional changes. Reviewed-by: Jason Ekstrand <[email protected]>
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-12 14:27:42 +0000