freedreno: Work around UBWC flakiness.

[mesa.git] / src / gallium / drivers / freedreno / freedreno_resource.c

/*
 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
 *
 * 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.
 *
 * Authors:
 *    Rob Clark <robclark@freedesktop.org>
 */

#include "util/format/u_format.h"
#include "util/format/u_format_rgtc.h"
#include "util/format/u_format_zs.h"
#include "util/u_inlines.h"
#include "util/u_transfer.h"
#include "util/u_string.h"
#include "util/u_surface.h"
#include "util/set.h"
#include "util/u_drm.h"

#include "freedreno_resource.h"
#include "freedreno_batch_cache.h"
#include "freedreno_blitter.h"
#include "freedreno_fence.h"
#include "freedreno_screen.h"
#include "freedreno_surface.h"
#include "freedreno_context.h"
#include "freedreno_query_hw.h"
#include "freedreno_util.h"

#include "drm-uapi/drm_fourcc.h"
#include <errno.h>

/* XXX this should go away, needed for 'struct winsys_handle' */
#include "state_tracker/drm_driver.h"

/* A private modifier for now, so we have a way to request tiled but not
 * compressed.  It would perhaps be good to get real modifiers for the
 * tiled formats, but would probably need to do some work to figure out
 * the layout(s) of the tiled modes, and whether they are the same
 * across generations.
 */
#define FD_FORMAT_MOD_QCOM_TILED        fourcc_mod_code(QCOM, 0xffffffff)

/**
 * Go through the entire state and see if the resource is bound
 * anywhere. If it is, mark the relevant state as dirty. This is
 * called on realloc_bo to ensure the neccessary state is re-
 * emitted so the GPU looks at the new backing bo.
 */
static void
rebind_resource(struct fd_context *ctx, struct pipe_resource *prsc)
{
        /* VBOs */
        for (unsigned i = 0; i < ctx->vtx.vertexbuf.count && !(ctx->dirty & FD_DIRTY_VTXBUF); i++) {
                if (ctx->vtx.vertexbuf.vb[i].buffer.resource == prsc)
                        ctx->dirty |= FD_DIRTY_VTXBUF;
        }

        /* per-shader-stage resources: */
        for (unsigned stage = 0; stage < PIPE_SHADER_TYPES; stage++) {
                /* Constbufs.. note that constbuf[0] is normal uniforms emitted in
                 * cmdstream rather than by pointer..
                 */
                const unsigned num_ubos = util_last_bit(ctx->constbuf[stage].enabled_mask);
                for (unsigned i = 1; i < num_ubos; i++) {
                        if (ctx->dirty_shader[stage] & FD_DIRTY_SHADER_CONST)
                                break;
                        if (ctx->constbuf[stage].cb[i].buffer == prsc)
                                ctx->dirty_shader[stage] |= FD_DIRTY_SHADER_CONST;
                }

                /* Textures */
                for (unsigned i = 0; i < ctx->tex[stage].num_textures; i++) {
                        if (ctx->dirty_shader[stage] & FD_DIRTY_SHADER_TEX)
                                break;
                        if (ctx->tex[stage].textures[i] && (ctx->tex[stage].textures[i]->texture == prsc))
                                ctx->dirty_shader[stage] |= FD_DIRTY_SHADER_TEX;
                }

                /* Images */
                const unsigned num_images = util_last_bit(ctx->shaderimg[stage].enabled_mask);
                for (unsigned i = 0; i < num_images; i++) {
                        if (ctx->dirty_shader[stage] & FD_DIRTY_SHADER_IMAGE)
                                break;
                        if (ctx->shaderimg[stage].si[i].resource == prsc)
                                ctx->dirty_shader[stage] |= FD_DIRTY_SHADER_IMAGE;
                }

                /* SSBOs */
                const unsigned num_ssbos = util_last_bit(ctx->shaderbuf[stage].enabled_mask);
                for (unsigned i = 0; i < num_ssbos; i++) {
                        if (ctx->dirty_shader[stage] & FD_DIRTY_SHADER_SSBO)
                                break;
                        if (ctx->shaderbuf[stage].sb[i].buffer == prsc)
                                ctx->dirty_shader[stage] |= FD_DIRTY_SHADER_SSBO;
                }
        }
}

static void
realloc_bo(struct fd_resource *rsc, uint32_t size)
{
        struct pipe_resource *prsc = &rsc->base;
        struct fd_screen *screen = fd_screen(rsc->base.screen);
        uint32_t flags = DRM_FREEDRENO_GEM_CACHE_WCOMBINE |
                        DRM_FREEDRENO_GEM_TYPE_KMEM |
                        COND(prsc->bind & PIPE_BIND_SCANOUT, DRM_FREEDRENO_GEM_SCANOUT);
                        /* TODO other flags? */

        /* if we start using things other than write-combine,
         * be sure to check for PIPE_RESOURCE_FLAG_MAP_COHERENT
         */

        if (rsc->bo)
                fd_bo_del(rsc->bo);

        rsc->bo = fd_bo_new(screen->dev, size, flags, "%ux%ux%u@%u:%x",
                        prsc->width0, prsc->height0, prsc->depth0, rsc->layout.cpp, prsc->bind);

        /* Zero out the UBWC area on allocation.  This fixes intermittent failures
         * with UBWC, which I suspect are due to the HW having a hard time
         * interpreting arbitrary values populating the flags buffer when the BO
         * was recycled through the bo cache (instead of fresh allocations from
         * the kernel, which are zeroed).  sleep(1) in this spot didn't work
         * around the issue, but any memset value seems to.
         */
        if (rsc->layout.ubwc) {
                void *buf = fd_bo_map(rsc->bo);
                memset(buf, 0, rsc->layout.slices[0].offset);
        }

        rsc->seqno = p_atomic_inc_return(&screen->rsc_seqno);
        util_range_set_empty(&rsc->valid_buffer_range);
        fd_bc_invalidate_resource(rsc, true);
}

static void
do_blit(struct fd_context *ctx, const struct pipe_blit_info *blit, bool fallback)
{
        struct pipe_context *pctx = &ctx->base;

        /* TODO size threshold too?? */
        if (fallback || !fd_blit(pctx, blit)) {
                /* do blit on cpu: */
                util_resource_copy_region(pctx,
                                blit->dst.resource, blit->dst.level, blit->dst.box.x,
                                blit->dst.box.y, blit->dst.box.z,
                                blit->src.resource, blit->src.level, &blit->src.box);
        }
}

/**
 * @rsc: the resource to shadow
 * @level: the level to discard (if box != NULL, otherwise ignored)
 * @box: the box to discard (or NULL if none)
 * @modifier: the modifier for the new buffer state
 */
static bool
fd_try_shadow_resource(struct fd_context *ctx, struct fd_resource *rsc,
                unsigned level, const struct pipe_box *box, uint64_t modifier)
{
        struct pipe_context *pctx = &ctx->base;
        struct pipe_resource *prsc = &rsc->base;
        bool fallback = false;

        if (prsc->next)
                return false;

        /* TODO: somehow munge dimensions and format to copy unsupported
         * render target format to something that is supported?
         */
        if (!pctx->screen->is_format_supported(pctx->screen,
                        prsc->format, prsc->target, prsc->nr_samples,
                        prsc->nr_storage_samples,
                        PIPE_BIND_RENDER_TARGET))
                fallback = true;

        /* do shadowing back-blits on the cpu for buffers: */
        if (prsc->target == PIPE_BUFFER)
                fallback = true;

        bool discard_whole_level = box && util_texrange_covers_whole_level(prsc, level,
                box->x, box->y, box->z, box->width, box->height, box->depth);

        /* TODO need to be more clever about current level */
        if ((prsc->target >= PIPE_TEXTURE_2D) && box && !discard_whole_level)
                return false;

        struct pipe_resource *pshadow =
                pctx->screen->resource_create_with_modifiers(pctx->screen,
                                prsc, &modifier, 1);

        if (!pshadow)
                return false;

        assert(!ctx->in_shadow);
        ctx->in_shadow = true;

        /* get rid of any references that batch-cache might have to us (which
         * should empty/destroy rsc->batches hashset)
         */
        fd_bc_invalidate_resource(rsc, false);

        mtx_lock(&ctx->screen->lock);

        /* Swap the backing bo's, so shadow becomes the old buffer,
         * blit from shadow to new buffer.  From here on out, we
         * cannot fail.
         *
         * Note that we need to do it in this order, otherwise if
         * we go down cpu blit path, the recursive transfer_map()
         * sees the wrong status..
         */
        struct fd_resource *shadow = fd_resource(pshadow);

        DBG("shadow: %p (%d) -> %p (%d)\n", rsc, rsc->base.reference.count,
                        shadow, shadow->base.reference.count);

        /* TODO valid_buffer_range?? */
        swap(rsc->bo,        shadow->bo);
        swap(rsc->write_batch,   shadow->write_batch);
        swap(rsc->layout, shadow->layout);
        rsc->seqno = p_atomic_inc_return(&ctx->screen->rsc_seqno);

        /* at this point, the newly created shadow buffer is not referenced
         * by any batches, but the existing rsc (probably) is.  We need to
         * transfer those references over:
         */
        debug_assert(shadow->batch_mask == 0);
        struct fd_batch *batch;
        foreach_batch(batch, &ctx->screen->batch_cache, rsc->batch_mask) {
                struct set_entry *entry = _mesa_set_search(batch->resources, rsc);
                _mesa_set_remove(batch->resources, entry);
                _mesa_set_add(batch->resources, shadow);
        }
        swap(rsc->batch_mask, shadow->batch_mask);

        mtx_unlock(&ctx->screen->lock);

        struct pipe_blit_info blit = {};
        blit.dst.resource = prsc;
        blit.dst.format   = prsc->format;
        blit.src.resource = pshadow;
        blit.src.format   = pshadow->format;
        blit.mask = util_format_get_mask(prsc->format);
        blit.filter = PIPE_TEX_FILTER_NEAREST;

#define set_box(field, val) do {     \
                blit.dst.field = (val);      \
                blit.src.field = (val);      \
        } while (0)

        /* blit the other levels in their entirety: */
        for (unsigned l = 0; l <= prsc->last_level; l++) {
                if (box && l == level)
                        continue;

                /* just blit whole level: */
                set_box(level, l);
                set_box(box.width,  u_minify(prsc->width0, l));
                set_box(box.height, u_minify(prsc->height0, l));
                set_box(box.depth,  u_minify(prsc->depth0, l));

                for (int i = 0; i < prsc->array_size; i++) {
                        set_box(box.z, i);
                        do_blit(ctx, &blit, fallback);
                }
        }

        /* deal w/ current level specially, since we might need to split
         * it up into a couple blits:
         */
        if (box && !discard_whole_level) {
                set_box(level, level);

                switch (prsc->target) {
                case PIPE_BUFFER:
                case PIPE_TEXTURE_1D:
                        set_box(box.y, 0);
                        set_box(box.z, 0);
                        set_box(box.height, 1);
                        set_box(box.depth, 1);

                        if (box->x > 0) {
                                set_box(box.x, 0);
                                set_box(box.width, box->x);

                                do_blit(ctx, &blit, fallback);
                        }
                        if ((box->x + box->width) < u_minify(prsc->width0, level)) {
                                set_box(box.x, box->x + box->width);
                                set_box(box.width, u_minify(prsc->width0, level) - (box->x + box->width));

                                do_blit(ctx, &blit, fallback);
                        }
                        break;
                case PIPE_TEXTURE_2D:
                        /* TODO */
                default:
                        unreachable("TODO");
                }
        }

        ctx->in_shadow = false;

        pipe_resource_reference(&pshadow, NULL);

        return true;
}

/**
 * Uncompress an UBWC compressed buffer "in place".  This works basically
 * like resource shadowing, creating a new resource, and doing an uncompress
 * blit, and swapping the state between shadow and original resource so it
 * appears to the state tracker as if nothing changed.
 */
void
fd_resource_uncompress(struct fd_context *ctx, struct fd_resource *rsc)
{
        bool success =
                fd_try_shadow_resource(ctx, rsc, 0, NULL, FD_FORMAT_MOD_QCOM_TILED);

        /* shadow should not fail in any cases where we need to uncompress: */
        debug_assert(success);

        /*
         * TODO what if rsc is used in other contexts, we don't currently
         * have a good way to rebind_resource() in other contexts.  And an
         * app that is reading one resource in multiple contexts, isn't
         * going to expect that the resource is modified.
         *
         * Hopefully the edge cases where we need to uncompress are rare
         * enough that they mostly only show up in deqp.
         */

        rebind_resource(ctx, &rsc->base);
}

static struct fd_resource *
fd_alloc_staging(struct fd_context *ctx, struct fd_resource *rsc,
                unsigned level, const struct pipe_box *box)
{
        struct pipe_context *pctx = &ctx->base;
        struct pipe_resource tmpl = rsc->base;

        tmpl.width0  = box->width;
        tmpl.height0 = box->height;
        /* for array textures, box->depth is the array_size, otherwise
         * for 3d textures, it is the depth:
         */
        if (tmpl.array_size > 1) {
                if (tmpl.target == PIPE_TEXTURE_CUBE)
                        tmpl.target = PIPE_TEXTURE_2D_ARRAY;
                tmpl.array_size = box->depth;
                tmpl.depth0 = 1;
        } else {
                tmpl.array_size = 1;
                tmpl.depth0 = box->depth;
        }
        tmpl.last_level = 0;
        tmpl.bind |= PIPE_BIND_LINEAR;

        struct pipe_resource *pstaging =
                pctx->screen->resource_create(pctx->screen, &tmpl);
        if (!pstaging)
                return NULL;

        return fd_resource(pstaging);
}

static void
fd_blit_from_staging(struct fd_context *ctx, struct fd_transfer *trans)
{
        struct pipe_resource *dst = trans->base.resource;
        struct pipe_blit_info blit = {};

        blit.dst.resource = dst;
        blit.dst.format   = dst->format;
        blit.dst.level    = trans->base.level;
        blit.dst.box      = trans->base.box;
        blit.src.resource = trans->staging_prsc;
        blit.src.format   = trans->staging_prsc->format;
        blit.src.level    = 0;
        blit.src.box      = trans->staging_box;
        blit.mask = util_format_get_mask(trans->staging_prsc->format);
        blit.filter = PIPE_TEX_FILTER_NEAREST;

        do_blit(ctx, &blit, false);
}

static void
fd_blit_to_staging(struct fd_context *ctx, struct fd_transfer *trans)
{
        struct pipe_resource *src = trans->base.resource;
        struct pipe_blit_info blit = {};

        blit.src.resource = src;
        blit.src.format   = src->format;
        blit.src.level    = trans->base.level;
        blit.src.box      = trans->base.box;
        blit.dst.resource = trans->staging_prsc;
        blit.dst.format   = trans->staging_prsc->format;
        blit.dst.level    = 0;
        blit.dst.box      = trans->staging_box;
        blit.mask = util_format_get_mask(trans->staging_prsc->format);
        blit.filter = PIPE_TEX_FILTER_NEAREST;

        do_blit(ctx, &blit, false);
}

static void fd_resource_transfer_flush_region(struct pipe_context *pctx,
                struct pipe_transfer *ptrans,
                const struct pipe_box *box)
{
        struct fd_resource *rsc = fd_resource(ptrans->resource);

        if (ptrans->resource->target == PIPE_BUFFER)
                util_range_add(&rsc->base, &rsc->valid_buffer_range,
                                           ptrans->box.x + box->x,
                                           ptrans->box.x + box->x + box->width);
}

static void
flush_resource(struct fd_context *ctx, struct fd_resource *rsc, unsigned usage)
{
        struct fd_batch *write_batch = NULL;

        mtx_lock(&ctx->screen->lock);
        fd_batch_reference_locked(&write_batch, rsc->write_batch);
        mtx_unlock(&ctx->screen->lock);

        if (usage & PIPE_TRANSFER_WRITE) {
                struct fd_batch *batch, *batches[32] = {};
                uint32_t batch_mask;

                /* This is a bit awkward, probably a fd_batch_flush_locked()
                 * would make things simpler.. but we need to hold the lock
                 * to iterate the batches which reference this resource.  So
                 * we must first grab references under a lock, then flush.
                 */
                mtx_lock(&ctx->screen->lock);
                batch_mask = rsc->batch_mask;
                foreach_batch(batch, &ctx->screen->batch_cache, batch_mask)
                        fd_batch_reference_locked(&batches[batch->idx], batch);
                mtx_unlock(&ctx->screen->lock);

                foreach_batch(batch, &ctx->screen->batch_cache, batch_mask)
                        fd_batch_flush(batch);

                foreach_batch(batch, &ctx->screen->batch_cache, batch_mask) {
                        fd_batch_reference(&batches[batch->idx], NULL);
                }
                assert(rsc->batch_mask == 0);
        } else if (write_batch) {
                fd_batch_flush(write_batch);
        }

        fd_batch_reference(&write_batch, NULL);

        assert(!rsc->write_batch);
}

static void
fd_flush_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
{
        flush_resource(fd_context(pctx), fd_resource(prsc), PIPE_TRANSFER_READ);
}

static void
fd_resource_transfer_unmap(struct pipe_context *pctx,
                struct pipe_transfer *ptrans)
{
        struct fd_context *ctx = fd_context(pctx);
        struct fd_resource *rsc = fd_resource(ptrans->resource);
        struct fd_transfer *trans = fd_transfer(ptrans);

        if (trans->staging_prsc) {
                if (ptrans->usage & PIPE_TRANSFER_WRITE)
                        fd_blit_from_staging(ctx, trans);
                pipe_resource_reference(&trans->staging_prsc, NULL);
        }

        if (!(ptrans->usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
                fd_bo_cpu_fini(rsc->bo);
        }

        util_range_add(&rsc->base, &rsc->valid_buffer_range,
                                   ptrans->box.x,
                                   ptrans->box.x + ptrans->box.width);

        pipe_resource_reference(&ptrans->resource, NULL);
        slab_free(&ctx->transfer_pool, ptrans);
}

static void *
fd_resource_transfer_map(struct pipe_context *pctx,
                struct pipe_resource *prsc,
                unsigned level, unsigned usage,
                const struct pipe_box *box,
                struct pipe_transfer **pptrans)
{
        struct fd_context *ctx = fd_context(pctx);
        struct fd_resource *rsc = fd_resource(prsc);
        struct fdl_slice *slice = fd_resource_slice(rsc, level);
        struct fd_transfer *trans;
        struct pipe_transfer *ptrans;
        enum pipe_format format = prsc->format;
        uint32_t op = 0;
        uint32_t offset;
        char *buf;
        int ret = 0;

        DBG("prsc=%p, level=%u, usage=%x, box=%dx%d+%d,%d", prsc, level, usage,
                box->width, box->height, box->x, box->y);

        ptrans = slab_alloc(&ctx->transfer_pool);
        if (!ptrans)
                return NULL;

        /* slab_alloc_st() doesn't zero: */
        trans = fd_transfer(ptrans);
        memset(trans, 0, sizeof(*trans));

        pipe_resource_reference(&ptrans->resource, prsc);
        ptrans->level = level;
        ptrans->usage = usage;
        ptrans->box = *box;
        ptrans->stride = util_format_get_nblocksx(format, slice->pitch) * rsc->layout.cpp;
        ptrans->layer_stride = fd_resource_layer_stride(rsc, level);

        /* we always need a staging texture for tiled buffers:
         *
         * TODO we might sometimes want to *also* shadow the resource to avoid
         * splitting a batch.. for ex, mid-frame texture uploads to a tiled
         * texture.
         */
        if (rsc->layout.tile_mode) {
                struct fd_resource *staging_rsc;

                staging_rsc = fd_alloc_staging(ctx, rsc, level, box);
                if (staging_rsc) {
                        struct fdl_slice *staging_slice =
                                fd_resource_slice(staging_rsc, 0);
                        // TODO for PIPE_TRANSFER_READ, need to do untiling blit..
                        trans->staging_prsc = &staging_rsc->base;
                        trans->base.stride = util_format_get_nblocksx(format,
                                staging_slice->pitch) * staging_rsc->layout.cpp;
                        trans->base.layer_stride = fd_resource_layer_stride(staging_rsc, 0);
                        trans->staging_box = *box;
                        trans->staging_box.x = 0;
                        trans->staging_box.y = 0;
                        trans->staging_box.z = 0;

                        if (usage & PIPE_TRANSFER_READ) {
                                fd_blit_to_staging(ctx, trans);

                                fd_bo_cpu_prep(staging_rsc->bo, ctx->pipe,
                                                DRM_FREEDRENO_PREP_READ);
                        }

                        buf = fd_bo_map(staging_rsc->bo);
                        offset = 0;

                        *pptrans = ptrans;

                        ctx->stats.staging_uploads++;

                        return buf;
                }
        }

        if (ctx->in_shadow && !(usage & PIPE_TRANSFER_READ))
                usage |= PIPE_TRANSFER_UNSYNCHRONIZED;

        if (usage & PIPE_TRANSFER_READ)
                op |= DRM_FREEDRENO_PREP_READ;

        if (usage & PIPE_TRANSFER_WRITE)
                op |= DRM_FREEDRENO_PREP_WRITE;

        if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) {
                realloc_bo(rsc, fd_bo_size(rsc->bo));
                rebind_resource(ctx, prsc);
        } else if ((usage & PIPE_TRANSFER_WRITE) &&
                           prsc->target == PIPE_BUFFER &&
                           !util_ranges_intersect(&rsc->valid_buffer_range,
                                                                          box->x, box->x + box->width)) {
                /* We are trying to write to a previously uninitialized range. No need
                 * to wait.
                 */
        } else if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
                struct fd_batch *write_batch = NULL;

                /* hold a reference, so it doesn't disappear under us: */
                fd_context_lock(ctx);
                fd_batch_reference_locked(&write_batch, rsc->write_batch);
                fd_context_unlock(ctx);

                if ((usage & PIPE_TRANSFER_WRITE) && write_batch &&
                                write_batch->back_blit) {
                        /* if only thing pending is a back-blit, we can discard it: */
                        fd_batch_reset(write_batch);
                }

                /* If the GPU is writing to the resource, or if it is reading from the
                 * resource and we're trying to write to it, flush the renders.
                 */
                bool needs_flush = pending(rsc, !!(usage & PIPE_TRANSFER_WRITE));
                bool busy = needs_flush || (0 != fd_bo_cpu_prep(rsc->bo,
                                ctx->pipe, op | DRM_FREEDRENO_PREP_NOSYNC));

                /* if we need to flush/stall, see if we can make a shadow buffer
                 * to avoid this:
                 *
                 * TODO we could go down this path !reorder && !busy_for_read
                 * ie. we only *don't* want to go down this path if the blit
                 * will trigger a flush!
                 */
                if (ctx->screen->reorder && busy && !(usage & PIPE_TRANSFER_READ) &&
                                (usage & PIPE_TRANSFER_DISCARD_RANGE)) {
                        /* try shadowing only if it avoids a flush, otherwise staging would
                         * be better:
                         */
                        if (needs_flush && fd_try_shadow_resource(ctx, rsc, level,
                                                        box, DRM_FORMAT_MOD_LINEAR)) {
                                needs_flush = busy = false;
                                rebind_resource(ctx, prsc);
                                ctx->stats.shadow_uploads++;
                        } else {
                                struct fd_resource *staging_rsc;

                                if (needs_flush) {
                                        flush_resource(ctx, rsc, usage);
                                        needs_flush = false;
                                }

                                /* in this case, we don't need to shadow the whole resource,
                                 * since any draw that references the previous contents has
                                 * already had rendering flushed for all tiles.  So we can
                                 * use a staging buffer to do the upload.
                                 */
                                staging_rsc = fd_alloc_staging(ctx, rsc, level, box);
                                if (staging_rsc) {
                                        struct fdl_slice *staging_slice =
                                                fd_resource_slice(staging_rsc, 0);
                                        trans->staging_prsc = &staging_rsc->base;
                                        trans->base.stride = util_format_get_nblocksx(format,
                                                staging_slice->pitch) * staging_rsc->layout.cpp;
                                        trans->base.layer_stride =
                                                fd_resource_layer_stride(staging_rsc, 0);
                                        trans->staging_box = *box;
                                        trans->staging_box.x = 0;
                                        trans->staging_box.y = 0;
                                        trans->staging_box.z = 0;
                                        buf = fd_bo_map(staging_rsc->bo);
                                        offset = 0;

                                        *pptrans = ptrans;

                                        fd_batch_reference(&write_batch, NULL);

                                        ctx->stats.staging_uploads++;

                                        return buf;
                                }
                        }
                }

                if (needs_flush) {
                        flush_resource(ctx, rsc, usage);
                        needs_flush = false;
                }

                fd_batch_reference(&write_batch, NULL);

                /* The GPU keeps track of how the various bo's are being used, and
                 * will wait if necessary for the proper operation to have
                 * completed.
                 */
                if (busy) {
                        ret = fd_bo_cpu_prep(rsc->bo, ctx->pipe, op);
                        if (ret)
                                goto fail;
                }
        }

        buf = fd_bo_map(rsc->bo);
        offset =
                box->y / util_format_get_blockheight(format) * ptrans->stride +
                box->x / util_format_get_blockwidth(format) * rsc->layout.cpp +
                fd_resource_offset(rsc, level, box->z);

        if (usage & PIPE_TRANSFER_WRITE)
                rsc->valid = true;

        *pptrans = ptrans;

        return buf + offset;

fail:
        fd_resource_transfer_unmap(pctx, ptrans);
        return NULL;
}

static void
fd_resource_destroy(struct pipe_screen *pscreen,
                struct pipe_resource *prsc)
{
        struct fd_resource *rsc = fd_resource(prsc);
        fd_bc_invalidate_resource(rsc, true);
        if (rsc->bo)
                fd_bo_del(rsc->bo);
        if (rsc->scanout)
                renderonly_scanout_destroy(rsc->scanout, fd_screen(pscreen)->ro);

        util_range_destroy(&rsc->valid_buffer_range);
        FREE(rsc);
}

static uint64_t
fd_resource_modifier(struct fd_resource *rsc)
{
        if (!rsc->layout.tile_mode)
                return DRM_FORMAT_MOD_LINEAR;

        if (rsc->layout.ubwc_layer_size)
                return DRM_FORMAT_MOD_QCOM_COMPRESSED;

        /* TODO invent a modifier for tiled but not UBWC buffers: */
        return DRM_FORMAT_MOD_INVALID;
}

static bool
fd_resource_get_handle(struct pipe_screen *pscreen,
                struct pipe_context *pctx,
                struct pipe_resource *prsc,
                struct winsys_handle *handle,
                unsigned usage)
{
        struct fd_resource *rsc = fd_resource(prsc);

        handle->modifier = fd_resource_modifier(rsc);

        return fd_screen_bo_get_handle(pscreen, rsc->bo, rsc->scanout,
                        fd_resource_slice(rsc, 0)->pitch * rsc->layout.cpp, handle);
}

static uint32_t
setup_slices(struct fd_resource *rsc, uint32_t alignment, enum pipe_format format)
{
        struct pipe_resource *prsc = &rsc->base;
        struct fd_screen *screen = fd_screen(prsc->screen);
        enum util_format_layout layout = util_format_description(format)->layout;
        uint32_t pitchalign = screen->gmem_alignw;
        uint32_t level, size = 0;
        uint32_t width = prsc->width0;
        uint32_t height = prsc->height0;
        uint32_t depth = prsc->depth0;
        /* in layer_first layout, the level (slice) contains just one
         * layer (since in fact the layer contains the slices)
         */
        uint32_t layers_in_level = rsc->layout.layer_first ? 1 : prsc->array_size;

        for (level = 0; level <= prsc->last_level; level++) {
                struct fdl_slice *slice = fd_resource_slice(rsc, level);
                uint32_t blocks;

                if (layout == UTIL_FORMAT_LAYOUT_ASTC)
                        slice->pitch = width =
                                util_align_npot(width, pitchalign * util_format_get_blockwidth(format));
                else
                        slice->pitch = width = align(width, pitchalign);
                slice->offset = size;
                blocks = util_format_get_nblocks(format, width, height);
                /* 1d array and 2d array textures must all have the same layer size
                 * for each miplevel on a3xx. 3d textures can have different layer
                 * sizes for high levels, but the hw auto-sizer is buggy (or at least
                 * different than what this code does), so as soon as the layer size
                 * range gets into range, we stop reducing it.
                 */
                if (prsc->target == PIPE_TEXTURE_3D && (
                                        level == 1 ||
                                        (level > 1 && fd_resource_slice(rsc, level - 1)->size0 > 0xf000)))
                        slice->size0 = align(blocks * rsc->layout.cpp, alignment);
                else if (level == 0 || rsc->layout.layer_first || alignment == 1)
                        slice->size0 = align(blocks * rsc->layout.cpp, alignment);
                else
                        slice->size0 = fd_resource_slice(rsc, level - 1)->size0;

                size += slice->size0 * depth * layers_in_level;

                width = u_minify(width, 1);
                height = u_minify(height, 1);
                depth = u_minify(depth, 1);
        }

        return size;
}

static uint32_t
slice_alignment(enum pipe_texture_target target)
{
        /* on a3xx, 2d array and 3d textures seem to want their
         * layers aligned to page boundaries:
         */
        switch (target) {
        case PIPE_TEXTURE_3D:
        case PIPE_TEXTURE_1D_ARRAY:
        case PIPE_TEXTURE_2D_ARRAY:
                return 4096;
        default:
                return 1;
        }
}

/* cross generation texture layout to plug in to screen->setup_slices()..
 * replace with generation specific one as-needed.
 *
 * TODO for a4xx probably can extract out the a4xx specific logic int
 * a small fd4_setup_slices() wrapper that sets up layer_first, and then
 * calls this.
 */
uint32_t
fd_setup_slices(struct fd_resource *rsc)
{
        uint32_t alignment;

        alignment = slice_alignment(rsc->base.target);

        struct fd_screen *screen = fd_screen(rsc->base.screen);
        if (is_a4xx(screen)) {
                switch (rsc->base.target) {
                case PIPE_TEXTURE_3D:
                        rsc->layout.layer_first = false;
                        break;
                default:
                        rsc->layout.layer_first = true;
                        alignment = 1;
                        break;
                }
        }

        return setup_slices(rsc, alignment, rsc->base.format);
}

/* special case to resize query buf after allocated.. */
void
fd_resource_resize(struct pipe_resource *prsc, uint32_t sz)
{
        struct fd_resource *rsc = fd_resource(prsc);

        debug_assert(prsc->width0 == 0);
        debug_assert(prsc->target == PIPE_BUFFER);
        debug_assert(prsc->bind == PIPE_BIND_QUERY_BUFFER);

        prsc->width0 = sz;
        realloc_bo(rsc, fd_screen(prsc->screen)->setup_slices(rsc));
}

static void
fd_resource_layout_init(struct pipe_resource *prsc)
{
        struct fd_resource *rsc = fd_resource(prsc);
        struct fdl_layout *layout = &rsc->layout;

        layout->width0 = prsc->width0;
        layout->height0 = prsc->height0;
        layout->depth0 = prsc->depth0;

        layout->cpp = util_format_get_blocksize(prsc->format);
        layout->cpp *= fd_resource_nr_samples(prsc);
}

/**
 * Create a new texture object, using the given template info.
 */
static struct pipe_resource *
fd_resource_create_with_modifiers(struct pipe_screen *pscreen,
                const struct pipe_resource *tmpl,
                const uint64_t *modifiers, int count)
{
        struct fd_screen *screen = fd_screen(pscreen);
        struct fd_resource *rsc;
        struct pipe_resource *prsc;
        enum pipe_format format = tmpl->format;
        uint32_t size;

        /* when using kmsro, scanout buffers are allocated on the display device
         * create_with_modifiers() doesn't give us usage flags, so we have to
         * assume that all calls with modifiers are scanout-possible
         */
        if (screen->ro &&
                ((tmpl->bind & PIPE_BIND_SCANOUT) ||
                 !(count == 1 && modifiers[0] == DRM_FORMAT_MOD_INVALID))) {
                struct pipe_resource scanout_templat = *tmpl;
                struct renderonly_scanout *scanout;
                struct winsys_handle handle;

                /* apply freedreno alignment requirement */
                scanout_templat.width0 = align(tmpl->width0, screen->gmem_alignw);

                scanout = renderonly_scanout_for_resource(&scanout_templat,
                                                                                                  screen->ro, &handle);
                if (!scanout)
                        return NULL;

                renderonly_scanout_destroy(scanout, screen->ro);

                assert(handle.type == WINSYS_HANDLE_TYPE_FD);
                rsc = fd_resource(pscreen->resource_from_handle(pscreen, tmpl,
                                                                                                                &handle,
                                                                                                                PIPE_HANDLE_USAGE_FRAMEBUFFER_WRITE));
                close(handle.handle);
                if (!rsc)
                        return NULL;

                return &rsc->base;
        }

        rsc = CALLOC_STRUCT(fd_resource);
        prsc = &rsc->base;

        DBG("%p: target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
                        "nr_samples=%u, usage=%u, bind=%x, flags=%x", prsc,
                        tmpl->target, util_format_name(format),
                        tmpl->width0, tmpl->height0, tmpl->depth0,
                        tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
                        tmpl->usage, tmpl->bind, tmpl->flags);

        if (!rsc)
                return NULL;

        *prsc = *tmpl;
        fd_resource_layout_init(prsc);

#define LINEAR \
        (PIPE_BIND_SCANOUT | \
         PIPE_BIND_LINEAR  | \
         PIPE_BIND_DISPLAY_TARGET)

        bool linear = drm_find_modifier(DRM_FORMAT_MOD_LINEAR, modifiers, count);
        if (tmpl->bind & LINEAR)
                linear = true;

        if (fd_mesa_debug & FD_DBG_NOTILE)
                linear = true;

        /* Normally, for non-shared buffers, allow buffer compression if
         * not shared, otherwise only allow if QCOM_COMPRESSED modifier
         * is requested:
         *
         * TODO we should probably also limit tiled in a similar way,
         * except we don't have a format modifier for tiled.  (We probably
         * should.)
         */
        bool allow_ubwc = drm_find_modifier(DRM_FORMAT_MOD_INVALID, modifiers, count);
        if (tmpl->bind & PIPE_BIND_SHARED)
                allow_ubwc = drm_find_modifier(DRM_FORMAT_MOD_QCOM_COMPRESSED, modifiers, count);

        allow_ubwc &= !(fd_mesa_debug & FD_DBG_NOUBWC);

        pipe_reference_init(&prsc->reference, 1);

        prsc->screen = pscreen;

        if (screen->tile_mode &&
                        (tmpl->target != PIPE_BUFFER) &&
                        !linear) {
                rsc->layout.tile_mode = screen->tile_mode(prsc);
        }

        util_range_init(&rsc->valid_buffer_range);

        rsc->internal_format = format;

        rsc->layout.ubwc = rsc->layout.tile_mode && is_a6xx(screen) && allow_ubwc;

        if (prsc->target == PIPE_BUFFER) {
                assert(prsc->format == PIPE_FORMAT_R8_UNORM);
                size = prsc->width0;
                fdl_layout_buffer(&rsc->layout, size);
        } else {
                size = screen->setup_slices(rsc);
        }

        /* special case for hw-query buffer, which we need to allocate before we
         * know the size:
         */
        if (size == 0) {
                /* note, semi-intention == instead of & */
                debug_assert(prsc->bind == PIPE_BIND_QUERY_BUFFER);
                return prsc;
        }

        /* Set the layer size if the (non-a6xx) backend hasn't done so. */
        if (rsc->layout.layer_first && !rsc->layout.layer_size) {
                rsc->layout.layer_size = align(size, 4096);
                size = rsc->layout.layer_size * prsc->array_size;
        }

        if (fd_mesa_debug & FD_DBG_LAYOUT)
                fdl_dump_layout(&rsc->layout);

        realloc_bo(rsc, size);
        if (!rsc->bo)
                goto fail;

        return prsc;
fail:
        fd_resource_destroy(pscreen, prsc);
        return NULL;
}

static struct pipe_resource *
fd_resource_create(struct pipe_screen *pscreen,
                const struct pipe_resource *tmpl)
{
        const uint64_t mod = DRM_FORMAT_MOD_INVALID;
        return fd_resource_create_with_modifiers(pscreen, tmpl, &mod, 1);
}

/**
 * Create a texture from a winsys_handle. The handle is often created in
 * another process by first creating a pipe texture and then calling
 * resource_get_handle.
 */
static struct pipe_resource *
fd_resource_from_handle(struct pipe_screen *pscreen,
                const struct pipe_resource *tmpl,
                struct winsys_handle *handle, unsigned usage)
{
        struct fd_screen *screen = fd_screen(pscreen);
        struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
        struct fdl_slice *slice = fd_resource_slice(rsc, 0);
        struct pipe_resource *prsc = &rsc->base;
        uint32_t pitchalign = fd_screen(pscreen)->gmem_alignw;

        DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
                        "nr_samples=%u, usage=%u, bind=%x, flags=%x",
                        tmpl->target, util_format_name(tmpl->format),
                        tmpl->width0, tmpl->height0, tmpl->depth0,
                        tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
                        tmpl->usage, tmpl->bind, tmpl->flags);

        if (!rsc)
                return NULL;

        *prsc = *tmpl;
        fd_resource_layout_init(prsc);

        pipe_reference_init(&prsc->reference, 1);

        prsc->screen = pscreen;

        util_range_init(&rsc->valid_buffer_range);

        rsc->bo = fd_screen_bo_from_handle(pscreen, handle);
        if (!rsc->bo)
                goto fail;

        rsc->internal_format = tmpl->format;
        slice->pitch = handle->stride / rsc->layout.cpp;
        slice->offset = handle->offset;
        slice->size0 = handle->stride * prsc->height0;

        if ((slice->pitch < align(prsc->width0, pitchalign)) ||
                        (slice->pitch & (pitchalign - 1)))
                goto fail;

        assert(rsc->layout.cpp);

        if (screen->layout_resource_for_modifier(rsc, handle->modifier) < 0)
                goto fail;

        if (screen->ro) {
                rsc->scanout =
                        renderonly_create_gpu_import_for_resource(prsc, screen->ro, NULL);
                /* failure is expected in some cases.. */
        }

        rsc->valid = true;

        return prsc;

fail:
        fd_resource_destroy(pscreen, prsc);
        return NULL;
}

bool
fd_render_condition_check(struct pipe_context *pctx)
{
        struct fd_context *ctx = fd_context(pctx);

        if (!ctx->cond_query)
                return true;

        union pipe_query_result res = { 0 };
        bool wait =
                ctx->cond_mode != PIPE_RENDER_COND_NO_WAIT &&
                ctx->cond_mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;

        if (pctx->get_query_result(pctx, ctx->cond_query, wait, &res))
                        return (bool)res.u64 != ctx->cond_cond;

        return true;
}

static void
fd_invalidate_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
{
        struct fd_context *ctx = fd_context(pctx);
        struct fd_resource *rsc = fd_resource(prsc);

        /*
         * TODO I guess we could track that the resource is invalidated and
         * use that as a hint to realloc rather than stall in _transfer_map(),
         * even in the non-DISCARD_WHOLE_RESOURCE case?
         *
         * Note: we set dirty bits to trigger invalidate logic fd_draw_vbo
         */

        if (rsc->write_batch) {
                struct fd_batch *batch = rsc->write_batch;
                struct pipe_framebuffer_state *pfb = &batch->framebuffer;

                if (pfb->zsbuf && pfb->zsbuf->texture == prsc) {
                        batch->resolve &= ~(FD_BUFFER_DEPTH | FD_BUFFER_STENCIL);
                        ctx->dirty |= FD_DIRTY_ZSA;
                }

                for (unsigned i = 0; i < pfb->nr_cbufs; i++) {
                        if (pfb->cbufs[i] && pfb->cbufs[i]->texture == prsc) {
                                batch->resolve &= ~(PIPE_CLEAR_COLOR0 << i);
                                ctx->dirty |= FD_DIRTY_FRAMEBUFFER;
                        }
                }
        }

        rsc->valid = false;
}

static enum pipe_format
fd_resource_get_internal_format(struct pipe_resource *prsc)
{
        return fd_resource(prsc)->internal_format;
}

static void
fd_resource_set_stencil(struct pipe_resource *prsc,
                struct pipe_resource *stencil)
{
        fd_resource(prsc)->stencil = fd_resource(stencil);
}

static struct pipe_resource *
fd_resource_get_stencil(struct pipe_resource *prsc)
{
        struct fd_resource *rsc = fd_resource(prsc);
        if (rsc->stencil)
                return &rsc->stencil->base;
        return NULL;
}

static const struct u_transfer_vtbl transfer_vtbl = {
                .resource_create          = fd_resource_create,
                .resource_destroy         = fd_resource_destroy,
                .transfer_map             = fd_resource_transfer_map,
                .transfer_flush_region    = fd_resource_transfer_flush_region,
                .transfer_unmap           = fd_resource_transfer_unmap,
                .get_internal_format      = fd_resource_get_internal_format,
                .set_stencil              = fd_resource_set_stencil,
                .get_stencil              = fd_resource_get_stencil,
};

static const uint64_t supported_modifiers[] = {
        DRM_FORMAT_MOD_LINEAR,
};

static int
fd_layout_resource_for_modifier(struct fd_resource *rsc, uint64_t modifier)
{
        switch (modifier) {
        case DRM_FORMAT_MOD_LINEAR:
                return 0;
        default:
                return -1;
        }
}

void
fd_resource_screen_init(struct pipe_screen *pscreen)
{
        struct fd_screen *screen = fd_screen(pscreen);
        bool fake_rgtc = screen->gpu_id < 400;

        pscreen->resource_create = u_transfer_helper_resource_create;
        /* NOTE: u_transfer_helper does not yet support the _with_modifiers()
         * variant:
         */
        pscreen->resource_create_with_modifiers = fd_resource_create_with_modifiers;
        pscreen->resource_from_handle = fd_resource_from_handle;
        pscreen->resource_get_handle = fd_resource_get_handle;
        pscreen->resource_destroy = u_transfer_helper_resource_destroy;

        pscreen->transfer_helper = u_transfer_helper_create(&transfer_vtbl,
                        true, false, fake_rgtc, true);

        if (!screen->setup_slices)
                screen->setup_slices = fd_setup_slices;
        if (!screen->layout_resource_for_modifier)
                screen->layout_resource_for_modifier = fd_layout_resource_for_modifier;
        if (!screen->supported_modifiers) {
                screen->supported_modifiers = supported_modifiers;
                screen->num_supported_modifiers = ARRAY_SIZE(supported_modifiers);
        }
}

static void
fd_get_sample_position(struct pipe_context *context,
                         unsigned sample_count, unsigned sample_index,
                         float *pos_out)
{
        /* The following is copied from nouveau/nv50 except for position
         * values, which are taken from blob driver */
        static const uint8_t pos1[1][2] = { { 0x8, 0x8 } };
        static const uint8_t pos2[2][2] = {
                { 0xc, 0xc }, { 0x4, 0x4 } };
        static const uint8_t pos4[4][2] = {
                { 0x6, 0x2 }, { 0xe, 0x6 },
                { 0x2, 0xa }, { 0xa, 0xe } };
        /* TODO needs to be verified on supported hw */
        static const uint8_t pos8[8][2] = {
                { 0x9, 0x5 }, { 0x7, 0xb },
                { 0xd, 0x9 }, { 0x5, 0x3 },
                { 0x3, 0xd }, { 0x1, 0x7 },
                { 0xb, 0xf }, { 0xf, 0x1 } };

        const uint8_t (*ptr)[2];

        switch (sample_count) {
        case 1:
                ptr = pos1;
                break;
        case 2:
                ptr = pos2;
                break;
        case 4:
                ptr = pos4;
                break;
        case 8:
                ptr = pos8;
                break;
        default:
                assert(0);
                return;
        }

        pos_out[0] = ptr[sample_index][0] / 16.0f;
        pos_out[1] = ptr[sample_index][1] / 16.0f;
}

static void
fd_blit_pipe(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
{
        /* wrap fd_blit to return void */
        fd_blit(pctx, blit_info);
}

void
fd_resource_context_init(struct pipe_context *pctx)
{
        pctx->transfer_map = u_transfer_helper_transfer_map;
        pctx->transfer_flush_region = u_transfer_helper_transfer_flush_region;
        pctx->transfer_unmap = u_transfer_helper_transfer_unmap;
        pctx->buffer_subdata = u_default_buffer_subdata;
        pctx->texture_subdata = u_default_texture_subdata;
        pctx->create_surface = fd_create_surface;
        pctx->surface_destroy = fd_surface_destroy;
        pctx->resource_copy_region = fd_resource_copy_region;
        pctx->blit = fd_blit_pipe;
        pctx->flush_resource = fd_flush_resource;
        pctx->invalidate_resource = fd_invalidate_resource;
        pctx->get_sample_position = fd_get_sample_position;
}