*
**************************************************************************/
+#include <sys/errno.h>
#include "main/glheader.h"
#include "main/context.h"
#include "main/samplerobj.h"
#include "main/state.h"
#include "main/enums.h"
+#include "main/macros.h"
+#include "main/transformfeedback.h"
#include "tnl/tnl.h"
#include "vbo/vbo_context.h"
#include "swrast/swrast.h"
#include "swrast_setup/swrast_setup.h"
+#include "drivers/common/meta.h"
+#include "brw_blorp.h"
#include "brw_draw.h"
#include "brw_defines.h"
#include "brw_context.h"
#include "brw_state.h"
#include "intel_batchbuffer.h"
+#include "intel_fbo.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
#define FILE_DEBUG_FLAG DEBUG_PRIMS
-static GLuint prim_to_hw_prim[GL_POLYGON+1] = {
+const GLuint prim_to_hw_prim[GL_POLYGON+1] = {
_3DPRIM_POINTLIST,
_3DPRIM_LINELIST,
_3DPRIM_LINELOOP,
* programs be immune to the active primitive (ie. cope with all
* possibilities). That may not be realistic however.
*/
-static GLuint brw_set_prim(struct brw_context *brw,
- const struct _mesa_prim *prim)
+static void brw_set_prim(struct brw_context *brw,
+ const struct _mesa_prim *prim)
{
- struct gl_context *ctx = &brw->intel.ctx;
- GLenum mode = prim->mode;
+ struct gl_context *ctx = &brw->ctx;
+ uint32_t hw_prim = prim_to_hw_prim[prim->mode];
DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
/* Slight optimization to avoid the GS program when not needed:
*/
- if (mode == GL_QUAD_STRIP &&
+ if (prim->mode == GL_QUAD_STRIP &&
ctx->Light.ShadeModel != GL_FLAT &&
ctx->Polygon.FrontMode == GL_FILL &&
ctx->Polygon.BackMode == GL_FILL)
- mode = GL_TRIANGLE_STRIP;
+ hw_prim = _3DPRIM_TRISTRIP;
if (prim->mode == GL_QUADS && prim->count == 4 &&
ctx->Light.ShadeModel != GL_FLAT &&
ctx->Polygon.FrontMode == GL_FILL &&
ctx->Polygon.BackMode == GL_FILL) {
- mode = GL_TRIANGLE_FAN;
+ hw_prim = _3DPRIM_TRIFAN;
}
- if (mode != brw->primitive) {
- brw->primitive = mode;
+ if (hw_prim != brw->primitive) {
+ brw->primitive = hw_prim;
brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
- if (reduced_prim[mode] != brw->intel.reduced_primitive) {
- brw->intel.reduced_primitive = reduced_prim[mode];
+ if (reduced_prim[prim->mode] != brw->reduced_primitive) {
+ brw->reduced_primitive = reduced_prim[prim->mode];
brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
}
}
+}
+
+static void gen6_set_prim(struct brw_context *brw,
+ const struct _mesa_prim *prim)
+{
+ uint32_t hw_prim;
- return prim_to_hw_prim[mode];
+ DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
+
+ hw_prim = prim_to_hw_prim[prim->mode];
+
+ if (hw_prim != brw->primitive) {
+ brw->primitive = hw_prim;
+ brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
+ }
}
+/**
+ * The hardware is capable of removing dangling vertices on its own; however,
+ * prior to Gen6, we sometimes convert quads into trifans (and quad strips
+ * into tristrips), since pre-Gen6 hardware requires a GS to render quads.
+ * This function manually trims dangling vertices from a draw call involving
+ * quads so that those dangling vertices won't get drawn when we convert to
+ * trifans/tristrips.
+ */
static GLuint trim(GLenum prim, GLuint length)
{
if (prim == GL_QUAD_STRIP)
const struct _mesa_prim *prim,
uint32_t hw_prim)
{
- struct intel_context *intel = &brw->intel;
int verts_per_instance;
int vertex_access_type;
int start_vertex_location;
start_vertex_location += brw->vb.start_vertex_bias;
}
- verts_per_instance = trim(prim->mode, prim->count);
+ /* We only need to trim the primitive count on pre-Gen6. */
+ if (brw->gen < 6)
+ verts_per_instance = trim(prim->mode, prim->count);
+ else
+ verts_per_instance = prim->count;
/* If nothing to emit, just return. */
if (verts_per_instance == 0)
* and missed flushes of the render cache as it heads to other parts of
* the besides the draw code.
*/
- if (intel->always_flush_cache) {
- intel_batchbuffer_emit_mi_flush(intel);
+ if (brw->always_flush_cache) {
+ intel_batchbuffer_emit_mi_flush(brw);
}
BEGIN_BATCH(6);
vertex_access_type);
OUT_BATCH(verts_per_instance);
OUT_BATCH(start_vertex_location);
- OUT_BATCH(1); // instance count
- OUT_BATCH(0); // start instance location
+ OUT_BATCH(prim->num_instances);
+ OUT_BATCH(prim->base_instance);
OUT_BATCH(base_vertex_location);
ADVANCE_BATCH();
- intel->batch.need_workaround_flush = true;
+ brw->batch.need_workaround_flush = true;
- if (intel->always_flush_cache) {
- intel_batchbuffer_emit_mi_flush(intel);
+ if (brw->always_flush_cache) {
+ intel_batchbuffer_emit_mi_flush(brw);
}
}
const struct _mesa_prim *prim,
uint32_t hw_prim)
{
- struct intel_context *intel = &brw->intel;
int verts_per_instance;
int vertex_access_type;
int start_vertex_location;
start_vertex_location += brw->vb.start_vertex_bias;
}
- verts_per_instance = trim(prim->mode, prim->count);
+ verts_per_instance = prim->count;
/* If nothing to emit, just return. */
if (verts_per_instance == 0)
* and missed flushes of the render cache as it heads to other parts of
* the besides the draw code.
*/
- if (intel->always_flush_cache) {
- intel_batchbuffer_emit_mi_flush(intel);
+ if (brw->always_flush_cache) {
+ intel_batchbuffer_emit_mi_flush(brw);
}
BEGIN_BATCH(7);
OUT_BATCH(hw_prim | vertex_access_type);
OUT_BATCH(verts_per_instance);
OUT_BATCH(start_vertex_location);
- OUT_BATCH(1); // instance count
- OUT_BATCH(0); // start instance location
+ OUT_BATCH(prim->num_instances);
+ OUT_BATCH(prim->base_instance);
OUT_BATCH(base_vertex_location);
ADVANCE_BATCH();
- if (intel->always_flush_cache) {
- intel_batchbuffer_emit_mi_flush(intel);
+ if (brw->always_flush_cache) {
+ intel_batchbuffer_emit_mi_flush(brw);
}
}
static void brw_merge_inputs( struct brw_context *brw,
const struct gl_client_array *arrays[])
{
- struct brw_vertex_info old = brw->vb.info;
GLuint i;
for (i = 0; i < brw->vb.nr_buffers; i++) {
}
brw->vb.nr_buffers = 0;
- memset(&brw->vb.info, 0, sizeof(brw->vb.info));
-
for (i = 0; i < VERT_ATTRIB_MAX; i++) {
brw->vb.inputs[i].buffer = -1;
brw->vb.inputs[i].glarray = arrays[i];
brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
+ }
+}
+
+/*
+ * \brief Resolve buffers before drawing.
+ *
+ * Resolve the depth buffer's HiZ buffer and resolve the depth buffer of each
+ * enabled depth texture.
+ *
+ * (In the future, this will also perform MSAA resolves).
+ */
+static void
+brw_predraw_resolve_buffers(struct brw_context *brw)
+{
+ struct gl_context *ctx = &brw->ctx;
+ struct intel_renderbuffer *depth_irb;
+ struct intel_texture_object *tex_obj;
- if (arrays[i]->StrideB != 0)
- brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
- ((i%16) * 2);
+ /* Resolve the depth buffer's HiZ buffer. */
+ depth_irb = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH);
+ if (depth_irb)
+ intel_renderbuffer_resolve_hiz(brw, depth_irb);
+
+ /* Resolve depth buffer of each enabled depth texture, and color buffer of
+ * each fast-clear-enabled color texture.
+ */
+ for (int i = 0; i < BRW_MAX_TEX_UNIT; i++) {
+ if (!ctx->Texture.Unit[i]._ReallyEnabled)
+ continue;
+ tex_obj = intel_texture_object(ctx->Texture.Unit[i]._Current);
+ if (!tex_obj || !tex_obj->mt)
+ continue;
+ intel_miptree_all_slices_resolve_depth(brw, tex_obj->mt);
+ intel_miptree_resolve_color(brw, tex_obj->mt);
}
+}
- /* Raise statechanges if input sizes have changed. */
- if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
- brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
+/**
+ * \brief Call this after drawing to mark which buffers need resolving
+ *
+ * If the depth buffer was written to and if it has an accompanying HiZ
+ * buffer, then mark that it needs a depth resolve.
+ *
+ * If the color buffer is a multisample window system buffer, then
+ * mark that it needs a downsample.
+ */
+static void brw_postdraw_set_buffers_need_resolve(struct brw_context *brw)
+{
+ struct gl_context *ctx = &brw->ctx;
+ struct gl_framebuffer *fb = ctx->DrawBuffer;
+
+ struct intel_renderbuffer *front_irb = NULL;
+ struct intel_renderbuffer *back_irb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
+ struct intel_renderbuffer *depth_irb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
+ struct gl_renderbuffer_attachment *depth_att = &fb->Attachment[BUFFER_DEPTH];
+
+ if (brw->is_front_buffer_rendering)
+ front_irb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
+
+ if (front_irb)
+ intel_renderbuffer_set_needs_downsample(front_irb);
+ if (back_irb)
+ intel_renderbuffer_set_needs_downsample(back_irb);
+ if (depth_irb && ctx->Depth.Mask)
+ intel_renderbuffer_att_set_needs_depth_resolve(depth_att);
}
/* May fail if out of video memory for texture or vbo upload, or on
* fallback conditions.
*/
-static GLboolean brw_try_draw_prims( struct gl_context *ctx,
+static bool brw_try_draw_prims( struct gl_context *ctx,
const struct gl_client_array *arrays[],
const struct _mesa_prim *prim,
GLuint nr_prims,
GLuint min_index,
GLuint max_index )
{
- struct intel_context *intel = intel_context(ctx);
struct brw_context *brw = brw_context(ctx);
- GLboolean retval = GL_FALSE;
- GLboolean warn = GL_FALSE;
+ bool retval = true;
GLuint i;
+ bool fail_next = false;
if (ctx->NewState)
_mesa_update_state( ctx );
+ /* Find the highest sampler unit used by each shader program. A bit-count
+ * won't work since ARB programs use the texture unit number as the sampler
+ * index.
+ */
+ brw->wm.sampler_count = _mesa_fls(ctx->FragmentProgram._Current->Base.SamplersUsed);
+ brw->vs.sampler_count = _mesa_fls(ctx->VertexProgram._Current->Base.SamplersUsed);
+
/* We have to validate the textures *before* checking for fallbacks;
* otherwise, the software fallback won't be able to rely on the
* texture state, the firstLevel and lastLevel fields won't be
*/
brw_validate_textures( brw );
+ intel_prepare_render(brw);
+
+ /* This workaround has to happen outside of brw_upload_state() because it
+ * may flush the batchbuffer for a blit, affecting the state flags.
+ */
+ brw_workaround_depthstencil_alignment(brw, 0);
+
+ /* Resolves must occur after updating renderbuffers, updating context state,
+ * and finalizing textures but before setting up any hardware state for
+ * this draw call.
+ */
+ brw_predraw_resolve_buffers(brw);
+
/* Bind all inputs, derive varying and size information:
*/
brw_merge_inputs( brw, arrays );
brw->vb.max_index = max_index;
brw->state.dirty.brw |= BRW_NEW_VERTICES;
- /* Have to validate state quite late. Will rebuild tnl_program,
- * which depends on varying information.
- *
- * Note this is where brw->vs->prog_data.inputs_read is calculated,
- * so can't access it earlier.
- */
-
- intel_prepare_render(intel);
-
for (i = 0; i < nr_prims; i++) {
- uint32_t hw_prim;
int estimated_max_prim_size;
estimated_max_prim_size = 512; /* batchbuffer commands */
* we've got validated state that needs to be in the same batch as the
* primitives.
*/
- intel_batchbuffer_require_space(intel, estimated_max_prim_size, false);
+ intel_batchbuffer_require_space(brw, estimated_max_prim_size, false);
+ intel_batchbuffer_save_state(brw);
- hw_prim = brw_set_prim(brw, &prim[i]);
- if (brw->state.dirty.brw) {
- brw_validate_state(brw);
-
- /* Various fallback checks: */
- if (brw->intel.Fallback)
- goto out;
-
- /* Check that we can fit our state in with our existing batchbuffer, or
- * flush otherwise.
- */
- if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
- brw->state.validated_bo_count)) {
- static GLboolean warned;
- intel_batchbuffer_flush(intel);
-
- /* Validate the state after we flushed the batch (which would have
- * changed the set of dirty state). If we still fail to
- * check_aperture, warn of what's happening, but attempt to continue
- * on since it may succeed anyway, and the user would probably rather
- * see a failure and a warning than a fallback.
- */
- brw_validate_state(brw);
- if (!warned &&
- dri_bufmgr_check_aperture_space(brw->state.validated_bos,
- brw->state.validated_bo_count)) {
- warn = GL_TRUE;
- warned = GL_TRUE;
- }
- }
+ if (brw->num_instances != prim->num_instances) {
+ brw->num_instances = prim->num_instances;
+ brw->state.dirty.brw |= BRW_NEW_VERTICES;
+ }
+ if (brw->basevertex != prim->basevertex) {
+ brw->basevertex = prim->basevertex;
+ brw->state.dirty.brw |= BRW_NEW_VERTICES;
+ }
+ if (brw->gen < 6)
+ brw_set_prim(brw, &prim[i]);
+ else
+ gen6_set_prim(brw, &prim[i]);
- intel->no_batch_wrap = GL_TRUE;
+retry:
+ /* Note that before the loop, brw->state.dirty.brw was set to != 0, and
+ * that the state updated in the loop outside of this block is that in
+ * *_set_prim or intel_batchbuffer_flush(), which only impacts
+ * brw->state.dirty.brw.
+ */
+ if (brw->state.dirty.brw) {
+ brw->no_batch_wrap = true;
brw_upload_state(brw);
}
- if (intel->gen >= 7)
- gen7_emit_prim(brw, &prim[i], hw_prim);
+ if (brw->gen >= 7)
+ gen7_emit_prim(brw, &prim[i], brw->primitive);
else
- brw_emit_prim(brw, &prim[i], hw_prim);
-
- intel->no_batch_wrap = GL_FALSE;
-
- retval = GL_TRUE;
+ brw_emit_prim(brw, &prim[i], brw->primitive);
+
+ brw->no_batch_wrap = false;
+
+ if (dri_bufmgr_check_aperture_space(&brw->batch.bo, 1)) {
+ if (!fail_next) {
+ intel_batchbuffer_reset_to_saved(brw);
+ intel_batchbuffer_flush(brw);
+ fail_next = true;
+ goto retry;
+ } else {
+ if (intel_batchbuffer_flush(brw) == -ENOSPC) {
+ static bool warned = false;
+
+ if (!warned) {
+ fprintf(stderr, "i965: Single primitive emit exceeded"
+ "available aperture space\n");
+ warned = true;
+ }
+
+ retval = false;
+ }
+ }
+ }
}
- if (intel->always_flush_batch)
- intel_batchbuffer_flush(intel);
- out:
+ if (brw->always_flush_batch)
+ intel_batchbuffer_flush(brw);
brw_state_cache_check_size(brw);
-
- if (warn)
- fprintf(stderr, "i965: Single primitive emit potentially exceeded "
- "available aperture space\n");
-
- if (!retval)
- DBG("%s failed\n", __FUNCTION__);
+ brw_postdraw_set_buffers_need_resolve(brw);
return retval;
}
void brw_draw_prims( struct gl_context *ctx,
- const struct gl_client_array *arrays[],
const struct _mesa_prim *prim,
GLuint nr_prims,
const struct _mesa_index_buffer *ib,
GLboolean index_bounds_valid,
GLuint min_index,
- GLuint max_index )
+ GLuint max_index,
+ struct gl_transform_feedback_object *tfb_vertcount )
{
- GLboolean retval;
+ struct brw_context *brw = brw_context(ctx);
+ const struct gl_client_array **arrays = ctx->Array._DrawArrays;
if (!_mesa_check_conditional_render(ctx))
return;
- if (!vbo_all_varyings_in_vbos(arrays)) {
- if (!index_bounds_valid)
- vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
-
- /* Decide if we want to rebase. If so we end up recursing once
- * only into this function.
- */
- if (min_index != 0 && !vbo_any_varyings_in_vbos(arrays)) {
- vbo_rebase_prims(ctx, arrays,
- prim, nr_prims,
- ib, min_index, max_index,
- brw_draw_prims );
- return;
- }
+ /* Handle primitive restart if needed */
+ if (brw_handle_primitive_restart(ctx, prim, nr_prims, ib)) {
+ /* The draw was handled, so we can exit now */
+ return;
}
- /* Make a first attempt at drawing:
+ /* If we're going to have to upload any of the user's vertex arrays, then
+ * get the minimum and maximum of their index buffer so we know what range
+ * to upload.
*/
- retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
+ if (!vbo_all_varyings_in_vbos(arrays) && !index_bounds_valid)
+ vbo_get_minmax_indices(ctx, prim, ib, &min_index, &max_index, nr_prims);
- /* Otherwise, we really are out of memory. Pass the drawing
- * command to the software tnl module and which will in turn call
- * swrast to do the drawing.
+ /* Do GL_SELECT and GL_FEEDBACK rendering using swrast, even though it
+ * won't support all the extensions we support.
*/
- if (!retval) {
- _swsetup_Wakeup(ctx);
+ if (ctx->RenderMode != GL_RENDER) {
+ perf_debug("%s render mode not supported in hardware\n",
+ _mesa_lookup_enum_by_nr(ctx->RenderMode));
+ _swsetup_Wakeup(ctx);
+ _tnl_wakeup(ctx);
_tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
+ return;
}
+ /* Try drawing with the hardware, but don't do anything else if we can't
+ * manage it. swrast doesn't support our featureset, so we can't fall back
+ * to it.
+ */
+ brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
}
void brw_draw_init( struct brw_context *brw )
{
- struct gl_context *ctx = &brw->intel.ctx;
+ struct gl_context *ctx = &brw->ctx;
struct vbo_context *vbo = vbo_context(ctx);
int i;
projects / mesa.git / blobdiff