/**
* Build a struct pipe_blit_info object from the arguments used by the
* pipe::resource_copy_region() function.
*/
static void
build_blit_info(struct pipe_resource *dst_tex,
unsigned dst_level,
unsigned dst_x,
unsigned dst_y,
unsigned dst_z,
struct pipe_resource *src_tex,
unsigned src_level,
const struct pipe_box *src_box,
struct pipe_blit_info *blit)
{
memset(blit, 0, sizeof(*blit));
blit->src.format = src_tex->format;
blit->dst.format = dst_tex->format;
blit->mask = util_format_get_mask(blit->dst.format);
blit->filter = PIPE_TEX_FILTER_NEAREST;
blit->src.resource = src_tex;
blit->src.level = src_level;
blit->dst.resource = dst_tex;
blit->dst.level = dst_level;
blit->src.box = *src_box;
u_box_3d(dst_x, dst_y, dst_z, src_box->width, src_box->height,
src_box->depth, &blit->dst.box);
}
static void
i915_surface_copy_render(struct pipe_context *pipe,
struct pipe_resource *dst, unsigned dst_level,
unsigned dstx, unsigned dsty, unsigned dstz,
struct pipe_resource *src, unsigned src_level,
const struct pipe_box *src_box)
{
struct i915_context *i915 = i915_context(pipe);
unsigned src_width0 = src->width0;
unsigned src_height0 = src->height0;
unsigned dst_width0 = dst->width0;
unsigned dst_height0 = dst->height0;
struct pipe_box dstbox;
struct pipe_sampler_view src_templ, *src_view;
struct pipe_surface dst_templ, *dst_view;
const struct util_format_description *desc;
/* Fallback for buffers. */
if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER)
goto fallback;
/* Fallback for depth&stencil. XXX: see if we can use a proxy format */
desc = util_format_description(src->format);
if (desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS)
goto fallback;
desc = util_format_description(dst->format);
if (desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS)
goto fallback;
util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
util_blitter_default_src_texture(i915->blitter, &src_templ, src, src_level);
if (!util_blitter_is_copy_supported(i915->blitter, dst, src))
goto fallback;
i915_util_blitter_save_states(i915);
dst_view = i915_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
src_view = i915_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
abs(src_box->depth), &dstbox);
util_blitter_blit_generic(i915->blitter, dst_view, &dstbox,
src_view, src_box, src_width0, src_height0,
PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
FALSE);
return;
fallback:
util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
src, src_level, src_box);
}
/* Copy from a transfer's staging texture to a full GPU one. */
static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
{
struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
struct pipe_resource *texture = transfer->resource;
struct pipe_box sbox;
u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height, transfer->box.depth, &sbox);
ctx->resource_copy_region(ctx, texture, transfer->level,
transfer->box.x, transfer->box.y, transfer->box.z,
rtransfer->staging,
0, &sbox);
}
HRESULT NINE_WINAPI
NineVolume9_UnlockBox( struct NineVolume9 *This )
{
DBG("This=%p lock_count=%u\n", This, This->lock_count);
user_assert(This->lock_count, D3DERR_INVALIDCALL);
if (This->transfer) {
This->pipe->transfer_unmap(This->pipe, This->transfer);
This->transfer = NULL;
}
--This->lock_count;
if (This->data_conversion) {
struct pipe_transfer *transfer;
uint8_t *dst = This->data;
struct pipe_box box;
u_box_3d(0, 0, 0, This->desc.Width, This->desc.Height, This->desc.Depth,
&box);
if (!dst) {
dst = This->pipe->transfer_map(This->pipe,
This->resource,
This->level,
PIPE_TRANSFER_WRITE |
PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE,
&box, &transfer);
if (!dst)
return D3D_OK;
}
(void) util_format_translate_3d(This->info.format,
dst, This->data ? This->stride : transfer->stride,
This->data ? This->layer_stride : transfer->layer_stride,
0, 0, 0,
This->format_conversion,
This->data_conversion,
This->stride_conversion,
This->layer_stride_conversion,
0, 0, 0,
This->desc.Width, This->desc.Height,
This->desc.Height);
if (!This->data)
pipe_transfer_unmap(This->pipe, transfer);
}
return D3D_OK;
}
/* Copy a detiled texture to a tiled one. */
static void r300_copy_into_tiled_texture(struct pipe_context *ctx,
struct r300_transfer *r300transfer)
{
struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer;
struct pipe_resource *tex = transfer->resource;
struct pipe_box src_box;
u_box_3d(0, 0, 0,
transfer->box.width, transfer->box.height, transfer->box.depth,
&src_box);
ctx->resource_copy_region(ctx, tex, transfer->level,
transfer->box.x, transfer->box.y, transfer->box.z,
&r300transfer->linear_texture->b.b, 0, &src_box);
/* XXX remove this. */
r300_flush(ctx, 0, NULL);
}
/* Copy from a transfer's staging texture to a full GPU one. */
static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
{
struct r600_common_context *rctx = (struct r600_common_context*)ctx;
struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
struct pipe_resource *dst = transfer->resource;
struct pipe_resource *src = &rtransfer->staging->b.b;
struct pipe_box sbox;
u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height, transfer->box.depth, &sbox);
if (dst->nr_samples > 1) {
r600_copy_region_with_blit(ctx, dst, transfer->level,
transfer->box.x, transfer->box.y, transfer->box.z,
src, 0, &sbox);
return;
}
rctx->dma_copy(ctx, dst, transfer->level,
transfer->box.x, transfer->box.y, transfer->box.z,
src, 0, &sbox);
}
static void
swr_transfer_unmap(struct pipe_context *pipe, struct pipe_transfer *transfer)
{
assert(transfer->resource);
struct swr_resource *spr = swr_resource(transfer->resource);
/* if we're mapping the depth/stencil, copy in stencil for the section
* being written out
*/
if (transfer->usage & PIPE_TRANSFER_WRITE &&
!(transfer->usage & PIPE_TRANSFER_FLUSH_EXPLICIT) &&
spr->has_depth && spr->has_stencil) {
struct pipe_box box;
u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height,
transfer->box.depth, &box);
swr_transfer_flush_region(pipe, transfer, &box);
}
pipe_resource_reference(&transfer->resource, NULL);
FREE(transfer);
}
inline void
NineVolume9_AddDirtyRegion( struct NineVolume9 *This,
const struct pipe_box *box )
{
struct pipe_box cover_a, cover_b;
float vol[2];
if (!box) {
u_box_3d(0, 0, 0, This->desc.Width, This->desc.Height,
This->desc.Depth, &This->dirty_box[0]);
memset(&This->dirty_box[1], 0, sizeof(This->dirty_box[1]));
return;
}
if (!This->dirty_box[0].width) {
This->dirty_box[0] = *box;
return;
}
u_box_union_3d(&cover_a, &This->dirty_box[0], box);
vol[0] = u_box_volume_3d(&cover_a);
if (This->dirty_box[1].width == 0) {
vol[1] = u_box_volume_3d(&This->dirty_box[0]);
if (vol[0] > (vol[1] * 1.5f))
This->dirty_box[1] = *box;
else
This->dirty_box[0] = cover_a;
} else {
u_box_union_3d(&cover_b, &This->dirty_box[1], box);
vol[1] = u_box_volume_3d(&cover_b);
if (vol[0] > vol[1])
This->dirty_box[1] = cover_b;
else
This->dirty_box[0] = cover_a;
}
}
static void r600_resource_copy_region(struct pipe_context *ctx,
struct pipe_resource *dst,
unsigned dst_level,
unsigned dstx, unsigned dsty, unsigned dstz,
struct pipe_resource *src,
unsigned src_level,
const struct pipe_box *src_box)
{
struct r600_context *rctx = (struct r600_context *)ctx;
struct pipe_surface *dst_view, dst_templ;
struct pipe_sampler_view src_templ, *src_view;
unsigned dst_width, dst_height, src_width0, src_height0, src_widthFL, src_heightFL;
unsigned src_force_level = 0;
struct pipe_box sbox, dstbox;
/* Handle buffers first. */
if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
if ((src->bind & PIPE_BIND_GLOBAL) ||
(dst->bind & PIPE_BIND_GLOBAL)) {
r600_copy_global_buffer(ctx, dst, dstx, src, src_box);
} else {
r600_copy_buffer(ctx, dst, dstx, src, src_box);
}
return;
}
assert(u_max_sample(dst) == u_max_sample(src));
/* The driver doesn't decompress resources automatically while
* u_blitter is rendering. */
if (!r600_decompress_subresource(ctx, src, src_level,
src_box->z, src_box->z + src_box->depth - 1)) {
return; /* error */
}
dst_width = u_minify(dst->width0, dst_level);
dst_height = u_minify(dst->height0, dst_level);
src_width0 = src->width0;
src_height0 = src->height0;
src_widthFL = u_minify(src->width0, src_level);
src_heightFL = u_minify(src->height0, src_level);
util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
util_blitter_default_src_texture(&src_templ, src, src_level);
if (util_format_is_compressed(src->format)) {
unsigned blocksize = util_format_get_blocksize(src->format);
if (blocksize == 8)
src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
else
src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
dst_templ.format = src_templ.format;
dst_width = util_format_get_nblocksx(dst->format, dst_width);
dst_height = util_format_get_nblocksy(dst->format, dst_height);
src_width0 = util_format_get_nblocksx(src->format, src_width0);
src_height0 = util_format_get_nblocksy(src->format, src_height0);
src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
src_heightFL = util_format_get_nblocksy(src->format, src_heightFL);
dstx = util_format_get_nblocksx(dst->format, dstx);
dsty = util_format_get_nblocksy(dst->format, dsty);
sbox.x = util_format_get_nblocksx(src->format, src_box->x);
sbox.y = util_format_get_nblocksy(src->format, src_box->y);
sbox.z = src_box->z;
sbox.width = util_format_get_nblocksx(src->format, src_box->width);
sbox.height = util_format_get_nblocksy(src->format, src_box->height);
sbox.depth = src_box->depth;
src_box = &sbox;
src_force_level = src_level;
} else if (!util_blitter_is_copy_supported(rctx->blitter, dst, src)) {
if (util_format_is_subsampled_422(src->format)) {
src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
dst_width = util_format_get_nblocksx(dst->format, dst_width);
src_width0 = util_format_get_nblocksx(src->format, src_width0);
src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
dstx = util_format_get_nblocksx(dst->format, dstx);
sbox = *src_box;
sbox.x = util_format_get_nblocksx(src->format, src_box->x);
sbox.width = util_format_get_nblocksx(src->format, src_box->width);
src_box = &sbox;
} else {
unsigned blocksize = util_format_get_blocksize(src->format);
switch (blocksize) {
case 1:
dst_templ.format = PIPE_FORMAT_R8_UNORM;
src_templ.format = PIPE_FORMAT_R8_UNORM;
break;
case 2:
dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
src_templ.format = PIPE_FORMAT_R8G8_UNORM;
break;
case 4:
dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
break;
case 8:
dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
break;
case 16:
dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
break;
default:
fprintf(stderr, "Unhandled format %s with blocksize %u\n",
util_format_short_name(src->format), blocksize);
assert(0);
}
}
}
dst_view = r600_create_surface_custom(ctx, dst, &dst_templ, dst_width, dst_height);
if (rctx->b.chip_class >= EVERGREEN) {
src_view = evergreen_create_sampler_view_custom(ctx, src, &src_templ,
src_width0, src_height0,
src_force_level);
} else {
src_view = r600_create_sampler_view_custom(ctx, src, &src_templ,
src_widthFL, src_heightFL);
}
u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
abs(src_box->depth), &dstbox);
/* Copy. */
r600_blitter_begin(ctx, R600_COPY_TEXTURE);
util_blitter_blit_generic(rctx->blitter, dst_view, &dstbox,
src_view, src_box, src_width0, src_height0,
PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
r600_blitter_end(ctx);
pipe_surface_reference(&dst_view, NULL);
pipe_sampler_view_reference(&src_view, NULL);
}
/* Copy a block of pixels from one surface to another. */
static void r300_resource_copy_region(struct pipe_context *pipe,
struct pipe_resource *dst,
unsigned dst_level,
unsigned dstx, unsigned dsty, unsigned dstz,
struct pipe_resource *src,
unsigned src_level,
const struct pipe_box *src_box)
{
struct pipe_screen *screen = pipe->screen;
struct r300_context *r300 = r300_context(pipe);
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
unsigned src_width0 = r300_resource(src)->tex.width0;
unsigned src_height0 = r300_resource(src)->tex.height0;
unsigned dst_width0 = r300_resource(dst)->tex.width0;
unsigned dst_height0 = r300_resource(dst)->tex.height0;
unsigned layout;
struct pipe_box box, dstbox;
struct pipe_sampler_view src_templ, *src_view;
struct pipe_surface dst_templ, *dst_view;
/* Fallback for buffers. */
if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) ||
!r300_is_blit_supported(dst->format)) {
util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
src, src_level, src_box);
return;
}
/* Can't read MSAA textures. */
if (src->nr_samples > 1 || dst->nr_samples > 1) {
return;
}
/* The code below changes the texture format so that the copy can be done
* on hardware. E.g. depth-stencil surfaces are copied as RGBA
* colorbuffers. */
util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
util_blitter_default_src_texture(&src_templ, src, src_level);
layout = util_format_description(dst_templ.format)->layout;
/* Handle non-renderable plain formats. */
if (layout == UTIL_FORMAT_LAYOUT_PLAIN &&
(!screen->is_format_supported(screen, src_templ.format, src->target,
src->nr_samples,
PIPE_BIND_SAMPLER_VIEW) ||
!screen->is_format_supported(screen, dst_templ.format, dst->target,
dst->nr_samples,
PIPE_BIND_RENDER_TARGET))) {
switch (util_format_get_blocksize(dst_templ.format)) {
case 1:
dst_templ.format = PIPE_FORMAT_I8_UNORM;
break;
case 2:
dst_templ.format = PIPE_FORMAT_B4G4R4A4_UNORM;
break;
case 4:
dst_templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
break;
case 8:
dst_templ.format = PIPE_FORMAT_R16G16B16A16_UNORM;
break;
default:
debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
"r300: copy_region: Software fallback doesn't work for tiled textures.\n",
util_format_short_name(dst_templ.format));
}
src_templ.format = dst_templ.format;
}
/* Handle compressed formats. */
if (layout == UTIL_FORMAT_LAYOUT_S3TC ||
layout == UTIL_FORMAT_LAYOUT_RGTC) {
assert(src_templ.format == dst_templ.format);
box = *src_box;
src_box = &box;
dst_width0 = align(dst_width0, 4);
dst_height0 = align(dst_height0, 4);
src_width0 = align(src_width0, 4);
src_height0 = align(src_height0, 4);
box.width = align(box.width, 4);
box.height = align(box.height, 4);
switch (util_format_get_blocksize(dst_templ.format)) {
case 8:
/* one 4x4 pixel block has 8 bytes.
* we set 1 pixel = 4 bytes ===> 1 block corrensponds to 2 pixels. */
dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
dst_width0 = dst_width0 / 2;
src_width0 = src_width0 / 2;
dstx /= 2;
box.x /= 2;
box.width /= 2;
break;
case 16:
/* one 4x4 pixel block has 16 bytes.
* we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
break;
}
src_templ.format = dst_templ.format;
dst_height0 = dst_height0 / 4;
src_height0 = src_height0 / 4;
dsty /= 4;
box.y /= 4;
box.height /= 4;
}
/* Fallback for textures. */
if (!screen->is_format_supported(screen, dst_templ.format,
dst->target, dst->nr_samples,
PIPE_BIND_RENDER_TARGET) ||
!screen->is_format_supported(screen, src_templ.format,
src->target, src->nr_samples,
PIPE_BIND_SAMPLER_VIEW)) {
assert(0 && "this shouldn't happen, update r300_is_blit_supported");
util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
src, src_level, src_box);
return;
}
/* Decompress ZMASK. */
if (r300->zmask_in_use && !r300->locked_zbuffer) {
if (fb->zsbuf->texture == src ||
fb->zsbuf->texture == dst) {
r300_decompress_zmask(r300);
}
}
dst_view = r300_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
src_view = r300_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
abs(src_box->depth), &dstbox);
r300_blitter_begin(r300, R300_COPY);
util_blitter_blit_generic(r300->blitter, dst_view, &dstbox,
src_view, src_box, src_width0, src_height0,
PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
r300_blitter_end(r300);
pipe_surface_reference(&dst_view, NULL);
pipe_sampler_view_reference(&src_view, NULL);
}
HRESULT WINAPI
NineVolume9_LockBox( struct NineVolume9 *This,
D3DLOCKED_BOX *pLockedVolume,
const D3DBOX *pBox,
DWORD Flags )
{
struct pipe_resource *resource = This->resource;
struct pipe_box box;
unsigned usage;
DBG("This=%p(%p) pLockedVolume=%p pBox=%p[%u..%u,%u..%u,%u..%u] Flags=%s\n",
This, This->base.container, pLockedVolume, pBox,
pBox ? pBox->Left : 0, pBox ? pBox->Right : 0,
pBox ? pBox->Top : 0, pBox ? pBox->Bottom : 0,
pBox ? pBox->Front : 0, pBox ? pBox->Back : 0,
nine_D3DLOCK_to_str(Flags));
user_assert(This->desc.Pool != D3DPOOL_DEFAULT ||
(This->desc.Usage & D3DUSAGE_DYNAMIC), D3DERR_INVALIDCALL);
user_assert(!((Flags & D3DLOCK_DISCARD) && (Flags & D3DLOCK_READONLY)),
D3DERR_INVALIDCALL);
user_assert(This->lock_count == 0, D3DERR_INVALIDCALL);
user_assert(pLockedVolume, E_POINTER);
if (pBox && This->desc.Pool == D3DPOOL_DEFAULT &&
util_format_is_compressed(This->info.format)) {
const unsigned w = util_format_get_blockwidth(This->info.format);
const unsigned h = util_format_get_blockheight(This->info.format);
user_assert(!(pBox->Left % w) && !(pBox->Right % w) &&
!(pBox->Top % h) && !(pBox->Bottom % h),
D3DERR_INVALIDCALL);
}
if (Flags & D3DLOCK_DISCARD) {
usage = PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD_RANGE;
} else {
usage = (Flags & D3DLOCK_READONLY) ?
PIPE_TRANSFER_READ : PIPE_TRANSFER_READ_WRITE;
}
if (Flags & D3DLOCK_DONOTWAIT)
usage |= PIPE_TRANSFER_DONTBLOCK;
if (pBox) {
d3dbox_to_pipe_box(&box, pBox);
if (u_box_clip_2d(&box, &box, This->desc.Width, This->desc.Height) < 0) {
DBG("Locked volume intersection empty.\n");
return D3DERR_INVALIDCALL;
}
} else {
u_box_3d(0, 0, 0, This->desc.Width, This->desc.Height, This->desc.Depth,
&box);
}
if (This->data) {
pLockedVolume->RowPitch = This->stride;
pLockedVolume->SlicePitch = This->layer_stride;
pLockedVolume->pBits =
NineVolume9_GetSystemMemPointer(This, box.x, box.y, box.z);
} else {
pLockedVolume->pBits =
This->pipe->transfer_map(This->pipe, resource, This->level, usage,
&box, &This->transfer);
if (!This->transfer) {
if (Flags & D3DLOCK_DONOTWAIT)
return D3DERR_WASSTILLDRAWING;
return D3DERR_DRIVERINTERNALERROR;
}
pLockedVolume->RowPitch = This->transfer->stride;
pLockedVolume->SlicePitch = This->transfer->layer_stride;
}
if (!(Flags & (D3DLOCK_NO_DIRTY_UPDATE | D3DLOCK_READONLY))) {
NineVolume9_MarkContainerDirty(This);
if (This->desc.Pool == D3DPOOL_MANAGED)
NineVolume9_AddDirtyRegion(This, &box);
}
++This->lock_count;
return D3D_OK;
}
static void si_resource_copy_region(struct pipe_context *ctx,
struct pipe_resource *dst,
unsigned dst_level,
unsigned dstx, unsigned dsty, unsigned dstz,
struct pipe_resource *src,
unsigned src_level,
const struct pipe_box *src_box)
{
struct si_context *sctx = (struct si_context *)ctx;
struct r600_texture *rdst = (struct r600_texture*)dst;
struct pipe_surface *dst_view, dst_templ;
struct pipe_sampler_view src_templ, *src_view;
struct texture_orig_info orig_info[2];
struct pipe_box sbox, dstbox;
boolean restore_orig[2];
/* Fallback for buffers. */
if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width);
return;
}
memset(orig_info, 0, sizeof(orig_info));
/* The driver doesn't decompress resources automatically while
* u_blitter is rendering. */
si_decompress_subresource(ctx, src, src_level,
src_box->z, src_box->z + src_box->depth - 1);
restore_orig[0] = restore_orig[1] = FALSE;
if (util_format_is_compressed(src->format) &&
util_format_is_compressed(dst->format)) {
si_compressed_to_blittable(src, src_level, &orig_info[0]);
restore_orig[0] = TRUE;
sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
sbox.y = util_format_get_nblocksy(orig_info[0].format, src_box->y);
sbox.z = src_box->z;
sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
sbox.height = util_format_get_nblocksy(orig_info[0].format, src_box->height);
sbox.depth = src_box->depth;
src_box = &sbox;
si_compressed_to_blittable(dst, dst_level, &orig_info[1]);
restore_orig[1] = TRUE;
/* translate the dst box as well */
dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
dsty = util_format_get_nblocksy(orig_info[1].format, dsty);
} else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src)) {
if (util_format_is_subsampled_422(src->format)) {
/* XXX untested */
si_change_format(src, src_level, &orig_info[0],
PIPE_FORMAT_R8G8B8A8_UINT);
si_change_format(dst, dst_level, &orig_info[1],
PIPE_FORMAT_R8G8B8A8_UINT);
sbox = *src_box;
sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
src_box = &sbox;
dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
restore_orig[0] = TRUE;
restore_orig[1] = TRUE;
} else {
unsigned blocksize = util_format_get_blocksize(src->format);
switch (blocksize) {
case 1:
si_change_format(src, src_level, &orig_info[0],
PIPE_FORMAT_R8_UNORM);
si_change_format(dst, dst_level, &orig_info[1],
PIPE_FORMAT_R8_UNORM);
break;
case 2:
si_change_format(src, src_level, &orig_info[0],
PIPE_FORMAT_R8G8_UNORM);
si_change_format(dst, dst_level, &orig_info[1],
PIPE_FORMAT_R8G8_UNORM);
break;
case 4:
si_change_format(src, src_level, &orig_info[0],
PIPE_FORMAT_R8G8B8A8_UNORM);
si_change_format(dst, dst_level, &orig_info[1],
PIPE_FORMAT_R8G8B8A8_UNORM);
break;
case 8:
si_change_format(src, src_level, &orig_info[0],
PIPE_FORMAT_R16G16B16A16_UINT);
si_change_format(dst, dst_level, &orig_info[1],
PIPE_FORMAT_R16G16B16A16_UINT);
break;
case 16:
si_change_format(src, src_level, &orig_info[0],
PIPE_FORMAT_R32G32B32A32_UINT);
si_change_format(dst, dst_level, &orig_info[1],
PIPE_FORMAT_R32G32B32A32_UINT);
break;
default:
fprintf(stderr, "Unhandled format %s with blocksize %u\n",
util_format_short_name(src->format), blocksize);
assert(0);
}
restore_orig[0] = TRUE;
restore_orig[1] = TRUE;
}
}
/* Initialize the surface. */
util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
rdst->surface.level[dst_level].npix_x,
rdst->surface.level[dst_level].npix_y);
/* Initialize the sampler view. */
util_blitter_default_src_texture(&src_templ, src, src_level);
src_view = ctx->create_sampler_view(ctx, src, &src_templ);
u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
abs(src_box->depth), &dstbox);
/* Copy. */
si_blitter_begin(ctx, SI_COPY);
util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
src_view, src_box, src->width0, src->height0,
PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
si_blitter_end(ctx);
pipe_surface_reference(&dst_view, NULL);
pipe_sampler_view_reference(&src_view, NULL);
if (restore_orig[0])
si_reset_blittable_to_orig(src, src_level, &orig_info[0]);
if (restore_orig[1])
si_reset_blittable_to_orig(dst, dst_level, &orig_info[1]);
}
void r600_test_dma(struct r600_common_screen *rscreen)
{
struct pipe_screen *screen = &rscreen->b;
struct pipe_context *ctx = screen->context_create(screen, NULL, 0);
struct r600_common_context *rctx = (struct r600_common_context*)ctx;
uint64_t max_alloc_size;
unsigned i, iterations, num_partial_copies, max_levels, max_tex_side;
unsigned num_pass = 0, num_fail = 0;
max_levels = screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS);
max_tex_side = 1 << (max_levels - 1);
/* Max 128 MB allowed for both textures. */
max_alloc_size = 128 * 1024 * 1024;
/* the seed for random test parameters */
srand(0x9b47d95b);
/* the seed for random pixel data */
s_rand_xorshift128plus(seed_xorshift128plus, false);
iterations = 1000000000; /* just kill it when you are bored */
num_partial_copies = 30;
/* These parameters are randomly generated per test:
* - whether to do one whole-surface copy or N partial copies per test
* - which tiling modes to use (LINEAR_ALIGNED, 1D, 2D)
* - which texture dimensions to use
* - whether to use VRAM (all tiling modes) and GTT (staging, linear
* only) allocations
* - random initial pixels in src
* - generate random subrectangle copies for partial blits
*/
for (i = 0; i < iterations; i++) {
struct pipe_resource tsrc = {}, tdst = {}, *src, *dst;
struct r600_texture *rdst;
struct r600_texture *rsrc;
struct cpu_texture src_cpu, dst_cpu;
unsigned bpp, max_width, max_height, max_depth, j, num;
unsigned gfx_blits = 0, dma_blits = 0, max_tex_side_gen;
unsigned max_tex_layers;
bool pass;
bool do_partial_copies = rand() & 1;
/* generate a random test case */
tsrc.target = tdst.target = PIPE_TEXTURE_2D_ARRAY;
tsrc.depth0 = tdst.depth0 = 1;
bpp = 1 << (rand() % 5);
tsrc.format = tdst.format = get_format_from_bpp(bpp);
max_tex_side_gen = generate_max_tex_side(max_tex_side);
max_tex_layers = rand() % 4 ? 1 : 5;
tsrc.width0 = (rand() % max_tex_side_gen) + 1;
tsrc.height0 = (rand() % max_tex_side_gen) + 1;
tsrc.array_size = (rand() % max_tex_layers) + 1;
/* Have a 1/4 chance of getting power-of-two dimensions. */
if (rand() % 4 == 0) {
tsrc.width0 = util_next_power_of_two(tsrc.width0);
tsrc.height0 = util_next_power_of_two(tsrc.height0);
}
if (!do_partial_copies) {
/* whole-surface copies only, same dimensions */
tdst = tsrc;
} else {
max_tex_side_gen = generate_max_tex_side(max_tex_side);
max_tex_layers = rand() % 4 ? 1 : 5;
/* many partial copies, dimensions can be different */
tdst.width0 = (rand() % max_tex_side_gen) + 1;
tdst.height0 = (rand() % max_tex_side_gen) + 1;
tdst.array_size = (rand() % max_tex_layers) + 1;
/* Have a 1/4 chance of getting power-of-two dimensions. */
if (rand() % 4 == 0) {
tdst.width0 = util_next_power_of_two(tdst.width0);
tdst.height0 = util_next_power_of_two(tdst.height0);
}
}
/* check texture sizes */
if ((uint64_t)tsrc.width0 * tsrc.height0 * tsrc.array_size * bpp +
(uint64_t)tdst.width0 * tdst.height0 * tdst.array_size * bpp >
max_alloc_size) {
/* too large, try again */
i--;
continue;
}
/* VRAM + the tiling mode depends on dimensions (3/4 of cases),
* or GTT + linear only (1/4 of cases)
*/
tsrc.usage = rand() % 4 ? PIPE_USAGE_DEFAULT : PIPE_USAGE_STAGING;
tdst.usage = rand() % 4 ? PIPE_USAGE_DEFAULT : PIPE_USAGE_STAGING;
/* Allocate textures (both the GPU and CPU copies).
* The CPU will emulate what the GPU should be doing.
*/
src = screen->resource_create(screen, &tsrc);
dst = screen->resource_create(screen, &tdst);
assert(src);
assert(dst);
rdst = (struct r600_texture*)dst;
rsrc = (struct r600_texture*)src;
alloc_cpu_texture(&src_cpu, &tsrc, bpp);
alloc_cpu_texture(&dst_cpu, &tdst, bpp);
printf("%4u: dst = (%5u x %5u x %u, %s), "
" src = (%5u x %5u x %u, %s), bpp = %2u, ",
i, tdst.width0, tdst.height0, tdst.array_size,
array_mode_to_string(rscreen, &rdst->surface),
tsrc.width0, tsrc.height0, tsrc.array_size,
array_mode_to_string(rscreen, &rsrc->surface), bpp);
fflush(stdout);
/* set src pixels */
set_random_pixels(ctx, src, &src_cpu);
/* clear dst pixels */
rctx->clear_buffer(ctx, dst, 0, rdst->surface.surf_size, 0, true);
memset(dst_cpu.ptr, 0, dst_cpu.layer_stride * tdst.array_size);
/* preparation */
max_width = MIN2(tsrc.width0, tdst.width0);
max_height = MIN2(tsrc.height0, tdst.height0);
max_depth = MIN2(tsrc.array_size, tdst.array_size);
num = do_partial_copies ? num_partial_copies : 1;
for (j = 0; j < num; j++) {
int width, height, depth;
int srcx, srcy, srcz, dstx, dsty, dstz;
struct pipe_box box;
unsigned old_num_draw_calls = rctx->num_draw_calls;
unsigned old_num_dma_calls = rctx->num_dma_calls;
if (!do_partial_copies) {
/* copy whole src to dst */
width = max_width;
height = max_height;
depth = max_depth;
srcx = srcy = srcz = dstx = dsty = dstz = 0;
} else {
/* random sub-rectangle copies from src to dst */
depth = (rand() % max_depth) + 1;
srcz = rand() % (tsrc.array_size - depth + 1);
dstz = rand() % (tdst.array_size - depth + 1);
/* special code path to hit the tiled partial copies */
if (!rsrc->surface.is_linear &&
!rdst->surface.is_linear &&
rand() & 1) {
if (max_width < 8 || max_height < 8)
continue;
width = ((rand() % (max_width / 8)) + 1) * 8;
height = ((rand() % (max_height / 8)) + 1) * 8;
srcx = rand() % (tsrc.width0 - width + 1) & ~0x7;
srcy = rand() % (tsrc.height0 - height + 1) & ~0x7;
dstx = rand() % (tdst.width0 - width + 1) & ~0x7;
dsty = rand() % (tdst.height0 - height + 1) & ~0x7;
} else {
/* just make sure that it doesn't divide by zero */
assert(max_width > 0 && max_height > 0);
width = (rand() % max_width) + 1;
height = (rand() % max_height) + 1;
srcx = rand() % (tsrc.width0 - width + 1);
srcy = rand() % (tsrc.height0 - height + 1);
dstx = rand() % (tdst.width0 - width + 1);
dsty = rand() % (tdst.height0 - height + 1);
}
/* special code path to hit out-of-bounds reads in L2T */
if (rsrc->surface.is_linear &&
!rdst->surface.is_linear &&
rand() % 4 == 0) {
srcx = 0;
srcy = 0;
srcz = 0;
}
}
/* GPU copy */
u_box_3d(srcx, srcy, srcz, width, height, depth, &box);
rctx->dma_copy(ctx, dst, 0, dstx, dsty, dstz, src, 0, &box);
/* See which engine was used. */
gfx_blits += rctx->num_draw_calls > old_num_draw_calls;
dma_blits += rctx->num_dma_calls > old_num_dma_calls;
/* CPU copy */
util_copy_box(dst_cpu.ptr, tdst.format, dst_cpu.stride,
dst_cpu.layer_stride,
dstx, dsty, dstz, width, height, depth,
src_cpu.ptr, src_cpu.stride,
src_cpu.layer_stride,
srcx, srcy, srcz);
}
pass = compare_textures(ctx, dst, &dst_cpu, bpp);
if (pass)
num_pass++;
else
num_fail++;
printf("BLITs: GFX = %2u, DMA = %2u, %s [%u/%u]\n",
gfx_blits, dma_blits, pass ? "pass" : "fail",
num_pass, num_pass+num_fail);
/* cleanup */
pipe_resource_reference(&src, NULL);
pipe_resource_reference(&dst, NULL);
free(src_cpu.ptr);
free(dst_cpu.ptr);
}
ctx->destroy(ctx);
exit(0);
}
void si_resource_copy_region(struct pipe_context *ctx,
struct pipe_resource *dst,
unsigned dst_level,
unsigned dstx, unsigned dsty, unsigned dstz,
struct pipe_resource *src,
unsigned src_level,
const struct pipe_box *src_box)
{
struct si_context *sctx = (struct si_context *)ctx;
struct pipe_surface *dst_view, dst_templ;
struct pipe_sampler_view src_templ, *src_view;
unsigned dst_width, dst_height, src_width0, src_height0;
unsigned src_force_level = 0;
struct pipe_box sbox, dstbox;
/* Handle buffers first. */
if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width);
return;
}
assert(u_max_sample(dst) == u_max_sample(src));
/* The driver doesn't decompress resources automatically while
* u_blitter is rendering. */
si_decompress_subresource(ctx, src, PIPE_MASK_RGBAZS, src_level,
src_box->z, src_box->z + src_box->depth - 1);
dst_width = u_minify(dst->width0, dst_level);
dst_height = u_minify(dst->height0, dst_level);
src_width0 = src->width0;
src_height0 = src->height0;
util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
util_blitter_default_src_texture(&src_templ, src, src_level);
if (util_format_is_compressed(src->format) ||
util_format_is_compressed(dst->format)) {
unsigned blocksize = util_format_get_blocksize(src->format);
if (blocksize == 8)
src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
else
src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
dst_templ.format = src_templ.format;
dst_width = util_format_get_nblocksx(dst->format, dst_width);
dst_height = util_format_get_nblocksy(dst->format, dst_height);
src_width0 = util_format_get_nblocksx(src->format, src_width0);
src_height0 = util_format_get_nblocksy(src->format, src_height0);
dstx = util_format_get_nblocksx(dst->format, dstx);
dsty = util_format_get_nblocksy(dst->format, dsty);
sbox.x = util_format_get_nblocksx(src->format, src_box->x);
sbox.y = util_format_get_nblocksy(src->format, src_box->y);
sbox.z = src_box->z;
sbox.width = util_format_get_nblocksx(src->format, src_box->width);
sbox.height = util_format_get_nblocksy(src->format, src_box->height);
sbox.depth = src_box->depth;
src_box = &sbox;
src_force_level = src_level;
} else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src) ||
/* also *8_SNORM has precision issues, use UNORM instead */
util_format_is_snorm8(src->format)) {
if (util_format_is_subsampled_422(src->format)) {
src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
dst_width = util_format_get_nblocksx(dst->format, dst_width);
src_width0 = util_format_get_nblocksx(src->format, src_width0);
dstx = util_format_get_nblocksx(dst->format, dstx);
sbox = *src_box;
sbox.x = util_format_get_nblocksx(src->format, src_box->x);
sbox.width = util_format_get_nblocksx(src->format, src_box->width);
src_box = &sbox;
} else {
unsigned blocksize = util_format_get_blocksize(src->format);
switch (blocksize) {
case 1:
dst_templ.format = PIPE_FORMAT_R8_UNORM;
src_templ.format = PIPE_FORMAT_R8_UNORM;
break;
case 2:
dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
src_templ.format = PIPE_FORMAT_R8G8_UNORM;
break;
case 4:
dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
break;
case 8:
dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
break;
case 16:
dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
break;
default:
fprintf(stderr, "Unhandled format %s with blocksize %u\n",
util_format_short_name(src->format), blocksize);
assert(0);
}
}
}
/* Initialize the surface. */
dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
dst_width, dst_height);
/* Initialize the sampler view. */
src_view = si_create_sampler_view_custom(ctx, src, &src_templ,
src_width0, src_height0,
src_force_level);
u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
abs(src_box->depth), &dstbox);
/* Copy. */
si_blitter_begin(ctx, SI_COPY);
util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
src_view, src_box, src_width0, src_height0,
PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
FALSE);
si_blitter_end(ctx);
pipe_surface_reference(&dst_view, NULL);
pipe_sampler_view_reference(&src_view, NULL);
}
HRESULT NINE_WINAPI
NineVolume9_LockBox( struct NineVolume9 *This,
D3DLOCKED_BOX *pLockedVolume,
const D3DBOX *pBox,
DWORD Flags )
{
struct pipe_resource *resource = This->resource;
struct pipe_box box;
unsigned usage;
DBG("This=%p(%p) pLockedVolume=%p pBox=%p[%u..%u,%u..%u,%u..%u] Flags=%s\n",
This, This->base.container, pLockedVolume, pBox,
pBox ? pBox->Left : 0, pBox ? pBox->Right : 0,
pBox ? pBox->Top : 0, pBox ? pBox->Bottom : 0,
pBox ? pBox->Front : 0, pBox ? pBox->Back : 0,
nine_D3DLOCK_to_str(Flags));
/* check if it's already locked */
user_assert(This->lock_count == 0, D3DERR_INVALIDCALL);
/* set pBits to NULL after lock_count check */
user_assert(pLockedVolume, E_POINTER);
pLockedVolume->pBits = NULL;
user_assert(This->desc.Pool != D3DPOOL_DEFAULT ||
(This->desc.Usage & D3DUSAGE_DYNAMIC), D3DERR_INVALIDCALL);
user_assert(!((Flags & D3DLOCK_DISCARD) && (Flags & D3DLOCK_READONLY)),
D3DERR_INVALIDCALL);
if (pBox && compressed_format (This->desc.Format)) { /* For volume all pools are checked */
const unsigned w = util_format_get_blockwidth(This->info.format);
const unsigned h = util_format_get_blockheight(This->info.format);
user_assert((pBox->Left == 0 && pBox->Right == This->desc.Width &&
pBox->Top == 0 && pBox->Bottom == This->desc.Height) ||
(!(pBox->Left % w) && !(pBox->Right % w) &&
!(pBox->Top % h) && !(pBox->Bottom % h)),
D3DERR_INVALIDCALL);
}
if (Flags & D3DLOCK_DISCARD) {
usage = PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD_RANGE;
} else {
usage = (Flags & D3DLOCK_READONLY) ?
PIPE_TRANSFER_READ : PIPE_TRANSFER_READ_WRITE;
}
if (Flags & D3DLOCK_DONOTWAIT)
usage |= PIPE_TRANSFER_DONTBLOCK;
if (pBox) {
user_assert(pBox->Right > pBox->Left, D3DERR_INVALIDCALL);
user_assert(pBox->Bottom > pBox->Top, D3DERR_INVALIDCALL);
user_assert(pBox->Back > pBox->Front, D3DERR_INVALIDCALL);
user_assert(pBox->Right <= This->desc.Width, D3DERR_INVALIDCALL);
user_assert(pBox->Bottom <= This->desc.Height, D3DERR_INVALIDCALL);
user_assert(pBox->Back <= This->desc.Depth, D3DERR_INVALIDCALL);
d3dbox_to_pipe_box(&box, pBox);
if (u_box_clip_2d(&box, &box, This->desc.Width, This->desc.Height) < 0) {
DBG("Locked volume intersection empty.\n");
return D3DERR_INVALIDCALL;
}
} else {
u_box_3d(0, 0, 0, This->desc.Width, This->desc.Height, This->desc.Depth,
&box);
}
if (This->data_conversion) {
/* For now we only have uncompressed formats here */
pLockedVolume->RowPitch = This->stride_conversion;
pLockedVolume->SlicePitch = This->layer_stride_conversion;
pLockedVolume->pBits = This->data_conversion + box.z * This->layer_stride_conversion +
box.y * This->stride_conversion +
util_format_get_stride(This->format_conversion, box.x);
} else if (This->data) {
pLockedVolume->RowPitch = This->stride;
pLockedVolume->SlicePitch = This->layer_stride;
pLockedVolume->pBits =
NineVolume9_GetSystemMemPointer(This, box.x, box.y, box.z);
} else {
pLockedVolume->pBits =
This->pipe->transfer_map(This->pipe, resource, This->level, usage,
&box, &This->transfer);
if (!This->transfer) {
if (Flags & D3DLOCK_DONOTWAIT)
return D3DERR_WASSTILLDRAWING;
return D3DERR_DRIVERINTERNALERROR;
}
pLockedVolume->RowPitch = This->transfer->stride;
pLockedVolume->SlicePitch = This->transfer->layer_stride;
}
if (!(Flags & (D3DLOCK_NO_DIRTY_UPDATE | D3DLOCK_READONLY))) {
NineVolume9_MarkContainerDirty(This);
NineVolume9_AddDirtyRegion(This, &box);
}
++This->lock_count;
return D3D_OK;
}
