*/
#include "util/macros.h"
+#include "util/u_vbuf.h"
#include "panfrost-quirks.h"
tp->postfix.occlusion_counter = 0;
}
+static unsigned
+panfrost_translate_index_size(unsigned size)
+{
+ switch (size) {
+ case 1:
+ return MALI_DRAW_INDEXED_UINT8;
+
+ case 2:
+ return MALI_DRAW_INDEXED_UINT16;
+
+ case 4:
+ return MALI_DRAW_INDEXED_UINT32;
+
+ default:
+ unreachable("Invalid index size");
+ }
+}
+
+/* Gets a GPU address for the associated index buffer. Only gauranteed to be
+ * good for the duration of the draw (transient), could last longer. Also get
+ * the bounds on the index buffer for the range accessed by the draw. We do
+ * these operations together because there are natural optimizations which
+ * require them to be together. */
+
+static mali_ptr
+panfrost_get_index_buffer_bounded(struct panfrost_context *ctx,
+ const struct pipe_draw_info *info,
+ unsigned *min_index, unsigned *max_index)
+{
+ struct panfrost_resource *rsrc = pan_resource(info->index.resource);
+ struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
+ off_t offset = info->start * info->index_size;
+ bool needs_indices = true;
+ mali_ptr out = 0;
+
+ if (info->max_index != ~0u) {
+ *min_index = info->min_index;
+ *max_index = info->max_index;
+ needs_indices = false;
+ }
+
+ if (!info->has_user_indices) {
+ /* Only resources can be directly mapped */
+ panfrost_batch_add_bo(batch, rsrc->bo,
+ PAN_BO_ACCESS_SHARED |
+ PAN_BO_ACCESS_READ |
+ PAN_BO_ACCESS_VERTEX_TILER);
+ out = rsrc->bo->gpu + offset;
+
+ /* Check the cache */
+ needs_indices = !panfrost_minmax_cache_get(rsrc->index_cache,
+ info->start,
+ info->count,
+ min_index,
+ max_index);
+ } else {
+ /* Otherwise, we need to upload to transient memory */
+ const uint8_t *ibuf8 = (const uint8_t *) info->index.user;
+ out = panfrost_upload_transient(batch, ibuf8 + offset,
+ info->count *
+ info->index_size);
+ }
+
+ if (needs_indices) {
+ /* Fallback */
+ u_vbuf_get_minmax_index(&ctx->base, info, min_index, max_index);
+
+ if (!info->has_user_indices)
+ panfrost_minmax_cache_add(rsrc->index_cache,
+ info->start, info->count,
+ *min_index, *max_index);
+ }
+
+ return out;
+}
+
+void
+panfrost_vt_set_draw_info(struct panfrost_context *ctx,
+ const struct pipe_draw_info *info,
+ enum mali_draw_mode draw_mode,
+ struct midgard_payload_vertex_tiler *vp,
+ struct midgard_payload_vertex_tiler *tp,
+ unsigned *vertex_count,
+ unsigned *padded_count)
+{
+ tp->prefix.draw_mode = draw_mode;
+
+ unsigned draw_flags = 0;
+
+ if (panfrost_writes_point_size(ctx))
+ draw_flags |= MALI_DRAW_VARYING_SIZE;
+
+ if (info->primitive_restart)
+ draw_flags |= MALI_DRAW_PRIMITIVE_RESTART_FIXED_INDEX;
+
+ /* These doesn't make much sense */
+
+ draw_flags |= 0x3000;
+
+ if (info->index_size) {
+ unsigned min_index = 0, max_index = 0;
+
+ tp->prefix.indices = panfrost_get_index_buffer_bounded(ctx,
+ info,
+ &min_index,
+ &max_index);
+
+ /* Use the corresponding values */
+ *vertex_count = max_index - min_index + 1;
+ tp->offset_start = vp->offset_start = min_index + info->index_bias;
+ tp->prefix.offset_bias_correction = -min_index;
+ tp->prefix.index_count = MALI_POSITIVE(info->count);
+ draw_flags |= panfrost_translate_index_size(info->index_size);
+ } else {
+ tp->prefix.indices = 0;
+ *vertex_count = ctx->vertex_count;
+ tp->offset_start = vp->offset_start = info->start;
+ tp->prefix.offset_bias_correction = 0;
+ tp->prefix.index_count = MALI_POSITIVE(ctx->vertex_count);
+ }
+
+ tp->prefix.unknown_draw = draw_flags;
+
+ /* Encode the padded vertex count */
+
+ if (info->instance_count > 1) {
+ *padded_count = panfrost_padded_vertex_count(*vertex_count);
+
+ unsigned shift = __builtin_ctz(ctx->padded_count);
+ unsigned k = ctx->padded_count >> (shift + 1);
+
+ tp->instance_shift = vp->instance_shift = shift;
+ tp->instance_odd = vp->instance_odd = k;
+ } else {
+ *padded_count = *vertex_count;
+
+ /* Reset instancing state */
+ tp->instance_shift = vp->instance_shift = 0;
+ tp->instance_odd = vp->instance_odd = 0;
+ }
+}
+
static void
panfrost_shader_meta_init(struct panfrost_context *ctx,
enum pipe_shader_type st,
meta->midgard1.uniform_buffer_count = panfrost_ubo_count(ctx, st);
}
-unsigned
+static unsigned
panfrost_translate_compare_func(enum pipe_compare_func in)
{
switch (in) {
}
}
+static unsigned
+translate_tex_wrap(enum pipe_tex_wrap w)
+{
+ switch (w) {
+ case PIPE_TEX_WRAP_REPEAT:
+ return MALI_WRAP_REPEAT;
+
+ case PIPE_TEX_WRAP_CLAMP:
+ return MALI_WRAP_CLAMP;
+
+ case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+ return MALI_WRAP_CLAMP_TO_EDGE;
+
+ case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+ return MALI_WRAP_CLAMP_TO_BORDER;
+
+ case PIPE_TEX_WRAP_MIRROR_REPEAT:
+ return MALI_WRAP_MIRRORED_REPEAT;
+
+ case PIPE_TEX_WRAP_MIRROR_CLAMP:
+ return MALI_WRAP_MIRRORED_CLAMP;
+
+ case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+ return MALI_WRAP_MIRRORED_CLAMP_TO_EDGE;
+
+ case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+ return MALI_WRAP_MIRRORED_CLAMP_TO_BORDER;
+
+ default:
+ unreachable("Invalid wrap");
+ }
+}
+
+void panfrost_sampler_desc_init(const struct pipe_sampler_state *cso,
+ struct mali_sampler_descriptor *hw)
+{
+ unsigned func = panfrost_translate_compare_func(cso->compare_func);
+ bool min_nearest = cso->min_img_filter == PIPE_TEX_FILTER_NEAREST;
+ bool mag_nearest = cso->mag_img_filter == PIPE_TEX_FILTER_NEAREST;
+ bool mip_linear = cso->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR;
+ unsigned min_filter = min_nearest ? MALI_SAMP_MIN_NEAREST : 0;
+ unsigned mag_filter = mag_nearest ? MALI_SAMP_MAG_NEAREST : 0;
+ unsigned mip_filter = mip_linear ?
+ (MALI_SAMP_MIP_LINEAR_1 | MALI_SAMP_MIP_LINEAR_2) : 0;
+ unsigned normalized = cso->normalized_coords ? MALI_SAMP_NORM_COORDS : 0;
+
+ *hw = (struct mali_sampler_descriptor) {
+ .filter_mode = min_filter | mag_filter | mip_filter |
+ normalized,
+ .wrap_s = translate_tex_wrap(cso->wrap_s),
+ .wrap_t = translate_tex_wrap(cso->wrap_t),
+ .wrap_r = translate_tex_wrap(cso->wrap_r),
+ .compare_func = panfrost_flip_compare_func(func),
+ .border_color = {
+ cso->border_color.f[0],
+ cso->border_color.f[1],
+ cso->border_color.f[2],
+ cso->border_color.f[3]
+ },
+ .min_lod = FIXED_16(cso->min_lod, false), /* clamp at 0 */
+ .max_lod = FIXED_16(cso->max_lod, false),
+ .lod_bias = FIXED_16(cso->lod_bias, true), /* can be negative */
+ .seamless_cube_map = cso->seamless_cube_map,
+ };
+
+ /* If necessary, we disable mipmapping in the sampler descriptor by
+ * clamping the LOD as tight as possible (from 0 to epsilon,
+ * essentially -- remember these are fixed point numbers, so
+ * epsilon=1/256) */
+
+ if (cso->min_mip_filter == PIPE_TEX_MIPFILTER_NONE)
+ hw->max_lod = hw->min_lod + 1;
+}
+
static void
panfrost_make_stencil_state(const struct pipe_stencil_state *in,
struct mali_stencil_test *out)
vtp->postfix.shared_memory = panfrost_upload_transient(batch, &shared,
sizeof(shared));
}
+
+static mali_ptr
+panfrost_get_tex_desc(struct panfrost_batch *batch,
+ enum pipe_shader_type st,
+ struct panfrost_sampler_view *view)
+{
+ if (!view)
+ return (mali_ptr) 0;
+
+ struct pipe_sampler_view *pview = &view->base;
+ struct panfrost_resource *rsrc = pan_resource(pview->texture);
+
+ /* Add the BO to the job so it's retained until the job is done. */
+
+ panfrost_batch_add_bo(batch, rsrc->bo,
+ PAN_BO_ACCESS_SHARED | PAN_BO_ACCESS_READ |
+ panfrost_bo_access_for_stage(st));
+
+ panfrost_batch_add_bo(batch, view->bo,
+ PAN_BO_ACCESS_SHARED | PAN_BO_ACCESS_READ |
+ panfrost_bo_access_for_stage(st));
+
+ return view->bo->gpu;
+}
+
+void
+panfrost_emit_texture_descriptors(struct panfrost_batch *batch,
+ enum pipe_shader_type stage,
+ struct midgard_payload_vertex_tiler *vtp)
+{
+ struct panfrost_context *ctx = batch->ctx;
+
+ if (!ctx->sampler_view_count[stage])
+ return;
+
+ uint64_t trampolines[PIPE_MAX_SHADER_SAMPLER_VIEWS];
+
+ for (int i = 0; i < ctx->sampler_view_count[stage]; ++i)
+ trampolines[i] = panfrost_get_tex_desc(batch, stage,
+ ctx->sampler_views[stage][i]);
+
+ vtp->postfix.texture_trampoline = panfrost_upload_transient(batch,
+ trampolines,
+ sizeof(uint64_t) *
+ ctx->sampler_view_count[stage]);
+}
+
+void
+panfrost_emit_sampler_descriptors(struct panfrost_batch *batch,
+ enum pipe_shader_type stage,
+ struct midgard_payload_vertex_tiler *vtp)
+{
+ struct panfrost_context *ctx = batch->ctx;
+
+ if (!ctx->sampler_count[stage])
+ return;
+
+ size_t desc_size = sizeof(struct mali_sampler_descriptor);
+ size_t transfer_size = desc_size * ctx->sampler_count[stage];
+ struct panfrost_transfer transfer = panfrost_allocate_transient(batch,
+ transfer_size);
+ struct mali_sampler_descriptor *desc = (struct mali_sampler_descriptor *)transfer.cpu;
+
+ for (int i = 0; i < ctx->sampler_count[stage]; ++i)
+ desc[i] = ctx->samplers[stage][i]->hw;
+
+ vtp->postfix.sampler_descriptor = transfer.gpu;
+}
+
+void
+panfrost_emit_vertex_attr_meta(struct panfrost_batch *batch,
+ struct midgard_payload_vertex_tiler *vp)
+{
+ struct panfrost_context *ctx = batch->ctx;
+
+ if (!ctx->vertex)
+ return;
+
+ struct panfrost_vertex_state *so = ctx->vertex;
+
+ panfrost_vertex_state_upd_attr_offs(ctx, vp);
+ vp->postfix.attribute_meta = panfrost_upload_transient(batch, so->hw,
+ sizeof(*so->hw) *
+ PAN_MAX_ATTRIBUTE);
+}
+
+void
+panfrost_emit_vertex_tiler_jobs(struct panfrost_batch *batch,
+ struct midgard_payload_vertex_tiler *vp,
+ struct midgard_payload_vertex_tiler *tp)
+{
+ struct panfrost_context *ctx = batch->ctx;
+ bool wallpapering = ctx->wallpaper_batch && batch->tiler_dep;
+
+ if (wallpapering) {
+ /* Inject in reverse order, with "predicted" job indices.
+ * THIS IS A HACK XXX */
+ panfrost_new_job(batch, JOB_TYPE_TILER, false,
+ batch->job_index + 2, tp, sizeof(*tp), true);
+ panfrost_new_job(batch, JOB_TYPE_VERTEX, false, 0,
+ vp, sizeof(*vp), true);
+ return;
+ }
+
+ /* If rasterizer discard is enable, only submit the vertex */
+
+ bool rasterizer_discard = ctx->rasterizer &&
+ ctx->rasterizer->base.rasterizer_discard;
+
+ unsigned vertex = panfrost_new_job(batch, JOB_TYPE_VERTEX, false, 0,
+ vp, sizeof(*vp), false);
+
+ if (rasterizer_discard)
+ return;
+
+ panfrost_new_job(batch, JOB_TYPE_TILER, false, vertex, tp, sizeof(*tp),
+ false);
+}