From: Kenneth Graunke Date: Thu, 17 Aug 2017 07:44:41 +0000 (-0700) Subject: i965: Use ISL for emitting null surface states. X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=5ae983c85b8af7553223c1d3b07199ebb181d0d3;p=mesa.git i965: Use ISL for emitting null surface states. We handle the Sandybridge multisampled 2D surface hack here, rather than in ISL, because it requires allocating a BO, and is kind of messy. Reviewed-by: Jason Ekstrand --- diff --git a/src/mesa/drivers/dri/i965/Makefile.sources b/src/mesa/drivers/dri/i965/Makefile.sources index 425c883de84..9687eb957e1 100644 --- a/src/mesa/drivers/dri/i965/Makefile.sources +++ b/src/mesa/drivers/dri/i965/Makefile.sources @@ -74,10 +74,8 @@ i965_FILES = \ gen7_misc_state.c \ gen7_sol_state.c \ gen7_urb.c \ - gen7_wm_surface_state.c \ gen8_depth_state.c \ gen8_multisample_state.c \ - gen8_surface_state.c \ hsw_queryobj.c \ hsw_sol.c \ intel_batchbuffer.c \ diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c index d97a24fbf82..d157f059704 100644 --- a/src/mesa/drivers/dri/i965/brw_context.c +++ b/src/mesa/drivers/dri/i965/brw_context.c @@ -866,10 +866,10 @@ brwCreateContext(gl_api api, brw->gs.base.stage = MESA_SHADER_GEOMETRY; brw->wm.base.stage = MESA_SHADER_FRAGMENT; if (brw->gen >= 8) { - gen8_init_vtable_surface_functions(brw); + gen6_init_vtable_surface_functions(brw); brw->vtbl.emit_depth_stencil_hiz = gen8_emit_depth_stencil_hiz; } else if (brw->gen >= 7) { - gen7_init_vtable_surface_functions(brw); + gen6_init_vtable_surface_functions(brw); brw->vtbl.emit_depth_stencil_hiz = gen7_emit_depth_stencil_hiz; } else if (brw->gen >= 6) { gen6_init_vtable_surface_functions(brw); diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index d41e6aa7bde..a227a61f654 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -635,11 +635,6 @@ struct brw_context struct gl_renderbuffer *rb, uint32_t flags, unsigned unit, uint32_t surf_index); - void (*emit_null_surface_state)(struct brw_context *brw, - unsigned width, - unsigned height, - unsigned samples, - uint32_t *out_offset); /** * Send the appropriate state packets to configure depth, stencil, and diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h index 46665aae12b..dc893e5b7bd 100644 --- a/src/mesa/drivers/dri/i965/brw_state.h +++ b/src/mesa/drivers/dri/i965/brw_state.h @@ -232,14 +232,6 @@ void brw_update_renderbuffer_surfaces(struct brw_context *brw, uint32_t render_target_start, uint32_t *surf_offset); -/* gen7_wm_surface_state.c */ -void gen7_check_surface_setup(uint32_t *surf, bool is_render_target); -void gen7_init_vtable_surface_functions(struct brw_context *brw); - -/* gen8_surface_state.c */ - -void gen8_init_vtable_surface_functions(struct brw_context *brw); - /* brw_sampler_state.c */ void brw_emit_sampler_state(struct brw_context *brw, uint32_t *sampler_state, diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c index 8537d6eeeb5..ab55eee1ce4 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c @@ -833,72 +833,48 @@ const struct brw_tracked_state brw_wm_pull_constants = { * hardware discard the target 0 color output.. */ static void -brw_emit_null_surface_state(struct brw_context *brw, - unsigned width, - unsigned height, - unsigned samples, - uint32_t *out_offset) +emit_null_surface_state(struct brw_context *brw, + unsigned width, + unsigned height, + unsigned samples, + uint32_t *out_offset) { - /* From the Sandy bridge PRM, Vol4 Part1 p71 (Surface Type: Programming - * Notes): - * - * A null surface will be used in instances where an actual surface is - * not bound. When a write message is generated to a null surface, no - * actual surface is written to. When a read message (including any - * sampling engine message) is generated to a null surface, the result - * is all zeros. Note that a null surface type is allowed to be used - * with all messages, even if it is not specificially indicated as - * supported. All of the remaining fields in surface state are ignored - * for null surfaces, with the following exceptions: + uint32_t *surf = brw_state_batch(brw, + brw->isl_dev.ss.size, + brw->isl_dev.ss.align, + out_offset); + + if (brw->gen != 6 || samples <= 1) { + isl_null_fill_state(&brw->isl_dev, surf, + isl_extent3d(width, height, 1)); + return; + } + + /* On Gen6, null render targets seem to cause GPU hangs when multisampling. + * So work around this problem by rendering into dummy color buffer. * - * - [DevSNB+]: Width, Height, Depth, and LOD fields must match the - * depth buffer’s corresponding state for all render target surfaces, - * including null. + * To decrease the amount of memory needed by the workaround buffer, we + * set its pitch to 128 bytes (the width of a Y tile). This means that + * the amount of memory needed for the workaround buffer is + * (width_in_tiles + height_in_tiles - 1) tiles. * - * - Surface Format must be R8G8B8A8_UNORM. + * Note that since the workaround buffer will be interpreted by the + * hardware as an interleaved multisampled buffer, we need to compute + * width_in_tiles and height_in_tiles by dividing the width and height + * by 16 rather than the normal Y-tile size of 32. */ - unsigned surface_type = BRW_SURFACE_NULL; - struct brw_bo *bo = NULL; - unsigned pitch_minus_1 = 0; - uint32_t multisampling_state = 0; - uint32_t *surf = brw_state_batch(brw, 6 * 4, 32, out_offset); + unsigned width_in_tiles = ALIGN(width, 16) / 16; + unsigned height_in_tiles = ALIGN(height, 16) / 16; + unsigned pitch_minus_1 = 127; + unsigned size_needed = (width_in_tiles + height_in_tiles - 1) * 4096; + brw_get_scratch_bo(brw, &brw->wm.multisampled_null_render_target_bo, + size_needed); - if (samples > 1) { - /* On Gen6, null render targets seem to cause GPU hangs when - * multisampling. So work around this problem by rendering into dummy - * color buffer. - * - * To decrease the amount of memory needed by the workaround buffer, we - * set its pitch to 128 bytes (the width of a Y tile). This means that - * the amount of memory needed for the workaround buffer is - * (width_in_tiles + height_in_tiles - 1) tiles. - * - * Note that since the workaround buffer will be interpreted by the - * hardware as an interleaved multisampled buffer, we need to compute - * width_in_tiles and height_in_tiles by dividing the width and height - * by 16 rather than the normal Y-tile size of 32. - */ - unsigned width_in_tiles = ALIGN(width, 16) / 16; - unsigned height_in_tiles = ALIGN(height, 16) / 16; - unsigned size_needed = (width_in_tiles + height_in_tiles - 1) * 4096; - brw_get_scratch_bo(brw, &brw->wm.multisampled_null_render_target_bo, - size_needed); - bo = brw->wm.multisampled_null_render_target_bo; - surface_type = BRW_SURFACE_2D; - pitch_minus_1 = 127; - multisampling_state = brw_get_surface_num_multisamples(samples); - } - - surf[0] = (surface_type << BRW_SURFACE_TYPE_SHIFT | + surf[0] = (BRW_SURFACE_2D << BRW_SURFACE_TYPE_SHIFT | ISL_FORMAT_B8G8R8A8_UNORM << BRW_SURFACE_FORMAT_SHIFT); - if (brw->gen < 6) { - surf[0] |= (1 << BRW_SURFACE_WRITEDISABLE_R_SHIFT | - 1 << BRW_SURFACE_WRITEDISABLE_G_SHIFT | - 1 << BRW_SURFACE_WRITEDISABLE_B_SHIFT | - 1 << BRW_SURFACE_WRITEDISABLE_A_SHIFT); - } - surf[1] = !bo ? 0 : - brw_emit_reloc(&brw->batch, *out_offset + 4, bo, 0, RELOC_WRITE); + surf[1] = brw_emit_reloc(&brw->batch, *out_offset + 4, + brw->wm.multisampled_null_render_target_bo, + 0, RELOC_WRITE); surf[2] = ((width - 1) << BRW_SURFACE_WIDTH_SHIFT | (height - 1) << BRW_SURFACE_HEIGHT_SHIFT); @@ -910,7 +886,7 @@ brw_emit_null_surface_state(struct brw_context *brw, */ surf[3] = (BRW_SURFACE_TILED | BRW_SURFACE_TILED_Y | pitch_minus_1 << BRW_SURFACE_PITCH_SHIFT); - surf[4] = multisampling_state; + surf[4] = BRW_SURFACE_MULTISAMPLECOUNT_4; surf[5] = 0; } @@ -1047,14 +1023,12 @@ brw_update_renderbuffer_surfaces(struct brw_context *brw, brw->vtbl.update_renderbuffer_surface( brw, fb->_ColorDrawBuffers[i], flags, i, surf_index); } else { - brw->vtbl.emit_null_surface_state(brw, w, h, s, - &surf_offset[surf_index]); + emit_null_surface_state(brw, w, h, s, &surf_offset[surf_index]); } } } else { const uint32_t surf_index = render_target_start; - brw->vtbl.emit_null_surface_state(brw, w, h, s, - &surf_offset[surf_index]); + emit_null_surface_state(brw, w, h, s, &surf_offset[surf_index]); } } @@ -1160,9 +1134,11 @@ update_renderbuffer_read_surfaces(struct brw_context *brw) 0); } else { - brw->vtbl.emit_null_surface_state( - brw, _mesa_geometric_width(fb), _mesa_geometric_height(fb), - _mesa_geometric_samples(fb), surf_offset); + emit_null_surface_state(brw, + _mesa_geometric_width(fb), + _mesa_geometric_height(fb), + _mesa_geometric_samples(fb), + surf_offset); } } @@ -1334,7 +1310,7 @@ brw_upload_ubo_surfaces(struct brw_context *brw, struct gl_program *prog, &ctx->UniformBufferBindings[prog->sh.UniformBlocks[i]->Binding]; if (binding->BufferObject == ctx->Shared->NullBufferObj) { - brw->vtbl.emit_null_surface_state(brw, 1, 1, 1, &ubo_surf_offsets[i]); + emit_null_surface_state(brw, 1, 1, 1, &ubo_surf_offsets[i]); } else { struct intel_buffer_object *intel_bo = intel_buffer_object(binding->BufferObject); @@ -1359,7 +1335,7 @@ brw_upload_ubo_surfaces(struct brw_context *brw, struct gl_program *prog, &ctx->ShaderStorageBufferBindings[prog->sh.ShaderStorageBlocks[i]->Binding]; if (binding->BufferObject == ctx->Shared->NullBufferObj) { - brw->vtbl.emit_null_surface_state(brw, 1, 1, 1, &ssbo_surf_offsets[i]); + emit_null_surface_state(brw, 1, 1, 1, &ssbo_surf_offsets[i]); } else { struct intel_buffer_object *intel_bo = intel_buffer_object(binding->BufferObject); @@ -1655,7 +1631,7 @@ update_image_surface(struct brw_context *brw, } } else { - brw->vtbl.emit_null_surface_state(brw, 1, 1, 1, surf_offset); + emit_null_surface_state(brw, 1, 1, 1, surf_offset); update_default_image_param(brw, u, surface_idx, param); } } @@ -1718,7 +1694,6 @@ void gen4_init_vtable_surface_functions(struct brw_context *brw) { brw->vtbl.update_renderbuffer_surface = gen4_update_renderbuffer_surface; - brw->vtbl.emit_null_surface_state = brw_emit_null_surface_state; } void diff --git a/src/mesa/drivers/dri/i965/gen7_wm_surface_state.c b/src/mesa/drivers/dri/i965/gen7_wm_surface_state.c deleted file mode 100644 index c9d777bcc57..00000000000 --- a/src/mesa/drivers/dri/i965/gen7_wm_surface_state.c +++ /dev/null @@ -1,180 +0,0 @@ -/* - * Copyright © 2011 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS - * IN THE SOFTWARE. - */ -#include "main/mtypes.h" -#include "main/blend.h" -#include "main/samplerobj.h" -#include "main/texformat.h" -#include "main/teximage.h" -#include "program/prog_parameter.h" -#include "program/prog_instruction.h" - -#include "intel_mipmap_tree.h" -#include "intel_batchbuffer.h" -#include "intel_tex.h" -#include "intel_fbo.h" -#include "intel_buffer_objects.h" - -#include "brw_context.h" -#include "brw_state.h" -#include "brw_defines.h" -#include "brw_wm.h" - -void -gen7_check_surface_setup(uint32_t *surf, bool is_render_target) -{ - unsigned num_multisamples = surf[4] & INTEL_MASK(5, 3); - unsigned multisampled_surface_storage_format = surf[4] & (1 << 6); - unsigned surface_array_spacing = surf[0] & (1 << 10); - bool is_multisampled = num_multisamples != GEN7_SURFACE_MULTISAMPLECOUNT_1; - - (void) surface_array_spacing; - - /* From the Ivybridge PRM, Volume 4 Part 1, page 66 (RENDER_SURFACE_STATE - * dword 0 bit 10 "Surface Array Spacing" Programming Notes): - * - * If Multisampled Surface Storage Format is MSFMT_MSS and Number of - * Multisamples is not MULTISAMPLECOUNT_1, this field must be set to - * ARYSPC_LOD0. - */ - if (multisampled_surface_storage_format == GEN7_SURFACE_MSFMT_MSS - && is_multisampled) - assert(surface_array_spacing == GEN7_SURFACE_ARYSPC_LOD0); - - /* From the Ivybridge PRM, Volume 4 Part 1, page 72 (RENDER_SURFACE_STATE - * dword 4 bit 6 "Multisampled Surface Storage" Programming Notes): - * - * All multisampled render target surfaces must have this field set to - * MSFMT_MSS. - * - * But also: - * - * This field is ignored if Number of Multisamples is MULTISAMPLECOUNT_1. - */ - if (is_render_target && is_multisampled) { - assert(multisampled_surface_storage_format == GEN7_SURFACE_MSFMT_MSS); - } - - /* From the Ivybridge PRM, Volume 4 Part 1, page 72 (RENDER_SURFACE_STATE - * dword 4 bit 6 "Multisampled Surface Storage Format" Errata): - * - * If the surface’s Number of Multisamples is MULTISAMPLECOUNT_8, Width - * is >= 8192 (meaning the actual surface width is >= 8193 pixels), this - * field must be set to MSFMT_MSS. - */ - uint32_t width = GET_FIELD(surf[2], GEN7_SURFACE_WIDTH) + 1; - if (num_multisamples == GEN7_SURFACE_MULTISAMPLECOUNT_8 && width >= 8193) { - assert(multisampled_surface_storage_format == GEN7_SURFACE_MSFMT_MSS); - } - - /* From the Ivybridge PRM, Volume 4 Part 1, page 72 (RENDER_SURFACE_STATE - * dword 4 bit 6 "Multisampled Surface Storage Format" Errata): - * - * If the surface’s Number of Multisamples is MULTISAMPLECOUNT_8, - * ((Depth+1) * (Height+1)) is > 4,194,304, OR if the surface’s Number of - * Multisamples is MULTISAMPLECOUNT_4, ((Depth+1) * (Height+1)) is > - * 8,388,608, this field must be set to MSFMT_DEPTH_STENCIL.This field - * must be set to MSFMT_DEPTH_STENCIL if Surface Format is one of the - * following: I24X8_UNORM, L24X8_UNORM, A24X8_UNORM, or - * R24_UNORM_X8_TYPELESS. - * - * But also (from the Programming Notes): - * - * This field is ignored if Number of Multisamples is MULTISAMPLECOUNT_1. - */ - uint32_t depth = GET_FIELD(surf[3], BRW_SURFACE_DEPTH) + 1; - uint32_t height = GET_FIELD(surf[2], GEN7_SURFACE_HEIGHT) + 1; - if (num_multisamples == GEN7_SURFACE_MULTISAMPLECOUNT_8 && - depth * height > 4194304) { - assert(multisampled_surface_storage_format == - GEN7_SURFACE_MSFMT_DEPTH_STENCIL); - } - if (num_multisamples == GEN7_SURFACE_MULTISAMPLECOUNT_4 && - depth * height > 8388608) { - assert(multisampled_surface_storage_format == - GEN7_SURFACE_MSFMT_DEPTH_STENCIL); - } - if (is_multisampled) { - switch (GET_FIELD(surf[0], BRW_SURFACE_FORMAT)) { - case ISL_FORMAT_I24X8_UNORM: - case ISL_FORMAT_L24X8_UNORM: - case ISL_FORMAT_A24X8_UNORM: - case ISL_FORMAT_R24_UNORM_X8_TYPELESS: - assert(multisampled_surface_storage_format == - GEN7_SURFACE_MSFMT_DEPTH_STENCIL); - } - } -} - -/** - * Creates a null surface. - * - * This is used when the shader doesn't write to any color output. An FB - * write to target 0 will still be emitted, because that's how the thread is - * terminated (and computed depth is returned), so we need to have the - * hardware discard the target 0 color output.. - */ -static void -gen7_emit_null_surface_state(struct brw_context *brw, - unsigned width, - unsigned height, - unsigned samples, - uint32_t *out_offset) -{ - /* From the Ivy bridge PRM, Vol4 Part1 p62 (Surface Type: Programming - * Notes): - * - * A null surface is used in instances where an actual surface is not - * bound. When a write message is generated to a null surface, no - * actual surface is written to. When a read message (including any - * sampling engine message) is generated to a null surface, the result - * is all zeros. Note that a null surface type is allowed to be used - * with all messages, even if it is not specificially indicated as - * supported. All of the remaining fields in surface state are ignored - * for null surfaces, with the following exceptions: Width, Height, - * Depth, LOD, and Render Target View Extent fields must match the - * depth buffer’s corresponding state for all render target surfaces, - * including null. - */ - uint32_t *surf = brw_state_batch(brw, 8 * 4, 32, out_offset); - memset(surf, 0, 8 * 4); - - /* From the Ivybridge PRM, Volume 4, Part 1, page 65, - * Tiled Surface: Programming Notes: - * "If Surface Type is SURFTYPE_NULL, this field must be TRUE." - */ - surf[0] = BRW_SURFACE_NULL << BRW_SURFACE_TYPE_SHIFT | - ISL_FORMAT_B8G8R8A8_UNORM << BRW_SURFACE_FORMAT_SHIFT | - GEN7_SURFACE_TILING_Y; - - surf[2] = SET_FIELD(width - 1, GEN7_SURFACE_WIDTH) | - SET_FIELD(height - 1, GEN7_SURFACE_HEIGHT); - - gen7_check_surface_setup(surf, true /* is_render_target */); -} - -void -gen7_init_vtable_surface_functions(struct brw_context *brw) -{ - brw->vtbl.update_renderbuffer_surface = brw_update_renderbuffer_surface; - brw->vtbl.emit_null_surface_state = gen7_emit_null_surface_state; -} diff --git a/src/mesa/drivers/dri/i965/gen8_surface_state.c b/src/mesa/drivers/dri/i965/gen8_surface_state.c deleted file mode 100644 index c2ac7c74a61..00000000000 --- a/src/mesa/drivers/dri/i965/gen8_surface_state.c +++ /dev/null @@ -1,82 +0,0 @@ -/* - * Copyright © 2012 Intel Corporation - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS - * IN THE SOFTWARE. - */ - -#include "main/blend.h" -#include "main/mtypes.h" -#include "main/samplerobj.h" -#include "main/texformat.h" -#include "main/teximage.h" -#include "program/prog_parameter.h" -#include "program/prog_instruction.h" - -#include "intel_mipmap_tree.h" -#include "intel_batchbuffer.h" -#include "intel_tex.h" -#include "intel_fbo.h" -#include "intel_buffer_objects.h" -#include "intel_image.h" - -#include "brw_context.h" -#include "brw_state.h" -#include "brw_defines.h" -#include "brw_wm.h" -#include "isl/isl.h" - -static uint32_t * -gen8_allocate_surface_state(struct brw_context *brw, uint32_t *out_offset) -{ - uint32_t *surf = brw_state_batch(brw, 64, 64, out_offset); - memset(surf, 0, 64); - return surf; -} - -/** - * Creates a null surface. - * - * This is used when the shader doesn't write to any color output. An FB - * write to target 0 will still be emitted, because that's how the thread is - * terminated (and computed depth is returned), so we need to have the - * hardware discard the target 0 color output.. - */ -static void -gen8_emit_null_surface_state(struct brw_context *brw, - unsigned width, - unsigned height, - unsigned samples, - uint32_t *out_offset) -{ - uint32_t *surf = gen8_allocate_surface_state(brw, out_offset); - - surf[0] = BRW_SURFACE_NULL << BRW_SURFACE_TYPE_SHIFT | - ISL_FORMAT_B8G8R8A8_UNORM << BRW_SURFACE_FORMAT_SHIFT | - GEN8_SURFACE_TILING_Y; - surf[2] = SET_FIELD(width - 1, GEN7_SURFACE_WIDTH) | - SET_FIELD(height - 1, GEN7_SURFACE_HEIGHT); -} - -void -gen8_init_vtable_surface_functions(struct brw_context *brw) -{ - brw->vtbl.update_renderbuffer_surface = brw_update_renderbuffer_surface; - brw->vtbl.emit_null_surface_state = gen8_emit_null_surface_state; -}