t.polygon_list_size = panfrost_tiler_full_size(
width, height, t.hierarchy_mask, hierarchy);
- /* Sanity check */
-
if (vertex_count) {
- struct panfrost_bo *tiler_heap;
-
- tiler_heap = panfrost_batch_get_tiler_heap(batch);
t.polygon_list = panfrost_batch_get_polygon_list(batch,
header_size +
t.polygon_list_size);
- /* Allow the entire tiler heap */
- t.heap_start = tiler_heap->gpu;
- t.heap_end = tiler_heap->gpu + tiler_heap->size;
+ t.heap_start = device->tiler_heap->gpu;
+ t.heap_end = device->tiler_heap->gpu + device->tiler_heap->size;
} else {
struct panfrost_bo *tiler_dummy;
* fragment jobs.
*/
struct panfrost_batch *batch = panfrost_get_fresh_batch_for_fbo(ctx);
-
- panfrost_batch_add_fbo_bos(batch);
panfrost_batch_clear(batch, buffers, color, depth, stencil);
}
-/* Reset per-frame context, called on context initialisation as well as after
- * flushing a frame */
-
-void
-panfrost_invalidate_frame(struct panfrost_context *ctx)
-{
- /* TODO: When does this need to be handled? */
- ctx->active_queries = true;
-}
-
bool
panfrost_writes_point_size(struct panfrost_context *ctx)
{
return vs->writes_point_size && ctx->active_prim == PIPE_PRIM_POINTS;
}
-void
-panfrost_vertex_state_upd_attr_offs(struct panfrost_context *ctx,
- struct mali_vertex_tiler_postfix *vertex_postfix)
-{
- if (!ctx->vertex)
- return;
-
- struct panfrost_vertex_state *so = ctx->vertex;
-
- /* Fixup offsets for the second pass. Recall that the hardware
- * calculates attribute addresses as:
- *
- * addr = base + (stride * vtx) + src_offset;
- *
- * However, on Mali, base must be aligned to 64-bytes, so we
- * instead let:
- *
- * base' = base & ~63 = base - (base & 63)
- *
- * To compensate when using base' (see emit_vertex_data), we have
- * to adjust src_offset by the masked off piece:
- *
- * addr' = base' + (stride * vtx) + (src_offset + (base & 63))
- * = base - (base & 63) + (stride * vtx) + src_offset + (base & 63)
- * = base + (stride * vtx) + src_offset
- * = addr;
- *
- * QED.
- */
-
- unsigned start = vertex_postfix->offset_start;
-
- for (unsigned i = 0; i < so->num_elements; ++i) {
- unsigned vbi = so->pipe[i].vertex_buffer_index;
- struct pipe_vertex_buffer *buf = &ctx->vertex_buffers[vbi];
-
- /* Adjust by the masked off bits of the offset. Make sure we
- * read src_offset from so->hw (which is not GPU visible)
- * rather than target (which is) due to caching effects */
-
- unsigned src_offset = so->pipe[i].src_offset;
-
- /* BOs aligned to 4k so guaranteed aligned to 64 */
- src_offset += (buf->buffer_offset & 63);
-
- /* Also, somewhat obscurely per-instance data needs to be
- * offset in response to a delayed start in an indexed draw */
-
- if (so->pipe[i].instance_divisor && ctx->instance_count > 1 && start)
- src_offset -= buf->stride * start;
-
- so->hw[i].src_offset = src_offset;
- }
-}
-
-/* Compute number of UBOs active (more specifically, compute the highest UBO
- * number addressable -- if there are gaps, include them in the count anyway).
- * We always include UBO #0 in the count, since we *need* uniforms enabled for
- * sysvals. */
-
-unsigned
-panfrost_ubo_count(struct panfrost_context *ctx, enum pipe_shader_type stage)
-{
- unsigned mask = ctx->constant_buffer[stage].enabled_mask | 1;
- return 32 - __builtin_clz(mask);
-}
-
/* The entire frame is in memory -- send it off to the kernel! */
void
/* Check if we're scissoring at all */
- if (!(ctx->rasterizer && ctx->rasterizer->base.scissor))
+ if (!ctx->rasterizer->base.scissor)
return false;
return (ss->minx == ss->maxx) || (ss->miny == ss->maxy);
assert(ctx->rasterizer != NULL);
if (!(ctx->draw_modes & (1 << mode))) {
- if (mode == PIPE_PRIM_QUADS && info->count == 4 && !ctx->rasterizer->base.flatshade) {
- mode = PIPE_PRIM_TRIANGLE_FAN;
- } else {
- if (info->count < 4) {
- /* Degenerate case? */
- return;
- }
-
- util_primconvert_save_rasterizer_state(ctx->primconvert, &ctx->rasterizer->base);
- util_primconvert_draw_vbo(ctx->primconvert, info);
+ if (info->count < 4) {
+ /* Degenerate case? */
return;
}
+
+ util_primconvert_save_rasterizer_state(ctx->primconvert, &ctx->rasterizer->base);
+ util_primconvert_draw_vbo(ctx->primconvert, info);
+ return;
}
- /* Now that we have a guaranteed terminating path, find the job.
- * Assignment commented out to prevent unused warning */
+ /* Now that we have a guaranteed terminating path, find the job. */
struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
-
- panfrost_batch_add_fbo_bos(batch);
panfrost_batch_set_requirements(batch);
/* Take into account a negative bias */
panfrost_statistics_record(ctx, info);
- /* Dispatch "compute jobs" for the vertex/tiler pair as (1,
- * vertex_count, 1) */
-
panfrost_pack_work_groups_fused(&vertex_prefix, &tiler_prefix,
1, vertex_count, info->instance_count,
1, 1, 1);
/* Emit all sort of descriptors. */
- panfrost_emit_vertex_data(batch, &vertex_postfix);
+ mali_ptr push_vert = 0, push_frag = 0, attribs = 0;
+
+ vertex_postfix.attribute_meta = panfrost_emit_vertex_data(batch, &attribs);
+ vertex_postfix.attributes = attribs;
panfrost_emit_varying_descriptor(batch,
ctx->padded_count *
ctx->instance_count,
&vertex_postfix, &tiler_postfix,
&primitive_size);
- panfrost_emit_shader_meta(batch, PIPE_SHADER_VERTEX, &vertex_postfix);
- panfrost_emit_shader_meta(batch, PIPE_SHADER_FRAGMENT, &tiler_postfix);
- panfrost_emit_vertex_attr_meta(batch, &vertex_postfix);
- panfrost_emit_sampler_descriptors(batch, PIPE_SHADER_VERTEX, &vertex_postfix);
- panfrost_emit_sampler_descriptors(batch, PIPE_SHADER_FRAGMENT, &tiler_postfix);
- panfrost_emit_texture_descriptors(batch, PIPE_SHADER_VERTEX, &vertex_postfix);
- panfrost_emit_texture_descriptors(batch, PIPE_SHADER_FRAGMENT, &tiler_postfix);
- panfrost_emit_const_buf(batch, PIPE_SHADER_VERTEX, &vertex_postfix);
- panfrost_emit_const_buf(batch, PIPE_SHADER_FRAGMENT, &tiler_postfix);
- panfrost_emit_viewport(batch, &tiler_postfix);
+ vertex_postfix.sampler_descriptor = panfrost_emit_sampler_descriptors(batch, PIPE_SHADER_VERTEX);
+ tiler_postfix.sampler_descriptor = panfrost_emit_sampler_descriptors(batch, PIPE_SHADER_FRAGMENT);
+ vertex_postfix.textures = panfrost_emit_texture_descriptors(batch, PIPE_SHADER_VERTEX);
+ tiler_postfix.textures = panfrost_emit_texture_descriptors(batch, PIPE_SHADER_FRAGMENT);
+ vertex_postfix.uniform_buffers = panfrost_emit_const_buf(batch, PIPE_SHADER_VERTEX, &push_vert);
+ tiler_postfix.uniform_buffers = panfrost_emit_const_buf(batch, PIPE_SHADER_FRAGMENT, &push_frag);
+ vertex_postfix.uniforms = push_vert;
+ tiler_postfix.uniforms = push_frag;
+ tiler_postfix.viewport = panfrost_emit_viewport(batch);
+
+ vertex_postfix.shader = panfrost_emit_compute_shader_meta(batch, PIPE_SHADER_VERTEX);
+ tiler_postfix.shader = panfrost_emit_frag_shader_meta(batch);
panfrost_vt_update_primitive_size(ctx, &tiler_prefix, &primitive_size);
so->base = *cso;
+ /* Gauranteed with the core GL call, so don't expose ARB_polygon_offset */
+ assert(cso->offset_clamp == 0.0);
+
return so;
}
if (!hwcso)
return;
- /* Gauranteed with the core GL call, so don't expose ARB_polygon_offset */
- assert(ctx->rasterizer->base.offset_clamp == 0.0);
-
/* Point sprites are emulated */
struct panfrost_shader_state *variant = panfrost_get_shader_state(ctx, PIPE_SHADER_FRAGMENT);
memcpy(so->pipe, elements, sizeof(*elements) * num_elements);
for (int i = 0; i < num_elements; ++i) {
- so->hw[i].index = i;
-
enum pipe_format fmt = elements[i].src_format;
const struct util_format_description *desc = util_format_description(fmt);
- so->hw[i].unknown1 = 0x2;
-
unsigned swizzle = 0;
if (dev->quirks & HAS_SWIZZLES)
swizzle = panfrost_translate_swizzle_4(desc->swizzle);
swizzle = panfrost_bifrost_swizzle(desc->nr_channels);
enum mali_format hw_format = panfrost_pipe_format_table[desc->format].hw;
- so->hw[i].format = (hw_format << 12) | swizzle;
+ so->formats[i] = (hw_format << 12) | swizzle;
assert(hw_format);
}
/* Let's also prepare vertex builtins */
- so->hw[PAN_VERTEX_ID].format = MALI_R32UI;
if (dev->quirks & HAS_SWIZZLES)
- so->hw[PAN_VERTEX_ID].format = (MALI_R32UI << 12) | panfrost_get_default_swizzle(1);
+ so->formats[PAN_VERTEX_ID] = (MALI_R32UI << 12) | panfrost_get_default_swizzle(1);
else
- so->hw[PAN_VERTEX_ID].format = (MALI_R32UI << 12) | panfrost_bifrost_swizzle(1);
+ so->formats[PAN_VERTEX_ID] = (MALI_R32UI << 12) | panfrost_bifrost_swizzle(1);
if (dev->quirks & HAS_SWIZZLES)
- so->hw[PAN_INSTANCE_ID].format = (MALI_R32UI << 12) | panfrost_get_default_swizzle(1);
+ so->formats[PAN_INSTANCE_ID] = (MALI_R32UI << 12) | panfrost_get_default_swizzle(1);
else
- so->hw[PAN_INSTANCE_ID].format = (MALI_R32UI << 12) | panfrost_bifrost_swizzle(1);
+ so->formats[PAN_INSTANCE_ID] = (MALI_R32UI << 12) | panfrost_bifrost_swizzle(1);
return so;
}
if (unlikely((dev->debug & PAN_DBG_PRECOMPILE) && cso->type == PIPE_SHADER_IR_NIR)) {
struct panfrost_context *ctx = pan_context(pctx);
- struct panfrost_shader_state state;
+ struct panfrost_shader_state state = { 0 };
uint64_t outputs_written;
panfrost_shader_compile(ctx, PIPE_SHADER_IR_NIR,
for (unsigned i = 0; i < cso->variant_count; ++i) {
struct panfrost_shader_state *shader_state = &cso->variants[i];
panfrost_bo_unreference(shader_state->bo);
+
+ if (shader_state->upload.rsrc)
+ pipe_resource_reference(&shader_state->upload.rsrc, NULL);
+
shader_state->bo = NULL;
}
free(cso->variants);
+
free(so);
}
util_copy_framebuffer_state(&ctx->pipe_framebuffer, fb);
ctx->batch = NULL;
- panfrost_invalidate_frame(ctx);
/* We may need to generate a new variant if the fragment shader is
* keyed to the framebuffer format (due to EXT_framebuffer_fetch) */
so->base = *zsa;
pan_pipe_to_stencil(&zsa->stencil[0], &so->stencil_front);
- pan_pipe_to_stencil(&zsa->stencil[1], &so->stencil_back);
-
so->stencil_mask_front = zsa->stencil[0].writemask;
- if (zsa->stencil[1].enabled)
+ if (zsa->stencil[1].enabled) {
+ pan_pipe_to_stencil(&zsa->stencil[1], &so->stencil_back);
so->stencil_mask_back = zsa->stencil[1].writemask;
- else
+ } else {
+ so->stencil_back = so->stencil_front;
so->stencil_mask_back = so->stencil_mask_front;
+ }
/* Alpha lowered by frontend */
assert(!zsa->alpha.enabled);
util_unreference_framebuffer_state(&panfrost->pipe_framebuffer);
u_upload_destroy(pipe->stream_uploader);
+ u_upload_destroy(panfrost->state_uploader);
ralloc_free(pipe);
}
gallium->stream_uploader = u_upload_create_default(gallium);
gallium->const_uploader = gallium->stream_uploader;
- assert(gallium->stream_uploader);
+
+ ctx->state_uploader = u_upload_create(gallium, 4096,
+ PIPE_BIND_CONSTANT_BUFFER, PIPE_USAGE_DYNAMIC, 0);
/* All of our GPUs support ES mode. Midgard supports additionally
* QUADS/QUAD_STRIPS/POLYGON. Bifrost supports just QUADS. */
/* Prepare for render! */
panfrost_batch_init(ctx);
- panfrost_invalidate_frame(ctx);
if (!(dev->quirks & IS_BIFROST)) {
for (unsigned c = 0; c < PIPE_MAX_COLOR_BUFS; ++c)
/* By default mask everything on */
ctx->sample_mask = ~0;
+ ctx->active_queries = true;
return gallium;
}