#include "panfrost-quirks.h"
-#include "pan_allocate.h"
+#include "pan_pool.h"
#include "pan_bo.h"
#include "pan_cmdstream.h"
#include "pan_context.h"
.scratchpad = panfrost_batch_get_scratchpad(batch, shift, dev->thread_tls_alloc, dev->core_count)->gpu,
.shared_workgroup_count = ~0,
};
- postfix->shared_memory = panfrost_upload_transient(batch, &shared, sizeof(shared));
+ postfix->shared_memory = panfrost_pool_upload(&batch->pool, &shared, sizeof(shared));
}
static void
sizeof(struct mali_single_framebuffer) :
sizeof(struct mali_framebuffer);
- batch->framebuffer = panfrost_allocate_transient(batch, size);
+ batch->framebuffer = panfrost_pool_alloc(&batch->pool, size);
/* Tag the pointer */
if (!(dev->quirks & MIDGARD_SFBD))
struct mali_vertex_tiler_postfix *postfix)
{
SET_BIT(postfix->gl_enables, MALI_OCCLUSION_QUERY, ctx->occlusion_query);
- if (ctx->occlusion_query)
+ if (ctx->occlusion_query) {
postfix->occlusion_counter = ctx->occlusion_query->bo->gpu;
- else
+ panfrost_batch_add_bo(ctx->batch, ctx->occlusion_query->bo,
+ PAN_BO_ACCESS_SHARED |
+ PAN_BO_ACCESS_RW |
+ PAN_BO_ACCESS_FRAGMENT);
+ } else {
postfix->occlusion_counter = 0;
+ }
}
void
} 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,
+ out = panfrost_pool_upload(&batch->pool, ibuf8 + offset,
info->count *
info->index_size);
}
fragmeta->depth_factor = 0.0f;
SET_BIT(fragmeta->unknown2_4, MALI_DEPTH_RANGE_A, false);
SET_BIT(fragmeta->unknown2_4, MALI_DEPTH_RANGE_B, false);
+ SET_BIT(fragmeta->unknown2_3, MALI_DEPTH_CLIP_NEAR, true);
+ SET_BIT(fragmeta->unknown2_3, MALI_DEPTH_CLIP_FAR, true);
return;
}
- bool msaa = ctx->rasterizer->base.multisample;
+ struct pipe_rasterizer_state *rast = &ctx->rasterizer->base;
+
+ bool msaa = rast->multisample;
/* TODO: Sample size */
SET_BIT(fragmeta->unknown2_3, MALI_HAS_MSAA, msaa);
SET_BIT(fragmeta->unknown2_4, MALI_NO_MSAA, !msaa);
- fragmeta->depth_units = ctx->rasterizer->base.offset_units * 2.0f;
- fragmeta->depth_factor = ctx->rasterizer->base.offset_scale;
+ fragmeta->depth_units = rast->offset_units * 2.0f;
+ fragmeta->depth_factor = rast->offset_scale;
/* XXX: Which bit is which? Does this maybe allow offseting not-tri? */
- SET_BIT(fragmeta->unknown2_4, MALI_DEPTH_RANGE_A,
- ctx->rasterizer->base.offset_tri);
- SET_BIT(fragmeta->unknown2_4, MALI_DEPTH_RANGE_B,
- ctx->rasterizer->base.offset_tri);
+ SET_BIT(fragmeta->unknown2_4, MALI_DEPTH_RANGE_A, rast->offset_tri);
+ SET_BIT(fragmeta->unknown2_4, MALI_DEPTH_RANGE_B, rast->offset_tri);
+
+ SET_BIT(fragmeta->unknown2_3, MALI_DEPTH_CLIP_NEAR, rast->depth_clip_near);
+ SET_BIT(fragmeta->unknown2_3, MALI_DEPTH_CLIP_FAR, rast->depth_clip_far);
}
static void
(dev->quirks & MIDGARD_SFBD) && ctx->blend &&
!ctx->blend->base.dither);
+ SET_BIT(fragmeta->unknown2_4, MALI_ALPHA_TO_COVERAGE,
+ ctx->blend->base.alpha_to_coverage);
+
/* Get blending setup */
unsigned rt_count = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
fs = panfrost_get_shader_state(ctx, PIPE_SHADER_FRAGMENT);
- fragmeta->alpha_coverage = ~MALI_ALPHA_COVERAGE(0.000000);
- fragmeta->unknown2_3 = MALI_DEPTH_FUNC(MALI_FUNC_ALWAYS) | 0x3010;
+ bool msaa = ctx->rasterizer && ctx->rasterizer->base.multisample;
+ fragmeta->coverage_mask = (msaa ? ctx->sample_mask : ~0) & 0xF;
+
+ fragmeta->unknown2_3 = MALI_DEPTH_FUNC(MALI_FUNC_ALWAYS) | 0x10;
fragmeta->unknown2_4 = 0x4e0;
/* unknown2_4 has 0x10 bit set on T6XX and T720. We don't know why this
SET_BIT(fragmeta->midgard1.flags_lo, MALI_EARLY_Z,
!fs->can_discard && !fs->writes_global &&
- !fs->writes_depth && !fs->writes_stencil);
+ !fs->writes_depth && !fs->writes_stencil &&
+ !ctx->blend->base.alpha_to_coverage);
/* Add the writes Z/S flags if needed. */
SET_BIT(fragmeta->midgard1.flags_lo, MALI_WRITES_Z, fs->writes_depth);
panfrost_frag_shader_meta_init(ctx, &meta, rts);
- xfer = panfrost_allocate_transient(batch, desc_size);
+ xfer = panfrost_pool_alloc(&batch->pool, desc_size);
memcpy(xfer.cpu, &meta, sizeof(meta));
memcpy(xfer.cpu + sizeof(meta), rts, rt_size * rt_count);
shader_ptr = xfer.gpu;
} else {
- shader_ptr = panfrost_upload_transient(batch, &meta,
+ shader_ptr = panfrost_pool_upload(&batch->pool, &meta,
sizeof(meta));
}
mvp->viewport0[1] = miny;
mvp->viewport1[1] = MALI_POSITIVE(maxy);
- mvp->clip_minz = minz;
- mvp->clip_maxz = maxz;
+ bool clip_near = true;
+ bool clip_far = true;
+
+ if (ctx->rasterizer) {
+ clip_near = ctx->rasterizer->base.depth_clip_near;
+ clip_far = ctx->rasterizer->base.depth_clip_far;
+ }
+
+ mvp->clip_minz = clip_near ? minz : -INFINITY;
+ mvp->clip_maxz = clip_far ? maxz : INFINITY;
}
void
mvp.viewport1[0] + 1,
mvp.viewport1[1] + 1);
- tiler_postfix->viewport = panfrost_upload_transient(batch, &mvp,
+ tiler_postfix->viewport = panfrost_pool_upload(&batch->pool, &mvp,
sizeof(mvp));
}
* PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT */
return rsrc->bo->gpu + cb->buffer_offset;
} else if (cb->user_buffer) {
- return panfrost_upload_transient(batch,
+ return panfrost_pool_upload(&batch->pool,
cb->user_buffer +
cb->buffer_offset,
cb->buffer_size);
size_t sys_size = sizeof(float) * 4 * ss->sysval_count;
size_t uniform_size = has_uniforms ? (buf->cb[0].buffer_size) : 0;
size_t size = sys_size + uniform_size;
- struct panfrost_transfer transfer = panfrost_allocate_transient(batch,
+ struct panfrost_transfer transfer = panfrost_pool_alloc(&batch->pool,
size);
/* Upload sysvals requested by the shader */
ubos[ubo] = MALI_MAKE_UBO(aligned / bytes_per_field, gpu);
}
- mali_ptr ubufs = panfrost_upload_transient(batch, ubos, sz);
+ mali_ptr ubufs = panfrost_pool_upload(&batch->pool, ubos, sz);
postfix->uniforms = transfer.gpu;
postfix->uniform_buffers = ubufs;
.shared_shift = util_logbase2(single_size) - 1
};
- vtp->postfix.shared_memory = panfrost_upload_transient(batch, &shared,
+ vtp->postfix.shared_memory = panfrost_pool_upload(&batch->pool, &shared,
sizeof(shared));
}
PAN_BO_ACCESS_SHARED | PAN_BO_ACCESS_READ |
panfrost_bo_access_for_stage(st));
- panfrost_batch_add_bo(batch, view->midgard_bo,
+ panfrost_batch_add_bo(batch, view->bo,
PAN_BO_ACCESS_SHARED | PAN_BO_ACCESS_READ |
panfrost_bo_access_for_stage(st));
- return view->midgard_bo->gpu;
+ return view->bo->gpu;
+}
+
+static void
+panfrost_update_sampler_view(struct panfrost_sampler_view *view,
+ struct pipe_context *pctx)
+{
+ struct panfrost_resource *rsrc = pan_resource(view->base.texture);
+ if (view->texture_bo != rsrc->bo->gpu ||
+ view->layout != rsrc->layout) {
+ panfrost_bo_unreference(view->bo);
+ panfrost_create_sampler_view_bo(view, pctx, &rsrc->base);
+ }
}
void
struct panfrost_sampler_view *view = ctx->sampler_views[stage][i];
struct pipe_sampler_view *pview = &view->base;
struct panfrost_resource *rsrc = pan_resource(pview->texture);
+ panfrost_update_sampler_view(view, &ctx->base);
/* Add the BOs to the job so they are retained until the job is done. */
PAN_BO_ACCESS_SHARED | PAN_BO_ACCESS_READ |
panfrost_bo_access_for_stage(stage));
- panfrost_batch_add_bo(batch, view->bifrost_bo,
+ panfrost_batch_add_bo(batch, view->bo,
PAN_BO_ACCESS_SHARED | PAN_BO_ACCESS_READ |
panfrost_bo_access_for_stage(stage));
memcpy(&descriptors[i], view->bifrost_descriptor, sizeof(*view->bifrost_descriptor));
}
- postfix->textures = panfrost_upload_transient(batch,
+ postfix->textures = panfrost_pool_upload(&batch->pool,
descriptors,
sizeof(struct bifrost_texture_descriptor) *
ctx->sampler_view_count[stage]);
} else {
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]);
+ for (int i = 0; i < ctx->sampler_view_count[stage]; ++i) {
+ struct panfrost_sampler_view *view = ctx->sampler_views[stage][i];
+
+ panfrost_update_sampler_view(view, &ctx->base);
+
+ trampolines[i] = panfrost_get_tex_desc(batch, stage, view);
+ }
- postfix->textures = panfrost_upload_transient(batch,
+ postfix->textures = panfrost_pool_upload(&batch->pool,
trampolines,
sizeof(uint64_t) *
ctx->sampler_view_count[stage]);
if (device->quirks & IS_BIFROST) {
size_t desc_size = sizeof(struct bifrost_sampler_descriptor);
size_t transfer_size = desc_size * ctx->sampler_count[stage];
- struct panfrost_transfer transfer = panfrost_allocate_transient(batch,
+ struct panfrost_transfer transfer = panfrost_pool_alloc(&batch->pool,
transfer_size);
struct bifrost_sampler_descriptor *desc = (struct bifrost_sampler_descriptor *)transfer.cpu;
} else {
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,
+ struct panfrost_transfer transfer = panfrost_pool_alloc(&batch->pool,
transfer_size);
struct mali_sampler_descriptor *desc = (struct mali_sampler_descriptor *)transfer.cpu;
struct panfrost_vertex_state *so = ctx->vertex;
panfrost_vertex_state_upd_attr_offs(ctx, vertex_postfix);
- vertex_postfix->attribute_meta = panfrost_upload_transient(batch, so->hw,
+ vertex_postfix->attribute_meta = panfrost_pool_upload(&batch->pool, so->hw,
sizeof(*so->hw) *
PAN_MAX_ATTRIBUTE);
}
/* Upload whatever we emitted and go */
- vertex_postfix->attributes = panfrost_upload_transient(batch, attrs,
+ vertex_postfix->attributes = panfrost_pool_upload(&batch->pool, attrs,
k * sizeof(*attrs));
}
slot->size = stride * count;
slot->shift = slot->extra_flags = 0;
- struct panfrost_transfer transfer = panfrost_allocate_transient(batch,
+ struct panfrost_transfer transfer = panfrost_pool_alloc(&batch->pool,
slot->size);
slot->elements = transfer.gpu | MALI_ATTR_LINEAR;
unsigned max_size = target->buffer_size;
unsigned expected_size = slot->stride * count;
- slot->size = MIN2(max_size, expected_size);
-
/* Grab the BO and bind it to the batch */
struct panfrost_bo *bo = pan_resource(target->buffer)->bo;
PAN_BO_ACCESS_VERTEX_TILER |
PAN_BO_ACCESS_FRAGMENT);
+ /* We will have an offset applied to get alignment */
mali_ptr addr = bo->gpu + target->buffer_offset + (offset * slot->stride);
- slot->elements = addr;
-}
-
-/* Given a shader and buffer indices, link varying metadata together */
-
-static bool
-is_special_varying(gl_varying_slot loc)
-{
- switch (loc) {
- case VARYING_SLOT_POS:
- case VARYING_SLOT_PSIZ:
- case VARYING_SLOT_PNTC:
- case VARYING_SLOT_FACE:
- return true;
- default:
- return false;
- }
-}
-
-static void
-panfrost_emit_varying_meta(void *outptr, struct panfrost_shader_state *ss,
- signed general, signed gl_Position,
- signed gl_PointSize, signed gl_PointCoord,
- signed gl_FrontFacing)
-{
- struct mali_attr_meta *out = (struct mali_attr_meta *) outptr;
-
- for (unsigned i = 0; i < ss->varying_count; ++i) {
- gl_varying_slot location = ss->varyings_loc[i];
- int index = -1;
-
- switch (location) {
- case VARYING_SLOT_POS:
- index = gl_Position;
- break;
- case VARYING_SLOT_PSIZ:
- index = gl_PointSize;
- break;
- case VARYING_SLOT_PNTC:
- index = gl_PointCoord;
- break;
- case VARYING_SLOT_FACE:
- index = gl_FrontFacing;
- break;
- default:
- index = general;
- break;
- }
-
- assert(index >= 0);
- out[i].index = index;
- }
+ slot->elements = (addr & ~63) | MALI_ATTR_LINEAR;
+ slot->size = MIN2(max_size, expected_size) + (addr & 63);
}
static bool
unsigned *streamout_offsets,
unsigned quirks,
unsigned *gen_offsets,
+ enum mali_format *gen_formats,
unsigned *gen_stride,
unsigned idx,
bool should_alloc,
bool is_fragment)
{
gl_varying_slot loc = stage->varyings_loc[idx];
- enum mali_format format = stage->varyings[idx].format;
+ enum mali_format format = stage->varyings[idx];
+
+ /* Override format to match linkage */
+ if (!should_alloc && gen_formats[idx])
+ format = gen_formats[idx];
if (has_point_coord(stage->point_sprite_mask, loc)) {
return pan_emit_vary_special(present, PAN_VARY_PNTCOORD, quirks);
if (should_alloc) {
/* We're linked, so allocate a space via a watermark allocation */
- gen_offsets[idx] = *gen_stride;
- offset = *gen_stride;
+ enum mali_format alt = other->varyings[other_idx];
+
+ /* Do interpolation at minimum precision */
+ unsigned size_main = pan_varying_size(format);
+ unsigned size_alt = pan_varying_size(alt);
+ unsigned size = MIN2(size_main, size_alt);
/* If a varying is marked for XFB but not actually captured, we
* should match the format to the format that would otherwise
if (xfb->so_mask & (1ull << loc)) {
struct pipe_stream_output *o = pan_get_so(&xfb->stream_output, loc);
format = pan_xfb_format(format, o->num_components);
+ size = pan_varying_size(format);
+ } else if (size == size_alt) {
+ format = alt;
}
- *gen_stride += pan_varying_size(format);
+ gen_offsets[idx] = *gen_stride;
+ gen_formats[other_idx] = format;
+ offset = *gen_stride;
+ *gen_stride += size;
}
return pan_emit_vary(present, PAN_VARY_GENERAL,
{
/* Load the shaders */
struct panfrost_context *ctx = batch->ctx;
- struct panfrost_device *device = pan_device(ctx->base.screen);
+ struct panfrost_device *dev = pan_device(ctx->base.screen);
struct panfrost_shader_state *vs, *fs;
- unsigned int num_gen_varyings = 0;
size_t vs_size, fs_size;
/* Allocate the varying descriptor */
vs_size = sizeof(struct mali_attr_meta) * vs->varying_count;
fs_size = sizeof(struct mali_attr_meta) * fs->varying_count;
- struct panfrost_transfer trans = panfrost_allocate_transient(batch,
+ struct panfrost_transfer trans = panfrost_pool_alloc(&batch->pool,
vs_size +
fs_size);
struct pipe_stream_output_info *so = &vs->stream_output;
+ unsigned present = pan_varying_present(vs, fs, dev->quirks);
/* Check if this varying is linked by us. This is the case for
* general-purpose, non-captured varyings. If it is, link it. If it's
* not, use the provided stream out information to determine the
* offset, since it was already linked for us. */
- for (unsigned i = 0; i < vs->varying_count; i++) {
- gl_varying_slot loc = vs->varyings_loc[i];
+ unsigned gen_offsets[32];
+ enum mali_format gen_formats[32];
+ memset(gen_offsets, 0, sizeof(gen_offsets));
+ memset(gen_formats, 0, sizeof(gen_formats));
- bool special = is_special_varying(loc);
- bool captured = ((vs->so_mask & (1ll << loc)) ? true : false);
+ unsigned gen_stride = 0;
+ assert(vs->varying_count < ARRAY_SIZE(gen_offsets));
+ assert(fs->varying_count < ARRAY_SIZE(gen_offsets));
- if (captured) {
- struct pipe_stream_output *o = pan_get_so(so, loc);
+ unsigned streamout_offsets[32];
- unsigned dst_offset = o->dst_offset * 4; /* dwords */
- vs->varyings[i].src_offset = dst_offset;
- } else if (!special) {
- vs->varyings[i].src_offset = 16 * (num_gen_varyings++);
- }
+ for (unsigned i = 0; i < ctx->streamout.num_targets; ++i) {
+ streamout_offsets[i] = panfrost_streamout_offset(
+ so->stride[i],
+ ctx->streamout.offsets[i],
+ ctx->streamout.targets[i]);
}
- /* Conversely, we need to set src_offset for the captured varyings.
- * Here, the layout is defined by the stream out info, not us */
-
- /* Link up with fragment varyings */
- bool reads_point_coord = fs->reads_point_coord;
-
- for (unsigned i = 0; i < fs->varying_count; i++) {
- gl_varying_slot loc = fs->varyings_loc[i];
- unsigned src_offset;
- signed vs_idx = -1;
-
- /* Link up */
- for (unsigned j = 0; j < vs->varying_count; ++j) {
- if (vs->varyings_loc[j] == loc) {
- vs_idx = j;
- break;
- }
- }
-
- /* Either assign or reuse */
- if (vs_idx >= 0)
- src_offset = vs->varyings[vs_idx].src_offset;
- else
- src_offset = 16 * (num_gen_varyings++);
-
- fs->varyings[i].src_offset = src_offset;
-
- if (has_point_coord(fs->point_sprite_mask, loc))
- reads_point_coord = true;
- }
-
- memcpy(trans.cpu, vs->varyings, vs_size);
- memcpy(trans.cpu + vs_size, fs->varyings, fs_size);
-
- union mali_attr varyings[PIPE_MAX_ATTRIBS] = {0};
+ struct mali_attr_meta *ovs = (struct mali_attr_meta *)trans.cpu;
+ struct mali_attr_meta *ofs = ovs + vs->varying_count;
- /* Figure out how many streamout buffers could be bound */
- unsigned so_count = ctx->streamout.num_targets;
for (unsigned i = 0; i < vs->varying_count; i++) {
- gl_varying_slot loc = vs->varyings_loc[i];
-
- bool captured = ((vs->so_mask & (1ll << loc)) ? true : false);
- if (!captured) continue;
+ ovs[i] = panfrost_emit_varying(vs, fs, vs, present,
+ ctx->streamout.num_targets, streamout_offsets,
+ dev->quirks,
+ gen_offsets, gen_formats, &gen_stride, i, true, false);
+ }
- struct pipe_stream_output *o = pan_get_so(so, loc);
- so_count = MAX2(so_count, o->output_buffer + 1);
+ for (unsigned i = 0; i < fs->varying_count; i++) {
+ ofs[i] = panfrost_emit_varying(fs, vs, vs, present,
+ ctx->streamout.num_targets, streamout_offsets,
+ dev->quirks,
+ gen_offsets, gen_formats, &gen_stride, i, false, true);
}
- signed idx = so_count;
- signed general = idx++;
- signed gl_Position = idx++;
- signed gl_PointSize = vs->writes_point_size ? (idx++) : -1;
- signed gl_PointCoord = reads_point_coord ? (idx++) : -1;
- signed gl_FrontFacing = fs->reads_face ? (idx++) : -1;
- signed gl_FragCoord = (fs->reads_frag_coord &&
- !(device->quirks & IS_BIFROST))
- ? (idx++) : -1;
+ unsigned xfb_base = pan_xfb_base(present);
+ struct panfrost_transfer T = panfrost_pool_alloc(&batch->pool,
+ sizeof(union mali_attr) * (xfb_base + ctx->streamout.num_targets));
+ union mali_attr *varyings = (union mali_attr *) T.cpu;
/* Emit the stream out buffers */
unsigned out_count = u_stream_outputs_for_vertices(ctx->active_prim,
ctx->vertex_count);
- for (unsigned i = 0; i < so_count; ++i) {
- if (i < ctx->streamout.num_targets) {
- panfrost_emit_streamout(batch, &varyings[i],
- so->stride[i],
- ctx->streamout.offsets[i],
- out_count,
- ctx->streamout.targets[i]);
- } else {
- /* Emit a dummy buffer */
- panfrost_emit_varyings(batch, &varyings[i],
- so->stride[i] * 4,
- out_count);
-
- /* Clear the attribute type */
- varyings[i].elements &= ~0xF;
- }
+ for (unsigned i = 0; i < ctx->streamout.num_targets; ++i) {
+ panfrost_emit_streamout(batch, &varyings[xfb_base + i],
+ so->stride[i],
+ ctx->streamout.offsets[i],
+ out_count,
+ ctx->streamout.targets[i]);
}
- panfrost_emit_varyings(batch, &varyings[general],
- num_gen_varyings * 16,
- vertex_count);
-
- mali_ptr varyings_p;
+ panfrost_emit_varyings(batch,
+ &varyings[pan_varying_index(present, PAN_VARY_GENERAL)],
+ gen_stride, vertex_count);
/* fp32 vec4 gl_Position */
- varyings_p = panfrost_emit_varyings(batch, &varyings[gl_Position],
- sizeof(float) * 4, vertex_count);
- tiler_postfix->position_varying = varyings_p;
-
-
- if (panfrost_writes_point_size(ctx)) {
- varyings_p = panfrost_emit_varyings(batch,
- &varyings[gl_PointSize],
- 2, vertex_count);
- primitive_size->pointer = varyings_p;
- }
-
- if (gl_PointCoord >= 0)
- varyings[gl_PointCoord].elements = MALI_VARYING_POINT_COORD;
-
- if (gl_FrontFacing >= 0)
- varyings[gl_FrontFacing].elements = MALI_VARYING_FRONT_FACING;
-
- if (gl_FragCoord >= 0)
- varyings[gl_FragCoord].elements = MALI_VARYING_FRAG_COORD;
-
- assert(!(device->quirks & IS_BIFROST) || !(reads_point_coord));
-
- /* Let's go ahead and link varying meta to the buffer in question, now
- * that that information is available. VARYING_SLOT_POS is mapped to
- * gl_FragCoord for fragment shaders but gl_Positionf or vertex shaders
- * */
-
- panfrost_emit_varying_meta(trans.cpu, vs, general, gl_Position,
- gl_PointSize, gl_PointCoord,
- gl_FrontFacing);
-
- panfrost_emit_varying_meta(trans.cpu + vs_size, fs, general,
- gl_FragCoord, gl_PointSize,
- gl_PointCoord, gl_FrontFacing);
-
- /* Replace streamout */
-
- struct mali_attr_meta *ovs = (struct mali_attr_meta *)trans.cpu;
- struct mali_attr_meta *ofs = ovs + vs->varying_count;
-
- for (unsigned i = 0; i < vs->varying_count; i++) {
- gl_varying_slot loc = vs->varyings_loc[i];
-
- /* If we write gl_PointSize from the vertex shader but don't
- * consume it, no memory will be allocated for it, so if we
- * attempted to write anyway we would dereference a NULL
- * pointer on the GPU. Midgard seems fine with this; Bifrost
- * faults. */
-
- if (loc == VARYING_SLOT_PSIZ && !panfrost_writes_point_size(ctx))
- ovs[i].format = MALI_VARYING_DISCARD;
-
- bool captured = ((vs->so_mask & (1ll << loc)) ? true : false);
- if (!captured)
- continue;
-
- struct pipe_stream_output *o = pan_get_so(so, loc);
- ovs[i].index = o->output_buffer;
-
- assert(o->stream == 0);
- ovs[i].format = (vs->varyings[i].format & ~MALI_NR_CHANNELS(4))
- | MALI_NR_CHANNELS(o->num_components);
-
- if (device->quirks & HAS_SWIZZLES)
- ovs[i].swizzle = panfrost_get_default_swizzle(o->num_components);
- else
- ovs[i].swizzle = panfrost_bifrost_swizzle(o->num_components);
-
- /* Link to the fragment */
- signed fs_idx = -1;
-
- /* Link up */
- for (unsigned j = 0; j < fs->varying_count; ++j) {
- if (fs->varyings_loc[j] == loc) {
- fs_idx = j;
- break;
- }
- }
-
- if (fs_idx >= 0) {
- ofs[fs_idx].index = ovs[i].index;
- ofs[fs_idx].format = ovs[i].format;
- ofs[fs_idx].swizzle = ovs[i].swizzle;
- }
- }
-
- /* Replace point sprite */
- for (unsigned i = 0; i < fs->varying_count; i++) {
- /* If we have a point sprite replacement, handle that here. We
- * have to translate location first. TODO: Flip y in shader.
- * We're already keying ... just time crunch .. */
-
- if (has_point_coord(fs->point_sprite_mask,
- fs->varyings_loc[i])) {
- ofs[i].index = gl_PointCoord;
-
- /* Swizzle out the z/w to 0/1 */
- ofs[i].format = MALI_RG16F;
- ofs[i].swizzle = panfrost_get_default_swizzle(2);
- }
+ tiler_postfix->position_varying = panfrost_emit_varyings(batch,
+ &varyings[pan_varying_index(present, PAN_VARY_POSITION)],
+ sizeof(float) * 4, vertex_count);
+
+ if (present & (1 << PAN_VARY_PSIZ)) {
+ primitive_size->pointer = panfrost_emit_varyings(batch,
+ &varyings[pan_varying_index(present, PAN_VARY_PSIZ)],
+ 2, vertex_count);
}
- /* Fix up unaligned addresses */
- for (unsigned i = 0; i < so_count; ++i) {
- if (varyings[i].elements < MALI_RECORD_SPECIAL)
- continue;
-
- unsigned align = (varyings[i].elements & 63);
-
- /* While we're at it, the SO buffers are linear */
-
- if (!align) {
- varyings[i].elements |= MALI_ATTR_LINEAR;
- continue;
- }
-
- /* We need to adjust alignment */
- varyings[i].elements &= ~63;
- varyings[i].elements |= MALI_ATTR_LINEAR;
- varyings[i].size += align;
-
- for (unsigned v = 0; v < vs->varying_count; ++v) {
- if (ovs[v].index != i)
- continue;
-
- ovs[v].src_offset = vs->varyings[v].src_offset + align;
- }
-
- for (unsigned f = 0; f < fs->varying_count; ++f) {
- if (ofs[f].index != i)
- continue;
-
- ofs[f].src_offset = fs->varyings[f].src_offset + align;
- }
- }
+ pan_emit_special_input(varyings, present, PAN_VARY_PNTCOORD, MALI_VARYING_POINT_COORD);
+ pan_emit_special_input(varyings, present, PAN_VARY_FACE, MALI_VARYING_FRONT_FACING);
+ pan_emit_special_input(varyings, present, PAN_VARY_FRAGCOORD, MALI_VARYING_FRAG_COORD);
- varyings_p = panfrost_upload_transient(batch, varyings,
- idx * sizeof(*varyings));
- vertex_postfix->varyings = varyings_p;
- tiler_postfix->varyings = varyings_p;
+ vertex_postfix->varyings = T.gpu;
+ tiler_postfix->varyings = T.gpu;
vertex_postfix->varying_meta = trans.gpu;
tiler_postfix->varying_meta = trans.gpu + vs_size;
0, 0,
};
- return panfrost_upload_transient(batch, locations, 96 * sizeof(uint16_t));
+ return panfrost_pool_upload(&batch->pool, locations, 96 * sizeof(uint16_t));
}