/*
- Copyright (C) Intel Corp. 2006. All Rights Reserved.
- Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
- develop this 3D driver.
-
- 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 COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS 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.
-
- **********************************************************************/
- /*
- * Authors:
- * Keith Whitwell <keith@tungstengraphics.com>
- */
-
-#include "main/glheader.h"
-#include "main/macros.h"
-#include "main/enums.h"
-
-#include "intel_batchbuffer.h"
-
-#include "brw_defines.h"
+ * Copyright © 2013 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.
+ */
+
+/**
+ * \file brw_vec4_gs.c
+ *
+ * State atom for client-programmable geometry shaders, and support code.
+ */
+
+#include "brw_gs.h"
#include "brw_context.h"
-#include "brw_eu.h"
-#include "brw_util.h"
+#include "brw_vec4_gs_visitor.h"
#include "brw_state.h"
-#include "brw_gs.h"
+#include "brw_ff_gs.h"
-#include "glsl/ralloc.h"
-static void compile_gs_prog( struct brw_context *brw,
- struct brw_gs_prog_key *key )
+static bool
+do_gs_prog(struct brw_context *brw,
+ struct gl_shader_program *prog,
+ struct brw_geometry_program *gp,
+ struct brw_gs_prog_key *key)
{
- struct intel_context *intel = &brw->intel;
+ struct brw_stage_state *stage_state = &brw->gs.base;
struct brw_gs_compile c;
- const GLuint *program;
- void *mem_ctx;
- GLuint program_size;
-
memset(&c, 0, sizeof(c));
-
c.key = *key;
- brw_compute_vue_map(&c.vue_map, intel, brw->vs.prog_data);
- c.nr_regs = (c.vue_map.num_slots + 1)/2;
+ c.gp = gp;
+
+ c.prog_data.include_primitive_id =
+ (gp->program.Base.InputsRead & VARYING_BIT_PRIMITIVE_ID) != 0;
+
+ c.prog_data.invocations = gp->program.Invocations;
- mem_ctx = NULL;
-
- /* Begin the compilation:
+ /* Allocate the references to the uniforms that will end up in the
+ * prog_data associated with the compiled program, and which will be freed
+ * by the state cache.
+ *
+ * Note: param_count needs to be num_uniform_components * 4, since we add
+ * padding around uniform values below vec4 size, so the worst case is that
+ * every uniform is a float which gets padded to the size of a vec4.
*/
- brw_init_compile(brw, &c.func, mem_ctx);
+ struct gl_shader *gs = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
+ int param_count = gs->num_uniform_components * 4;
- c.func.single_program_flow = 1;
+ /* We also upload clip plane data as uniforms */
+ param_count += MAX_CLIP_PLANES * 4;
- /* For some reason the thread is spawned with only 4 channels
- * unmasked.
+ c.prog_data.base.base.param =
+ rzalloc_array(NULL, const gl_constant_value *, param_count);
+ c.prog_data.base.base.pull_param =
+ rzalloc_array(NULL, const gl_constant_value *, param_count);
+ /* Setting nr_params here NOT to the size of the param and pull_param
+ * arrays, but to the number of uniform components vec4_visitor
+ * needs. vec4_visitor::setup_uniforms() will set it back to a proper value.
*/
- brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
+ c.prog_data.base.base.nr_params = ALIGN(param_count, 4) / 4 + gs->num_samplers;
- if (intel->gen >= 6) {
- unsigned num_verts;
- bool check_edge_flag;
- /* On Sandybridge, we use the GS for implementing transform feedback
- * (called "Stream Out" in the PRM).
- */
- switch (key->primitive) {
- case _3DPRIM_POINTLIST:
- num_verts = 1;
- check_edge_flag = false;
- break;
- case _3DPRIM_LINELIST:
- case _3DPRIM_LINESTRIP:
- case _3DPRIM_LINELOOP:
- num_verts = 2;
- check_edge_flag = false;
- break;
- case _3DPRIM_TRILIST:
- case _3DPRIM_TRIFAN:
- case _3DPRIM_TRISTRIP:
- case _3DPRIM_RECTLIST:
- num_verts = 3;
- check_edge_flag = false;
- break;
- case _3DPRIM_QUADLIST:
- case _3DPRIM_QUADSTRIP:
- case _3DPRIM_POLYGON:
- num_verts = 3;
- check_edge_flag = true;
- break;
- default:
- assert(!"Unexpected primitive type in Gen6 SOL program.");
- return;
+ if (brw->gen >= 7) {
+ if (gp->program.OutputType == GL_POINTS) {
+ /* When the output type is points, the geometry shader may output data
+ * to multiple streams, and EndPrimitive() has no effect. So we
+ * configure the hardware to interpret the control data as stream ID.
+ */
+ c.prog_data.control_data_format = GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_SID;
+
+ /* We only have to emit control bits if we are using streams */
+ if (prog->Geom.UsesStreams)
+ c.control_data_bits_per_vertex = 2;
+ else
+ c.control_data_bits_per_vertex = 0;
+ } else {
+ /* When the output type is triangle_strip or line_strip, EndPrimitive()
+ * may be used to terminate the current strip and start a new one
+ * (similar to primitive restart), and outputting data to multiple
+ * streams is not supported. So we configure the hardware to interpret
+ * the control data as EndPrimitive information (a.k.a. "cut bits").
+ */
+ c.prog_data.control_data_format = GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_CUT;
+
+ /* We only need to output control data if the shader actually calls
+ * EndPrimitive().
+ */
+ c.control_data_bits_per_vertex = gp->program.UsesEndPrimitive ? 1 : 0;
}
- gen6_sol_program(&c, key, num_verts, check_edge_flag);
} else {
- /* On Gen4-5, we use the GS to decompose certain types of primitives.
- * Note that primitives which don't require a GS program have already
- * been weeded out by now.
- */
- switch (key->primitive) {
- case _3DPRIM_QUADLIST:
- brw_gs_quads( &c, key );
- break;
- case _3DPRIM_QUADSTRIP:
- brw_gs_quad_strip( &c, key );
- break;
- case _3DPRIM_LINELOOP:
- brw_gs_lines( &c );
- break;
- default:
- ralloc_free(mem_ctx);
- return;
- }
+ /* There are no control data bits in gen6. */
+ c.control_data_bits_per_vertex = 0;
+
+ /* If it is using transform feedback, enable it */
+ if (prog->TransformFeedback.NumVarying)
+ c.prog_data.gen6_xfb_enabled = true;
+ else
+ c.prog_data.gen6_xfb_enabled = false;
}
+ c.control_data_header_size_bits =
+ gp->program.VerticesOut * c.control_data_bits_per_vertex;
+
+ /* 1 HWORD = 32 bytes = 256 bits */
+ c.prog_data.control_data_header_size_hwords =
+ ALIGN(c.control_data_header_size_bits, 256) / 256;
+
+ GLbitfield64 outputs_written = gp->program.Base.OutputsWritten;
- /* get the program
+ /* In order for legacy clipping to work, we need to populate the clip
+ * distance varying slots whenever clipping is enabled, even if the vertex
+ * shader doesn't write to gl_ClipDistance.
*/
- program = brw_get_program(&c.func, &program_size);
-
- if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
- int i;
-
- printf("gs:\n");
- for (i = 0; i < program_size / sizeof(struct brw_instruction); i++)
- brw_disasm(stdout, &((struct brw_instruction *)program)[i],
- intel->gen);
- printf("\n");
- }
-
- brw_upload_cache(&brw->cache, BRW_GS_PROG,
- &c.key, sizeof(c.key),
- program, program_size,
- &c.prog_data, sizeof(c.prog_data),
- &brw->gs.prog_offset, &brw->gs.prog_data);
- ralloc_free(mem_ctx);
-}
+ if (c.key.base.userclip_active) {
+ outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
+ outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
+ }
-static void populate_key( struct brw_context *brw,
- struct brw_gs_prog_key *key )
-{
- static const unsigned swizzle_for_offset[4] = {
- BRW_SWIZZLE4(0, 1, 2, 3),
- BRW_SWIZZLE4(1, 2, 3, 3),
- BRW_SWIZZLE4(2, 3, 3, 3),
- BRW_SWIZZLE4(3, 3, 3, 3)
- };
+ brw_compute_vue_map(brw, &c.prog_data.base.vue_map, outputs_written);
- struct gl_context *ctx = &brw->intel.ctx;
- struct intel_context *intel = &brw->intel;
+ /* Compute the output vertex size.
+ *
+ * From the Ivy Bridge PRM, Vol2 Part1 7.2.1.1 STATE_GS - Output Vertex
+ * Size (p168):
+ *
+ * [0,62] indicating [1,63] 16B units
+ *
+ * Specifies the size of each vertex stored in the GS output entry
+ * (following any Control Header data) as a number of 128-bit units
+ * (minus one).
+ *
+ * Programming Restrictions: The vertex size must be programmed as a
+ * multiple of 32B units with the following exception: Rendering is
+ * disabled (as per SOL stage state) and the vertex size output by the
+ * GS thread is 16B.
+ *
+ * If rendering is enabled (as per SOL state) the vertex size must be
+ * programmed as a multiple of 32B units. In other words, the only time
+ * software can program a vertex size with an odd number of 16B units
+ * is when rendering is disabled.
+ *
+ * Note: B=bytes in the above text.
+ *
+ * It doesn't seem worth the extra trouble to optimize the case where the
+ * vertex size is 16B (especially since this would require special-casing
+ * the GEN assembly that writes to the URB). So we just set the vertex
+ * size to a multiple of 32B (2 vec4's) in all cases.
+ *
+ * The maximum output vertex size is 62*16 = 992 bytes (31 hwords). We
+ * budget that as follows:
+ *
+ * 512 bytes for varyings (a varying component is 4 bytes and
+ * gl_MaxGeometryOutputComponents = 128)
+ * 16 bytes overhead for VARYING_SLOT_PSIZ (each varying slot is 16
+ * bytes)
+ * 16 bytes overhead for gl_Position (we allocate it a slot in the VUE
+ * even if it's not used)
+ * 32 bytes overhead for gl_ClipDistance (we allocate it 2 VUE slots
+ * whenever clip planes are enabled, even if the shader doesn't
+ * write to gl_ClipDistance)
+ * 16 bytes overhead since the VUE size must be a multiple of 32 bytes
+ * (see below)--this causes up to 1 VUE slot to be wasted
+ * 400 bytes available for varying packing overhead
+ *
+ * Worst-case varying packing overhead is 3/4 of a varying slot (12 bytes)
+ * per interpolation type, so this is plenty.
+ *
+ */
+ unsigned output_vertex_size_bytes = c.prog_data.base.vue_map.num_slots * 16;
+ assert(brw->gen == 6 ||
+ output_vertex_size_bytes <= GEN7_MAX_GS_OUTPUT_VERTEX_SIZE_BYTES);
+ c.prog_data.output_vertex_size_hwords =
+ ALIGN(output_vertex_size_bytes, 32) / 32;
- memset(key, 0, sizeof(*key));
+ /* Compute URB entry size. The maximum allowed URB entry size is 32k.
+ * That divides up as follows:
+ *
+ * 64 bytes for the control data header (cut indices or StreamID bits)
+ * 4096 bytes for varyings (a varying component is 4 bytes and
+ * gl_MaxGeometryTotalOutputComponents = 1024)
+ * 4096 bytes overhead for VARYING_SLOT_PSIZ (each varying slot is 16
+ * bytes/vertex and gl_MaxGeometryOutputVertices is 256)
+ * 4096 bytes overhead for gl_Position (we allocate it a slot in the VUE
+ * even if it's not used)
+ * 8192 bytes overhead for gl_ClipDistance (we allocate it 2 VUE slots
+ * whenever clip planes are enabled, even if the shader doesn't
+ * write to gl_ClipDistance)
+ * 4096 bytes overhead since the VUE size must be a multiple of 32
+ * bytes (see above)--this causes up to 1 VUE slot to be wasted
+ * 8128 bytes available for varying packing overhead
+ *
+ * Worst-case varying packing overhead is 3/4 of a varying slot per
+ * interpolation type, which works out to 3072 bytes, so this would allow
+ * us to accommodate 2 interpolation types without any danger of running
+ * out of URB space.
+ *
+ * In practice, the risk of running out of URB space is very small, since
+ * the above figures are all worst-case, and most of them scale with the
+ * number of output vertices. So we'll just calculate the amount of space
+ * we need, and if it's too large, fail to compile.
+ *
+ * The above is for gen7+ where we have a single URB entry that will hold
+ * all the output. In gen6, we will have to allocate URB entries for every
+ * vertex we emit, so our URB entries only need to be large enough to hold
+ * a single vertex. Also, gen6 does not have a control data header.
+ */
+ unsigned output_size_bytes;
+ if (brw->gen >= 7) {
+ output_size_bytes =
+ c.prog_data.output_vertex_size_hwords * 32 * gp->program.VerticesOut;
+ output_size_bytes += 32 * c.prog_data.control_data_header_size_hwords;
+ } else {
+ output_size_bytes = c.prog_data.output_vertex_size_hwords * 32;
+ }
- /* CACHE_NEW_VS_PROG (part of VUE map) */
- key->attrs = brw->vs.prog_data->outputs_written;
+ /* Broadwell stores "Vertex Count" as a full 8 DWord (32 byte) URB output,
+ * which comes before the control header.
+ */
+ if (brw->gen >= 8)
+ output_size_bytes += 32;
- /* BRW_NEW_PRIMITIVE */
- key->primitive = brw->primitive;
+ assert(output_size_bytes >= 1);
+ int max_output_size_bytes = GEN7_MAX_GS_URB_ENTRY_SIZE_BYTES;
+ if (brw->gen == 6)
+ max_output_size_bytes = GEN6_MAX_GS_URB_ENTRY_SIZE_BYTES;
+ if (output_size_bytes > max_output_size_bytes)
+ return false;
- /* _NEW_LIGHT */
- key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
- if (key->primitive == _3DPRIM_QUADLIST && ctx->Light.ShadeModel != GL_FLAT) {
- /* Provide consistent primitive order with brw_set_prim's
- * optimization of single quads to trifans.
- */
- key->pv_first = true;
+
+ /* URB entry sizes are stored as a multiple of 64 bytes in gen7+ and
+ * a multiple of 128 bytes in gen6.
+ */
+ if (brw->gen >= 7)
+ c.prog_data.base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64;
+ else
+ c.prog_data.base.urb_entry_size = ALIGN(output_size_bytes, 128) / 128;
+
+ c.prog_data.output_topology =
+ get_hw_prim_for_gl_prim(gp->program.OutputType);
+
+ brw_compute_vue_map(brw, &c.input_vue_map, c.key.input_varyings);
+
+ /* GS inputs are read from the VUE 256 bits (2 vec4's) at a time, so we
+ * need to program a URB read length of ceiling(num_slots / 2).
+ */
+ c.prog_data.base.urb_read_length = (c.input_vue_map.num_slots + 1) / 2;
+
+ void *mem_ctx = ralloc_context(NULL);
+ unsigned program_size;
+ const unsigned *program =
+ brw_gs_emit(brw, prog, &c, mem_ctx, &program_size);
+ if (program == NULL) {
+ ralloc_free(mem_ctx);
+ return false;
}
- /* CACHE_NEW_VS_PROG (part of VUE map)*/
- key->userclip_active = brw->vs.prog_data->userclip;
-
- if (intel->gen >= 7) {
- /* On Gen7 and later, we don't use GS (yet). */
- key->need_gs_prog = false;
- } else if (intel->gen == 6) {
- /* On Gen6, GS is used for transform feedback. */
- /* _NEW_TRANSFORM_FEEDBACK */
- if (ctx->TransformFeedback.CurrentObject->Active &&
- !ctx->TransformFeedback.CurrentObject->Paused) {
- const struct gl_shader_program *shaderprog =
- ctx->Shader.CurrentVertexProgram;
- const struct gl_transform_feedback_info *linked_xfb_info =
- &shaderprog->LinkedTransformFeedback;
- int i;
-
- /* Make sure that the VUE slots won't overflow the unsigned chars in
- * key->transform_feedback_bindings[].
- */
- STATIC_ASSERT(BRW_VERT_RESULT_MAX <= 256);
+ /* Scratch space is used for register spilling */
+ if (c.base.last_scratch) {
+ perf_debug("Geometry shader triggered register spilling. "
+ "Try reducing the number of live vec4 values to "
+ "improve performance.\n");
- /* Make sure that we don't need more binding table entries than we've
- * set aside for use in transform feedback. (We shouldn't, since we
- * set aside enough binding table entries to have one per component).
- */
- assert(linked_xfb_info->NumOutputs <= BRW_MAX_SOL_BINDINGS);
-
- key->need_gs_prog = true;
- key->num_transform_feedback_bindings = linked_xfb_info->NumOutputs;
- for (i = 0; i < key->num_transform_feedback_bindings; ++i) {
- key->transform_feedback_bindings[i] =
- linked_xfb_info->Outputs[i].OutputRegister;
- key->transform_feedback_swizzles[i] =
- swizzle_for_offset[linked_xfb_info->Outputs[i].ComponentOffset];
- }
- }
- /* On Gen6, GS is also used for rasterizer discard. */
- /* _NEW_RASTERIZER_DISCARD */
- if (ctx->RasterDiscard) {
- key->need_gs_prog = true;
- key->rasterizer_discard = true;
- }
- } else {
- /* Pre-gen6, GS is used to transform QUADLIST, QUADSTRIP, and LINELOOP
- * into simpler primitives.
- */
- key->need_gs_prog = (brw->primitive == _3DPRIM_QUADLIST ||
- brw->primitive == _3DPRIM_QUADSTRIP ||
- brw->primitive == _3DPRIM_LINELOOP);
+ c.prog_data.base.base.total_scratch
+ = brw_get_scratch_size(c.base.last_scratch*REG_SIZE);
+
+ brw_get_scratch_bo(brw, &stage_state->scratch_bo,
+ c.prog_data.base.base.total_scratch *
+ brw->max_gs_threads);
}
- /* For testing, the environment variable INTEL_FORCE_GS can be used to
- * force a GS program to be used, even if it's not necessary.
- */
- if (getenv("INTEL_FORCE_GS"))
- key->need_gs_prog = true;
+
+ brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
+ &c.key, sizeof(c.key),
+ program, program_size,
+ &c.prog_data, sizeof(c.prog_data),
+ &stage_state->prog_offset, &brw->gs.prog_data);
+ ralloc_free(mem_ctx);
+
+ return true;
}
-/* Calculate interpolants for triangle and line rasterization.
- */
+
static void
brw_upload_gs_prog(struct brw_context *brw)
{
+ struct gl_context *ctx = &brw->ctx;
+ struct brw_stage_state *stage_state = &brw->gs.base;
struct brw_gs_prog_key key;
- /* Populate the key:
- */
- populate_key(brw, &key);
+ /* BRW_NEW_GEOMETRY_PROGRAM */
+ struct brw_geometry_program *gp =
+ (struct brw_geometry_program *) brw->geometry_program;
- if (brw->gs.prog_active != key.need_gs_prog) {
- brw->state.dirty.cache |= CACHE_NEW_GS_PROG;
- brw->gs.prog_active = key.need_gs_prog;
- }
+ if (gp == NULL) {
+ /* No geometry shader. Vertex data just passes straight through. */
+ if (brw->state.dirty.brw & BRW_NEW_VUE_MAP_VS) {
+ brw->vue_map_geom_out = brw->vue_map_vs;
+ brw->state.dirty.brw |= BRW_NEW_VUE_MAP_GEOM_OUT;
+ }
- if (brw->gs.prog_active) {
- if (!brw_search_cache(&brw->cache, BRW_GS_PROG,
- &key, sizeof(key),
- &brw->gs.prog_offset, &brw->gs.prog_data)) {
- compile_gs_prog( brw, &key );
+ if (brw->gen == 6 &&
+ (brw->state.dirty.brw & BRW_NEW_TRANSFORM_FEEDBACK)) {
+ gen6_brw_upload_ff_gs_prog(brw);
+ return;
}
+
+ /* Other state atoms had better not try to access prog_data, since
+ * there's no GS program.
+ */
+ brw->gs.prog_data = NULL;
+ brw->gs.base.prog_data = NULL;
+
+ return;
+ }
+
+ struct gl_program *prog = &gp->program.Base;
+
+ memset(&key, 0, sizeof(key));
+
+ key.base.program_string_id = gp->id;
+ brw_setup_vec4_key_clip_info(brw, &key.base,
+ gp->program.Base.UsesClipDistanceOut);
+
+ /* _NEW_LIGHT | _NEW_BUFFERS */
+ key.base.clamp_vertex_color = ctx->Light._ClampVertexColor;
+
+ /* _NEW_TEXTURE */
+ brw_populate_sampler_prog_key_data(ctx, prog, stage_state->sampler_count,
+ &key.base.tex);
+
+ /* BRW_NEW_VUE_MAP_VS */
+ key.input_varyings = brw->vue_map_vs.slots_valid;
+
+ if (!brw_search_cache(&brw->cache, BRW_CACHE_GS_PROG,
+ &key, sizeof(key),
+ &stage_state->prog_offset, &brw->gs.prog_data)) {
+ bool success =
+ do_gs_prog(brw, ctx->_Shader->CurrentProgram[MESA_SHADER_GEOMETRY], gp,
+ &key);
+ assert(success);
+ (void)success;
+ }
+ brw->gs.base.prog_data = &brw->gs.prog_data->base.base;
+
+ if (memcmp(&brw->vs.prog_data->base.vue_map, &brw->vue_map_geom_out,
+ sizeof(brw->vue_map_geom_out)) != 0) {
+ brw->vue_map_geom_out = brw->gs.prog_data->base.vue_map;
+ brw->state.dirty.brw |= BRW_NEW_VUE_MAP_GEOM_OUT;
}
}
const struct brw_tracked_state brw_gs_prog = {
.dirty = {
- .mesa = (_NEW_LIGHT |
- _NEW_TRANSFORM_FEEDBACK |
- _NEW_RASTERIZER_DISCARD),
- .brw = BRW_NEW_PRIMITIVE,
- .cache = CACHE_NEW_VS_PROG
+ .mesa = _NEW_BUFFERS |
+ _NEW_LIGHT |
+ _NEW_TEXTURE,
+ .brw = BRW_NEW_GEOMETRY_PROGRAM |
+ BRW_NEW_TRANSFORM_FEEDBACK |
+ BRW_NEW_VUE_MAP_VS,
},
.emit = brw_upload_gs_prog
};
+
+
+bool
+brw_gs_precompile(struct gl_context *ctx,
+ struct gl_shader_program *shader_prog,
+ struct gl_program *prog)
+{
+ struct brw_context *brw = brw_context(ctx);
+ struct brw_gs_prog_key key;
+ uint32_t old_prog_offset = brw->gs.base.prog_offset;
+ struct brw_gs_prog_data *old_prog_data = brw->gs.prog_data;
+ bool success;
+
+ struct gl_geometry_program *gp = (struct gl_geometry_program *) prog;
+ struct brw_geometry_program *bgp = brw_geometry_program(gp);
+
+ memset(&key, 0, sizeof(key));
+
+ brw_vec4_setup_prog_key_for_precompile(ctx, &key.base, bgp->id, &gp->Base);
+
+ /* Assume that the set of varyings coming in from the vertex shader exactly
+ * matches what the geometry shader requires.
+ */
+ key.input_varyings = gp->Base.InputsRead;
+
+ success = do_gs_prog(brw, shader_prog, bgp, &key);
+
+ brw->gs.base.prog_offset = old_prog_offset;
+ brw->gs.prog_data = old_prog_data;
+
+ return success;
+}
+
+
+bool
+brw_gs_prog_data_compare(const void *in_a, const void *in_b)
+{
+ const struct brw_gs_prog_data *a = in_a;
+ const struct brw_gs_prog_data *b = in_b;
+
+ /* Compare the base structure. */
+ if (!brw_stage_prog_data_compare(&a->base.base, &b->base.base))
+ return false;
+
+ /* Compare the rest of the struct. */
+ const unsigned offset = sizeof(struct brw_stage_prog_data);
+ if (memcmp(((char *) a) + offset, ((char *) b) + offset,
+ sizeof(struct brw_gs_prog_data) - offset)) {
+ return false;
+ }
+
+ return true;
+}