static void
fd_resource_flush_z32s8(struct fd_transfer *trans, const struct pipe_box *box)
{
struct fd_resource *rsc = fd_resource(trans->base.resource);
struct fd_resource_slice *slice = fd_resource_slice(rsc, trans->base.level);
struct fd_resource_slice *sslice = fd_resource_slice(rsc->stencil, trans->base.level);
enum pipe_format format = trans->base.resource->format;
float *depth = fd_bo_map(rsc->bo) + slice->offset +
fd_resource_layer_offset(rsc, slice, trans->base.box.z) +
(trans->base.box.y + box->y) * slice->pitch * 4 + (trans->base.box.x + box->x) * 4;
uint8_t *stencil = fd_bo_map(rsc->stencil->bo) + sslice->offset +
fd_resource_layer_offset(rsc->stencil, sslice, trans->base.box.z) +
(trans->base.box.y + box->y) * sslice->pitch + trans->base.box.x + box->x;
if (format != PIPE_FORMAT_X32_S8X24_UINT)
util_format_z32_float_s8x24_uint_unpack_z_float(
depth, slice->pitch * 4,
trans->staging, trans->base.stride,
box->width, box->height);
util_format_z32_float_s8x24_uint_unpack_s_8uint(
stencil, sslice->pitch,
trans->staging, trans->base.stride,
box->width, box->height);
}
static void
emit_gmem2mem_surf(struct fd_batch *batch, uint32_t base,
struct pipe_surface *psurf, enum a5xx_blit_buf buf)
{
struct fd_ringbuffer *ring = batch->gmem;
struct fd_resource *rsc = fd_resource(psurf->texture);
struct fd_resource_slice *slice;
uint32_t offset;
slice = fd_resource_slice(rsc, psurf->u.tex.level);
offset = fd_resource_offset(rsc, psurf->u.tex.level,
psurf->u.tex.first_layer);
debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
OUT_PKT4(ring, REG_A5XX_RB_BLIT_FLAG_DST_LO, 4);
OUT_RING(ring, 0x00000000); /* RB_BLIT_FLAG_DST_LO */
OUT_RING(ring, 0x00000000); /* RB_BLIT_FLAG_DST_HI */
OUT_RING(ring, 0x00000000); /* RB_BLIT_FLAG_DST_PITCH */
OUT_RING(ring, 0x00000000); /* RB_BLIT_FLAG_DST_ARRAY_PITCH */
OUT_PKT4(ring, REG_A5XX_RB_RESOLVE_CNTL_3, 5);
OUT_RING(ring, 0x00000004); /* XXX RB_RESOLVE_CNTL_3 */
OUT_RELOCW(ring, rsc->bo, offset, 0, 0); /* RB_BLIT_DST_LO/HI */
OUT_RING(ring, A5XX_RB_BLIT_DST_PITCH(slice->pitch * rsc->cpp));
OUT_RING(ring, A5XX_RB_BLIT_DST_ARRAY_PITCH(slice->size0));
OUT_PKT4(ring, REG_A5XX_RB_BLIT_CNTL, 1);
OUT_RING(ring, A5XX_RB_BLIT_CNTL_BUF(buf));
fd5_emit_blit(batch->ctx, ring);
}
/* 2d array and 3d textures seem to want their layers aligned to
* page boundaries
*/
static uint32_t
setup_slices_array(struct fd_resource *rsc)
{
struct pipe_resource *prsc = &rsc->base.b;
uint32_t level, size = 0;
uint32_t width = prsc->width0;
uint32_t height = prsc->height0;
uint32_t depth = prsc->depth0;
for (level = 0; level <= prsc->last_level; level++) {
struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
slice->pitch = align(width, 32);
slice->offset = size;
slice->size0 = align(slice->pitch * height * rsc->cpp, 4096);
size += slice->size0 * depth * prsc->array_size;
width = u_minify(width, 1);
height = u_minify(height, 1);
depth = u_minify(depth, 1);
}
return size;
}
static void
emit_gmem2mem_surf(struct fd_batch *batch, uint32_t base,
struct pipe_surface *psurf)
{
struct fd_ringbuffer *ring = batch->gmem;
struct fd_resource *rsc = fd_resource(psurf->texture);
uint32_t swap = fmt2swap(psurf->format);
struct fd_resource_slice *slice =
fd_resource_slice(rsc, psurf->u.tex.level);
uint32_t offset =
fd_resource_offset(rsc, psurf->u.tex.level, psurf->u.tex.first_layer);
assert((slice->pitch & 31) == 0);
assert((offset & 0xfff) == 0);
if (!rsc->valid)
return;
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COLOR_INFO));
OUT_RING(ring, A2XX_RB_COLOR_INFO_SWAP(swap) |
A2XX_RB_COLOR_INFO_BASE(base) |
A2XX_RB_COLOR_INFO_FORMAT(fd2_pipe2color(psurf->format)));
OUT_PKT3(ring, CP_SET_CONSTANT, 5);
OUT_RING(ring, CP_REG(REG_A2XX_RB_COPY_CONTROL));
OUT_RING(ring, 0x00000000); /* RB_COPY_CONTROL */
OUT_RELOCW(ring, rsc->bo, offset, 0, 0); /* RB_COPY_DEST_BASE */
OUT_RING(ring, slice->pitch >> 5); /* RB_COPY_DEST_PITCH */
OUT_RING(ring, /* RB_COPY_DEST_INFO */
A2XX_RB_COPY_DEST_INFO_FORMAT(fd2_pipe2color(psurf->format)) |
A2XX_RB_COPY_DEST_INFO_LINEAR |
A2XX_RB_COPY_DEST_INFO_SWAP(swap) |
A2XX_RB_COPY_DEST_INFO_WRITE_RED |
A2XX_RB_COPY_DEST_INFO_WRITE_GREEN |
A2XX_RB_COPY_DEST_INFO_WRITE_BLUE |
A2XX_RB_COPY_DEST_INFO_WRITE_ALPHA);
if (!is_a20x(batch->ctx->screen)) {
OUT_WFI (ring);
OUT_PKT3(ring, CP_SET_CONSTANT, 3);
OUT_RING(ring, CP_REG(REG_A2XX_VGT_MAX_VTX_INDX));
OUT_RING(ring, 3); /* VGT_MAX_VTX_INDX */
OUT_RING(ring, 0); /* VGT_MIN_VTX_INDX */
}
fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
DI_SRC_SEL_AUTO_INDEX, 3, 0, INDEX_SIZE_IGN, 0, 0, NULL);
}
static uint32_t
setup_slices(struct fd_resource *rsc, uint32_t alignment, enum pipe_format format)
{
struct pipe_resource *prsc = &rsc->base.b;
enum util_format_layout layout = util_format_description(format)->layout;
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->layer_first ? 1 : prsc->array_size;
for (level = 0; level <= prsc->last_level; level++) {
struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
uint32_t blocks;
if (layout == UTIL_FORMAT_LAYOUT_ASTC)
slice->pitch = width =
util_align_npot(width, 32 * util_format_get_blockwidth(format));
else
slice->pitch = width = align(width, 32);
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 && rsc->slices[level - 1].size0 > 0xf000)))
slice->size0 = align(blocks * rsc->cpp, alignment);
else if (level == 0 || rsc->layer_first || alignment == 1)
slice->size0 = align(blocks * rsc->cpp, alignment);
else
slice->size0 = rsc->slices[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;
}
/* emit texture state for mem->gmem restore operation.. eventually it would
* be good to get rid of this and use normal CSO/etc state for more of these
* special cases..
*/
void
fd4_emit_gmem_restore_tex(struct fd_ringbuffer *ring, struct pipe_surface *psurf)
{
struct fd_resource *rsc = fd_resource(psurf->texture);
unsigned lvl = psurf->u.tex.level;
struct fd_resource_slice *slice = fd_resource_slice(rsc, lvl);
uint32_t offset = fd_resource_offset(rsc, lvl, psurf->u.tex.first_layer);
enum pipe_format format = fd4_gmem_restore_format(psurf->format);
debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
/* output sampler state: */
OUT_PKT3(ring, CP_LOAD_STATE, 4);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
CP_LOAD_STATE_0_NUM_UNIT(1));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
OUT_RING(ring, A4XX_TEX_SAMP_0_XY_MAG(A4XX_TEX_NEAREST) |
A4XX_TEX_SAMP_0_XY_MIN(A4XX_TEX_NEAREST) |
A4XX_TEX_SAMP_0_WRAP_S(A4XX_TEX_CLAMP_TO_EDGE) |
A4XX_TEX_SAMP_0_WRAP_T(A4XX_TEX_CLAMP_TO_EDGE) |
A4XX_TEX_SAMP_0_WRAP_R(A4XX_TEX_REPEAT));
OUT_RING(ring, 0x00000000);
/* emit texture state: */
OUT_PKT3(ring, CP_LOAD_STATE, 10);
OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(0) |
CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |
CP_LOAD_STATE_0_STATE_BLOCK(SB_FRAG_TEX) |
CP_LOAD_STATE_0_NUM_UNIT(1));
OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |
CP_LOAD_STATE_1_EXT_SRC_ADDR(0));
OUT_RING(ring, A4XX_TEX_CONST_0_FMT(fd4_pipe2tex(format)) |
A4XX_TEX_CONST_0_TYPE(A4XX_TEX_2D) |
fd4_tex_swiz(format, PIPE_SWIZZLE_RED, PIPE_SWIZZLE_GREEN,
PIPE_SWIZZLE_BLUE, PIPE_SWIZZLE_ALPHA));
OUT_RING(ring, A4XX_TEX_CONST_1_WIDTH(psurf->width) |
A4XX_TEX_CONST_1_HEIGHT(psurf->height));
OUT_RING(ring, A4XX_TEX_CONST_2_PITCH(slice->pitch * rsc->cpp));
OUT_RING(ring, 0x00000000);
OUT_RELOC(ring, rsc->bo, offset, 0, 0);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
}
static void
fd_resource_flush_rgtc(struct fd_transfer *trans, const struct pipe_box *box)
{
struct fd_resource *rsc = fd_resource(trans->base.resource);
struct fd_resource_slice *slice = fd_resource_slice(rsc, trans->base.level);
enum pipe_format format = trans->base.resource->format;
uint8_t *data = fd_bo_map(rsc->bo) + slice->offset +
fd_resource_layer_offset(rsc, slice, trans->base.box.z) +
((trans->base.box.y + box->y) * slice->pitch +
trans->base.box.x + box->x) * rsc->cpp;
uint8_t *source = trans->staging +
util_format_get_nblocksy(format, box->y) * trans->base.stride +
util_format_get_stride(format, box->x);
switch (format) {
case PIPE_FORMAT_RGTC1_UNORM:
case PIPE_FORMAT_RGTC1_SNORM:
case PIPE_FORMAT_LATC1_UNORM:
case PIPE_FORMAT_LATC1_SNORM:
util_format_rgtc1_unorm_unpack_rgba_8unorm(
data, slice->pitch * rsc->cpp,
source, trans->base.stride,
box->width, box->height);
break;
case PIPE_FORMAT_RGTC2_UNORM:
case PIPE_FORMAT_RGTC2_SNORM:
case PIPE_FORMAT_LATC2_UNORM:
case PIPE_FORMAT_LATC2_SNORM:
util_format_rgtc2_unorm_unpack_rgba_8unorm(
data, slice->pitch * rsc->cpp,
source, trans->base.stride,
box->width, box->height);
break;
default:
assert(!"Unexpected format\n");
break;
}
}
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 fd_resource_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 = util_slab_alloc(&ctx->transfer_pool);
if (!ptrans)
return NULL;
/* util_slab_alloc() 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->cpp;
ptrans->layer_stride = rsc->layer_first ? rsc->layer_size : slice->size0;
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));
if (rsc->stencil)
realloc_bo(rsc->stencil, fd_bo_size(rsc->stencil->bo));
fd_invalidate_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)) {
/* 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.
*/
if (((ptrans->usage & PIPE_TRANSFER_WRITE) &&
pending(rsc, FD_PENDING_READ | FD_PENDING_WRITE)) ||
pending(rsc, FD_PENDING_WRITE))
fd_context_render(pctx);
/* 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.
*/
ret = fd_bo_cpu_prep(rsc->bo, ctx->screen->pipe, op);
if (ret)
goto fail;
}
buf = fd_bo_map(rsc->bo);
if (!buf)
goto fail;
offset = slice->offset +
box->y / util_format_get_blockheight(format) * ptrans->stride +
box->x / util_format_get_blockwidth(format) * rsc->cpp +
fd_resource_layer_offset(rsc, slice, box->z);
if (prsc->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT ||
prsc->format == PIPE_FORMAT_X32_S8X24_UINT) {
assert(trans->base.box.depth == 1);
trans->base.stride = trans->base.box.width * rsc->cpp * 2;
trans->staging = malloc(trans->base.stride * trans->base.box.height);
if (!trans->staging)
goto fail;
/* if we're not discarding the whole range (or resource), we must copy
* the real data in.
*/
if (!(usage & (PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE |
PIPE_TRANSFER_DISCARD_RANGE))) {
struct fd_resource_slice *sslice =
fd_resource_slice(rsc->stencil, level);
void *sbuf = fd_bo_map(rsc->stencil->bo);
if (!sbuf)
goto fail;
float *depth = (float *)(buf + slice->offset +
fd_resource_layer_offset(rsc, slice, box->z) +
box->y * slice->pitch * 4 + box->x * 4);
uint8_t *stencil = sbuf + sslice->offset +
fd_resource_layer_offset(rsc->stencil, sslice, box->z) +
box->y * sslice->pitch + box->x;
if (format != PIPE_FORMAT_X32_S8X24_UINT)
util_format_z32_float_s8x24_uint_pack_z_float(
trans->staging, trans->base.stride,
depth, slice->pitch * 4,
box->width, box->height);
util_format_z32_float_s8x24_uint_pack_s_8uint(
trans->staging, trans->base.stride,
stencil, sslice->pitch,
box->width, box->height);
}
buf = trans->staging;
offset = 0;
} else if (rsc->internal_format != format &&
util_format_description(format)->layout == UTIL_FORMAT_LAYOUT_RGTC) {
assert(trans->base.box.depth == 1);
trans->base.stride = util_format_get_stride(
format, trans->base.box.width);
trans->staging = malloc(
util_format_get_2d_size(format, trans->base.stride,
trans->base.box.height));
if (!trans->staging)
goto fail;
/* if we're not discarding the whole range (or resource), we must copy
* the real data in.
*/
if (!(usage & (PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE |
PIPE_TRANSFER_DISCARD_RANGE))) {
uint8_t *rgba8 = (uint8_t *)buf + slice->offset +
fd_resource_layer_offset(rsc, slice, box->z) +
box->y * slice->pitch * rsc->cpp + box->x * rsc->cpp;
switch (format) {
case PIPE_FORMAT_RGTC1_UNORM:
case PIPE_FORMAT_RGTC1_SNORM:
case PIPE_FORMAT_LATC1_UNORM:
case PIPE_FORMAT_LATC1_SNORM:
util_format_rgtc1_unorm_pack_rgba_8unorm(
trans->staging, trans->base.stride,
rgba8, slice->pitch * rsc->cpp,
box->width, box->height);
break;
case PIPE_FORMAT_RGTC2_UNORM:
case PIPE_FORMAT_RGTC2_SNORM:
case PIPE_FORMAT_LATC2_UNORM:
case PIPE_FORMAT_LATC2_SNORM:
util_format_rgtc2_unorm_pack_rgba_8unorm(
trans->staging, trans->base.stride,
rgba8, slice->pitch * rsc->cpp,
box->width, box->height);
break;
default:
assert(!"Unexpected format");
break;
}
}
buf = trans->staging;
offset = 0;
}
*pptrans = ptrans;
return buf + offset;
fail:
fd_resource_transfer_unmap(pctx, ptrans);
return NULL;
}
static void
emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
struct pipe_surface **bufs, struct fd_gmem_stateobj *gmem)
{
enum a5xx_tile_mode tile_mode;
unsigned i;
if (gmem) {
tile_mode = TILE5_2;
} else {
tile_mode = TILE5_LINEAR;
}
for (i = 0; i < A5XX_MAX_RENDER_TARGETS; i++) {
enum a5xx_color_fmt format = 0;
enum a3xx_color_swap swap = WZYX;
bool srgb = false;
struct fd_resource *rsc = NULL;
struct fd_resource_slice *slice = NULL;
uint32_t stride = 0;
uint32_t size = 0;
uint32_t base = 0;
uint32_t offset = 0;
if ((i < nr_bufs) && bufs[i]) {
struct pipe_surface *psurf = bufs[i];
enum pipe_format pformat = psurf->format;
rsc = fd_resource(psurf->texture);
slice = fd_resource_slice(rsc, psurf->u.tex.level);
format = fd5_pipe2color(pformat);
swap = fd5_pipe2swap(pformat);
srgb = util_format_is_srgb(pformat);
debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
offset = fd_resource_offset(rsc, psurf->u.tex.level,
psurf->u.tex.first_layer);
if (gmem) {
stride = gmem->bin_w * rsc->cpp;
size = stride * gmem->bin_h;
base = gmem->cbuf_base[i];
} else {
stride = slice->pitch * rsc->cpp;
size = slice->size0;
}
}
OUT_PKT4(ring, REG_A5XX_RB_MRT_BUF_INFO(i), 5);
OUT_RING(ring, A5XX_RB_MRT_BUF_INFO_COLOR_FORMAT(format) |
A5XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(tile_mode) |
A5XX_RB_MRT_BUF_INFO_COLOR_SWAP(swap) |
0x800 | /* XXX 0x1000 for RECTLIST clear, 0x0 for BLIT.. */
COND(srgb, A5XX_RB_MRT_BUF_INFO_COLOR_SRGB));
OUT_RING(ring, A5XX_RB_MRT_PITCH(stride));
OUT_RING(ring, A5XX_RB_MRT_ARRAY_PITCH(size));
if (gmem || (i >= nr_bufs) || !bufs[i]) {
OUT_RING(ring, base); /* RB_MRT[i].BASE_LO */
OUT_RING(ring, 0x00000000); /* RB_MRT[i].BASE_HI */
} else {
debug_assert((offset + size) <= fd_bo_size(rsc->bo));
OUT_RELOCW(ring, rsc->bo, offset, 0, 0); /* BASE_LO/HI */
}
OUT_PKT4(ring, REG_A5XX_SP_FS_MRT_REG(i), 1);
OUT_RING(ring, A5XX_SP_FS_MRT_REG_COLOR_FORMAT(format));
/* when we support UBWC, these would be the system memory
* addr/pitch/etc:
*/
OUT_PKT4(ring, REG_A5XX_RB_MRT_FLAG_BUFFER(i), 4);
OUT_RING(ring, 0x00000000); /* RB_MRT_FLAG_BUFFER[i].ADDR_LO */
OUT_RING(ring, 0x00000000); /* RB_MRT_FLAG_BUFFER[i].ADDR_HI */
OUT_RING(ring, A5XX_RB_MRT_FLAG_BUFFER_PITCH(0));
OUT_RING(ring, A5XX_RB_MRT_FLAG_BUFFER_ARRAY_PITCH(0));
}
}
static void
emit_zs(struct fd_ringbuffer *ring, struct pipe_surface *zsbuf,
struct fd_gmem_stateobj *gmem)
{
if (zsbuf) {
struct fd_resource *rsc = fd_resource(zsbuf->texture);
enum a5xx_depth_format fmt = fd5_pipe2depth(zsbuf->format);
uint32_t cpp = rsc->cpp;
uint32_t stride = 0;
uint32_t size = 0;
if (gmem) {
stride = cpp * gmem->bin_w;
size = stride * gmem->bin_h;
} else {
struct fd_resource_slice *slice = fd_resource_slice(rsc, 0);
stride = slice->pitch * rsc->cpp;
size = slice->size0;
}
OUT_PKT4(ring, REG_A5XX_RB_DEPTH_BUFFER_INFO, 5);
OUT_RING(ring, A5XX_RB_DEPTH_BUFFER_INFO_DEPTH_FORMAT(fmt));
if (gmem) {
OUT_RING(ring, gmem->zsbuf_base[0]); /* RB_DEPTH_BUFFER_BASE_LO */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_BUFFER_BASE_HI */
} else {
OUT_RELOCW(ring, rsc->bo, 0, 0, 0); /* RB_DEPTH_BUFFER_BASE_LO/HI */
}
OUT_RING(ring, A5XX_RB_DEPTH_BUFFER_PITCH(stride));
OUT_RING(ring, A5XX_RB_DEPTH_BUFFER_ARRAY_PITCH(size));
OUT_PKT4(ring, REG_A5XX_GRAS_SU_DEPTH_BUFFER_INFO, 1);
OUT_RING(ring, A5XX_GRAS_SU_DEPTH_BUFFER_INFO_DEPTH_FORMAT(fmt));
OUT_PKT4(ring, REG_A5XX_RB_DEPTH_FLAG_BUFFER_BASE_LO, 3);
OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_BASE_LO */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_BASE_HI */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_PITCH */
if (rsc->stencil) {
if (gmem) {
stride = 1 * gmem->bin_w;
size = stride * gmem->bin_h;
} else {
struct fd_resource_slice *slice = fd_resource_slice(rsc->stencil, 0);
stride = slice->pitch * rsc->cpp;
size = slice->size0;
}
OUT_PKT4(ring, REG_A5XX_RB_STENCIL_INFO, 5);
OUT_RING(ring, A5XX_RB_STENCIL_INFO_SEPARATE_STENCIL);
if (gmem) {
OUT_RING(ring, gmem->zsbuf_base[1]); /* RB_STENCIL_BASE_LO */
OUT_RING(ring, 0x00000000); /* RB_STENCIL_BASE_HI */
} else {
OUT_RELOCW(ring, rsc->stencil->bo, 0, 0, 0); /* RB_STENCIL_BASE_LO/HI */
}
OUT_RING(ring, A5XX_RB_STENCIL_PITCH(stride));
OUT_RING(ring, A5XX_RB_STENCIL_ARRAY_PITCH(size));
} else {
OUT_PKT4(ring, REG_A5XX_RB_STENCIL_INFO, 1);
OUT_RING(ring, 0x00000000); /* RB_STENCIL_INFO */
}
} else {
OUT_PKT4(ring, REG_A5XX_RB_DEPTH_BUFFER_INFO, 5);
OUT_RING(ring, A5XX_RB_DEPTH_BUFFER_INFO_DEPTH_FORMAT(DEPTH5_NONE));
OUT_RING(ring, 0x00000000); /* RB_DEPTH_BUFFER_BASE_LO */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_BUFFER_BASE_HI */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_BUFFER_PITCH */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_BUFFER_ARRAY_PITCH */
OUT_PKT4(ring, REG_A5XX_GRAS_SU_DEPTH_BUFFER_INFO, 1);
OUT_RING(ring, A5XX_GRAS_SU_DEPTH_BUFFER_INFO_DEPTH_FORMAT(DEPTH5_NONE));
OUT_PKT4(ring, REG_A5XX_RB_DEPTH_FLAG_BUFFER_BASE_LO, 3);
OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_BASE_LO */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_BASE_HI */
OUT_RING(ring, 0x00000000); /* RB_DEPTH_FLAG_BUFFER_PITCH */
OUT_PKT4(ring, REG_A5XX_RB_STENCIL_INFO, 1);
OUT_RING(ring, 0x00000000); /* RB_STENCIL_INFO */
}
}
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 fd_resource_slice *slice = fd_resource_slice(rsc, level);
struct pipe_transfer *ptrans;
enum pipe_format format = prsc->format;
uint32_t op = 0;
char *buf;
int ret = 0;
ptrans = util_slab_alloc(&ctx->transfer_pool);
if (!ptrans)
return NULL;
/* util_slab_alloc() doesn't zero: */
memset(ptrans, 0, sizeof(*ptrans));
pipe_resource_reference(&ptrans->resource, prsc);
ptrans->level = level;
ptrans->usage = usage;
ptrans->box = *box;
ptrans->stride = slice->pitch * rsc->cpp;
ptrans->layer_stride = slice->size0;
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)
op |= DRM_FREEDRENO_PREP_NOSYNC;
/* some state trackers (at least XA) don't do this.. */
if (!(usage & (PIPE_TRANSFER_FLUSH_EXPLICIT | PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE)))
fd_resource_transfer_flush_region(pctx, ptrans, box);
if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
ret = fd_bo_cpu_prep(rsc->bo, ctx->screen->pipe, op);
if ((ret == -EBUSY) && (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE))
realloc_bo(rsc, fd_bo_size(rsc->bo));
else if (ret)
goto fail;
}
buf = fd_bo_map(rsc->bo);
if (!buf) {
fd_resource_transfer_unmap(pctx, ptrans);
return NULL;
}
*pptrans = ptrans;
return buf + slice->offset +
box->y / util_format_get_blockheight(format) * ptrans->stride +
box->x / util_format_get_blockwidth(format) * rsc->cpp +
box->z * slice->size0;
fail:
fd_resource_transfer_unmap(pctx, ptrans);
return NULL;
}
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 fd_resource_slice *slice = fd_resource_slice(rsc, level);
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", prsc, level, usage);
ptrans = util_slab_alloc(&ctx->transfer_pool);
if (!ptrans)
return NULL;
/* util_slab_alloc() doesn't zero: */
memset(ptrans, 0, sizeof(*ptrans));
pipe_resource_reference(&ptrans->resource, prsc);
ptrans->level = level;
ptrans->usage = usage;
ptrans->box = *box;
ptrans->stride = slice->pitch * rsc->cpp;
ptrans->layer_stride = slice->size0;
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));
fd_invalidate_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)) {
/* 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.
*/
if (rsc->dirty ||
((ptrans->usage & PIPE_TRANSFER_WRITE) && rsc->reading))
fd_context_render(pctx);
/* 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.
*/
ret = fd_bo_cpu_prep(rsc->bo, ctx->screen->pipe, op);
if (ret)
goto fail;
}
buf = fd_bo_map(rsc->bo);
if (!buf) {
fd_resource_transfer_unmap(pctx, ptrans);
return NULL;
}
*pptrans = ptrans;
if (rsc->layer_first) {
offset = slice->offset +
box->y / util_format_get_blockheight(format) * ptrans->stride +
box->x / util_format_get_blockwidth(format) * rsc->cpp +
box->z * rsc->layer_size;
} else {
offset = slice->offset +
box->y / util_format_get_blockheight(format) * ptrans->stride +
box->x / util_format_get_blockwidth(format) * rsc->cpp +
box->z * slice->size0;
}
return buf + offset;
fail:
fd_resource_transfer_unmap(pctx, ptrans);
return NULL;
}
static void
emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
struct pipe_surface **bufs, uint32_t *bases,
uint32_t bin_w, bool decode_srgb)
{
enum a4xx_tile_mode tile_mode;
unsigned i;
if (bin_w) {
tile_mode = 2;
} else {
tile_mode = TILE4_LINEAR;
}
for (i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
enum a4xx_color_fmt format = 0;
enum a3xx_color_swap swap = WZYX;
bool srgb = false;
struct fd_resource *rsc = NULL;
struct fd_resource_slice *slice = NULL;
uint32_t stride = 0;
uint32_t base = 0;
uint32_t offset = 0;
if ((i < nr_bufs) && bufs[i]) {
struct pipe_surface *psurf = bufs[i];
enum pipe_format pformat = psurf->format;
rsc = fd_resource(psurf->texture);
/* In case we're drawing to Z32F_S8, the "color" actually goes to
* the stencil
*/
if (rsc->stencil) {
rsc = rsc->stencil;
pformat = rsc->base.b.format;
bases++;
}
slice = fd_resource_slice(rsc, psurf->u.tex.level);
format = fd4_pipe2color(pformat);
swap = fd4_pipe2swap(pformat);
if (decode_srgb)
srgb = util_format_is_srgb(pformat);
else
pformat = util_format_linear(pformat);
debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
offset = fd_resource_offset(rsc, psurf->u.tex.level,
psurf->u.tex.first_layer);
if (bin_w) {
stride = bin_w * rsc->cpp;
if (bases) {
base = bases[i];
}
} else {
stride = slice->pitch * rsc->cpp;
}
} else if ((i < nr_bufs) && bases) {
base = bases[i];
}
OUT_PKT0(ring, REG_A4XX_RB_MRT_BUF_INFO(i), 3);
OUT_RING(ring, A4XX_RB_MRT_BUF_INFO_COLOR_FORMAT(format) |
A4XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(tile_mode) |
A4XX_RB_MRT_BUF_INFO_COLOR_BUF_PITCH(stride) |
A4XX_RB_MRT_BUF_INFO_COLOR_SWAP(swap) |
COND(srgb, A4XX_RB_MRT_BUF_INFO_COLOR_SRGB));
if (bin_w || (i >= nr_bufs) || !bufs[i]) {
OUT_RING(ring, base);
OUT_RING(ring, A4XX_RB_MRT_CONTROL3_STRIDE(stride));
} else {
OUT_RELOCW(ring, rsc->bo, offset, 0, 0);
/* RB_MRT[i].CONTROL3.STRIDE not emitted by c2d..
* not sure if we need to skip it for bypass or
* not.
*/
OUT_RING(ring, A4XX_RB_MRT_CONTROL3_STRIDE(0));
}
}
}
static void
emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
struct pipe_surface **bufs, uint32_t *bases, uint32_t bin_w)
{
enum a4xx_tile_mode tile_mode;
unsigned i;
if (bin_w) {
tile_mode = 2;
} else {
tile_mode = TILE4_LINEAR;
}
for (i = 0; i < 8; i++) {
enum a4xx_color_fmt format = 0;
enum a3xx_color_swap swap = WZYX;
struct fd_resource *rsc = NULL;
struct fd_resource_slice *slice = NULL;
uint32_t stride = 0;
uint32_t base = 0;
uint32_t offset = 0;
if ((i < nr_bufs) && bufs[i]) {
struct pipe_surface *psurf = bufs[i];
rsc = fd_resource(psurf->texture);
slice = fd_resource_slice(rsc, psurf->u.tex.level);
format = fd4_pipe2color(psurf->format);
swap = fd4_pipe2swap(psurf->format);
debug_assert(psurf->u.tex.first_layer == psurf->u.tex.last_layer);
offset = fd_resource_offset(rsc, psurf->u.tex.level,
psurf->u.tex.first_layer);
if (bin_w) {
stride = bin_w * rsc->cpp;
if (bases) {
base = bases[i];
}
} else {
stride = slice->pitch * rsc->cpp;
}
}
OUT_PKT0(ring, REG_A4XX_RB_MRT_BUF_INFO(i), 3);
OUT_RING(ring, A4XX_RB_MRT_BUF_INFO_COLOR_FORMAT(format) |
A4XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(tile_mode) |
A4XX_RB_MRT_BUF_INFO_COLOR_BUF_PITCH(stride) |
A4XX_RB_MRT_BUF_INFO_COLOR_SWAP(swap));
if (bin_w || (i >= nr_bufs)) {
OUT_RING(ring, base);
OUT_RING(ring, A4XX_RB_MRT_CONTROL3_STRIDE(stride));
} else {
OUT_RELOCW(ring, rsc->bo, offset, 0, 0);
/* RB_MRT[i].CONTROL3.STRIDE not emitted by c2d..
* not sure if we need to skip it for bypass or
* not.
*/
OUT_RING(ring, A4XX_RB_MRT_CONTROL3_STRIDE(0));
}
}
}