/* Break the COW tie to the pbo. Both the pbo and the region end up
* with a copy of the data.
*/
void
intel_region_cow(struct intel_context *intel, struct intel_region *region)
{
struct intel_buffer_object *pbo = region->pbo;
GLboolean was_locked = intel->locked;
if (intel == NULL)
return;
intel_region_release_pbo(intel, region);
assert(region->cpp * region->pitch * region->height == pbo->Base.Size);
DBG("%s (%d bytes)\n", __FUNCTION__, pbo->Base.Size);
/* Now blit from the texture buffer to the new buffer:
*/
was_locked = intel->locked;
if (!was_locked)
LOCK_HARDWARE(intel);
intelEmitCopyBlit(intel,
region->cpp,
region->pitch, region->buffer, 0, region->tiling,
region->pitch, pbo->buffer, 0, region->tiling,
0, 0, 0, 0,
region->pitch, region->height,
GL_COPY);
if (!was_locked)
UNLOCK_HARDWARE(intel);
}
/* Copy rectangular sub-regions. Need better logic about when to
* push buffers into AGP - will currently do so whenever possible.
*/
void
intel_region_copy(struct intel_context *intel,
struct intel_region *dst,
GLuint dst_offset,
GLuint dstx, GLuint dsty,
struct intel_region *src,
GLuint src_offset,
GLuint srcx, GLuint srcy, GLuint width, GLuint height)
{
DBG("%s\n", __FUNCTION__);
if (intel == NULL)
return;
if (dst->pbo) {
if (dstx == 0 &&
dsty == 0 && width == dst->pitch && height == dst->height)
intel_region_release_pbo(intel, dst);
else
intel_region_cow(intel, dst);
}
assert(src->cpp == dst->cpp);
intelEmitCopyBlit(intel,
dst->cpp,
src->pitch, src->buffer, src_offset, src->tiling,
dst->pitch, dst->buffer, dst_offset, dst->tiling,
srcx, srcy, dstx, dsty, width, height,
GL_COPY);
}
/* Break the COW tie to the pbo. Both the pbo and the region end up
* with a copy of the data.
*/
void
intel_region_cow(struct intel_context *intel, struct intel_region *region)
{
struct intel_buffer_object *pbo = region->pbo;
GLboolean ok;
intel_region_release_pbo(intel, region);
assert(region->cpp * region->pitch * region->height == pbo->Base.Size);
_DBG("%s %p (%d bytes)\n", __FUNCTION__, region, pbo->Base.Size);
/* Now blit from the texture buffer to the new buffer:
*/
LOCK_HARDWARE(intel);
ok = intelEmitCopyBlit(intel,
region->cpp,
region->pitch, pbo->buffer, 0, region->tiling,
region->pitch, region->buffer, 0, region->tiling,
0, 0, 0, 0,
region->pitch, region->height,
GL_COPY);
assert(ok);
UNLOCK_HARDWARE(intel);
}
/* We don't have a memmove-type blit like some other hardware, so we'll do a
* rectangular blit covering a large space, then emit 1-scanline blit at the
* end to cover the last if we need.
*/
void
intel_emit_linear_blit(struct intel_context *intel,
drm_intel_bo *dst_bo,
unsigned int dst_offset,
drm_intel_bo *src_bo,
unsigned int src_offset,
unsigned int size)
{
struct gl_context *ctx = &intel->ctx;
GLuint pitch, height;
bool ok;
/* The pitch given to the GPU must be DWORD aligned, and
* we want width to match pitch. Max width is (1 << 15 - 1),
* rounding that down to the nearest DWORD is 1 << 15 - 4
*/
pitch = ROUND_DOWN_TO(MIN2(size, (1 << 15) - 1), 4);
height = (pitch == 0) ? 1 : size / pitch;
ok = intelEmitCopyBlit(intel, 1,
pitch, src_bo, src_offset, I915_TILING_NONE,
pitch, dst_bo, dst_offset, I915_TILING_NONE,
0, 0, /* src x/y */
0, 0, /* dst x/y */
pitch, height, /* w, h */
GL_COPY);
if (!ok)
_mesa_problem(ctx, "Failed to linear blit %dx%d\n", pitch, height);
src_offset += pitch * height;
dst_offset += pitch * height;
size -= pitch * height;
assert (size < (1 << 15));
pitch = ALIGN(size, 4);
if (size != 0) {
ok = intelEmitCopyBlit(intel, 1,
pitch, src_bo, src_offset, I915_TILING_NONE,
pitch, dst_bo, dst_offset, I915_TILING_NONE,
0, 0, /* src x/y */
0, 0, /* dst x/y */
size, 1, /* w, h */
GL_COPY);
if (!ok)
_mesa_problem(ctx, "Failed to linear blit %dx%d\n", size, 1);
}
}
/* XXX: Do this for TexSubImage also:
*/
static GLboolean
try_pbo_upload(struct intel_context *intel,
struct intel_texture_image *intelImage,
const struct gl_pixelstore_attrib *unpack,
GLint internalFormat,
GLint width, GLint height,
GLenum format, GLenum type, const void *pixels)
{
struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
GLuint src_offset, src_stride;
GLuint dst_offset, dst_stride;
if (!pbo ||
intel->ctx._ImageTransferState ||
unpack->SkipPixels || unpack->SkipRows) {
_mesa_printf("%s: failure 1\n", __FUNCTION__);
return GL_FALSE;
}
src_offset = (GLuint) pixels;
if (unpack->RowLength > 0)
src_stride = unpack->RowLength;
else
src_stride = width;
dst_offset = intel_miptree_image_offset(intelImage->mt,
intelImage->face,
intelImage->level);
dst_stride = intelImage->mt->pitch;
intelFlush(&intel->ctx);
LOCK_HARDWARE(intel);
{
struct _DriBufferObject *src_buffer =
intel_bufferobj_buffer(intel, pbo, INTEL_READ);
struct _DriBufferObject *dst_buffer =
intel_region_buffer(intel->intelScreen, intelImage->mt->region,
INTEL_WRITE_FULL);
intelEmitCopyBlit(intel,
intelImage->mt->cpp,
src_stride, src_buffer, src_offset,
dst_stride, dst_buffer, dst_offset,
0, 0, 0, 0, width, height,
GL_COPY);
intel_batchbuffer_flush(intel->batch);
}
UNLOCK_HARDWARE(intel);
return GL_TRUE;
}
/* We don't have a memmove-type blit like some other hardware, so we'll do a
* rectangular blit covering a large space, then emit 1-scanline blit at the
* end to cover the last if we need.
*/
void
intel_emit_linear_blit(struct intel_context *intel,
drm_intel_bo *dst_bo,
unsigned int dst_offset,
drm_intel_bo *src_bo,
unsigned int src_offset,
unsigned int size)
{
GLuint pitch, height;
GLboolean ok;
/* The pitch given to the GPU must be DWORD aligned, and
* we want width to match pitch. Max width is (1 << 15 - 1),
* rounding that down to the nearest DWORD is 1 << 15 - 4
*/
pitch = MIN2(size, (1 << 15) - 4);
height = size / pitch;
ok = intelEmitCopyBlit(intel, 1,
pitch, src_bo, src_offset, I915_TILING_NONE,
pitch, dst_bo, dst_offset, I915_TILING_NONE,
0, 0, /* src x/y */
0, 0, /* dst x/y */
pitch, height, /* w, h */
GL_COPY);
assert(ok);
src_offset += pitch * height;
dst_offset += pitch * height;
size -= pitch * height;
assert (size < (1 << 15));
assert ((size & 3) == 0); /* Pitch must be DWORD aligned */
if (size != 0) {
ok = intelEmitCopyBlit(intel, 1,
size, src_bo, src_offset, I915_TILING_NONE,
size, dst_bo, dst_offset, I915_TILING_NONE,
0, 0, /* src x/y */
0, 0, /* dst x/y */
size, 1, /* w, h */
GL_COPY);
assert(ok);
}
}
/**
* Implements a rectangular block transfer (blit) of pixels between two
* miptrees.
*
* Our blitter can operate on 1, 2, or 4-byte-per-pixel data, with generous,
* but limited, pitches and sizes allowed.
*
* The src/dst coordinates are relative to the given level/slice of the
* miptree.
*
* If @src_flip or @dst_flip is set, then the rectangle within that miptree
* will be inverted (including scanline order) when copying. This is common
* in GL when copying between window system and user-created
* renderbuffers/textures.
*/
bool
intel_miptree_blit(struct intel_context *intel,
struct intel_mipmap_tree *src_mt,
int src_level, int src_slice,
uint32_t src_x, uint32_t src_y, bool src_flip,
struct intel_mipmap_tree *dst_mt,
int dst_level, int dst_slice,
uint32_t dst_x, uint32_t dst_y, bool dst_flip,
uint32_t width, uint32_t height,
GLenum logicop)
{
/* No sRGB decode or encode is done by the hardware blitter, which is
* consistent with what we want in the callers (glCopyTexSubImage(),
* glBlitFramebuffer(), texture validation, etc.).
*/
gl_format src_format = _mesa_get_srgb_format_linear(src_mt->format);
gl_format dst_format = _mesa_get_srgb_format_linear(dst_mt->format);
/* The blitter doesn't support doing any format conversions. We do also
* support blitting ARGB8888 to XRGB8888 (trivial, the values dropped into
* the X channel don't matter), and XRGB8888 to ARGB8888 by setting the A
* channel to 1.0 at the end.
*/
if (src_format != dst_format &&
((src_format != MESA_FORMAT_ARGB8888 &&
src_format != MESA_FORMAT_XRGB8888) ||
(dst_format != MESA_FORMAT_ARGB8888 &&
dst_format != MESA_FORMAT_XRGB8888))) {
perf_debug("%s: Can't use hardware blitter from %s to %s, "
"falling back.\n", __FUNCTION__,
_mesa_get_format_name(src_format),
_mesa_get_format_name(dst_format));
return false;
}
/* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
* Data Size Limitations):
*
* The BLT engine is capable of transferring very large quantities of
* graphics data. Any graphics data read from and written to the
* destination is permitted to represent a number of pixels that
* occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
* at the destination. The maximum number of pixels that may be
* represented per scan line’s worth of graphics data depends on the
* color depth.
*
* Furthermore, intelEmitCopyBlit (which is called below) uses a signed
* 16-bit integer to represent buffer pitch, so it can only handle buffer
* pitches < 32k.
*
* As a result of these two limitations, we can only use the blitter to do
* this copy when the region's pitch is less than 32k.
*/
if (src_mt->region->pitch > 32768 ||
dst_mt->region->pitch > 32768) {
perf_debug("Falling back due to >32k pitch\n");
return false;
}
if (src_flip)
src_y = src_mt->level[src_level].height - src_y - height;
if (dst_flip)
dst_y = dst_mt->level[dst_level].height - dst_y - height;
int src_pitch = src_mt->region->pitch;
if (src_flip != dst_flip)
src_pitch = -src_pitch;
uint32_t src_image_x, src_image_y;
intel_miptree_get_image_offset(src_mt, src_level, src_slice,
&src_image_x, &src_image_y);
src_x += src_image_x;
src_y += src_image_y;
uint32_t dst_image_x, dst_image_y;
intel_miptree_get_image_offset(dst_mt, dst_level, dst_slice,
&dst_image_x, &dst_image_y);
dst_x += dst_image_x;
dst_y += dst_image_y;
if (!intelEmitCopyBlit(intel,
src_mt->cpp,
src_pitch,
src_mt->region->bo, src_mt->offset,
src_mt->region->tiling,
dst_mt->region->pitch,
dst_mt->region->bo, dst_mt->offset,
dst_mt->region->tiling,
src_x, src_y,
dst_x, dst_y,
width, height,
logicop)) {
return false;
}
if (src_mt->format == MESA_FORMAT_XRGB8888 &&
dst_mt->format == MESA_FORMAT_ARGB8888) {
intel_miptree_set_alpha_to_one(intel, dst_mt,
dst_x, dst_y,
width, height);
}
return true;
}
bool
intel_copy_texsubimage(struct intel_context *intel,
struct intel_texture_image *intelImage,
GLint dstx, GLint dsty,
struct intel_renderbuffer *irb,
GLint x, GLint y, GLsizei width, GLsizei height)
{
struct gl_context *ctx = &intel->ctx;
struct intel_region *region;
const GLenum internalFormat = intelImage->base.Base.InternalFormat;
bool copy_supported = false;
bool copy_supported_with_alpha_override = false;
intel_prepare_render(intel);
if (!intelImage->mt || !irb || !irb->mt) {
if (unlikely(INTEL_DEBUG & DEBUG_PERF))
fprintf(stderr, "%s fail %p %p (0x%08x)\n",
__FUNCTION__, intelImage->mt, irb, internalFormat);
return false;
} else {
region = irb->mt->region;
assert(region);
}
/* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
* Data Size Limitations):
*
* The BLT engine is capable of transferring very large quantities of
* graphics data. Any graphics data read from and written to the
* destination is permitted to represent a number of pixels that
* occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
* at the destination. The maximum number of pixels that may be
* represented per scan line’s worth of graphics data depends on the
* color depth.
*
* Furthermore, intelEmitCopyBlit (which is called below) uses a signed
* 16-bit integer to represent buffer pitch, so it can only handle buffer
* pitches < 32k.
*
* As a result of these two limitations, we can only use the blitter to do
* this copy when the region's pitch is less than 32k.
*/
if (region->pitch >= 32768)
return false;
if (intelImage->base.Base.TexObject->Target == GL_TEXTURE_1D_ARRAY ||
intelImage->base.Base.TexObject->Target == GL_TEXTURE_2D_ARRAY) {
perf_debug("no support for array textures\n");
}
copy_supported = intelImage->base.Base.TexFormat == intel_rb_format(irb);
/* Converting ARGB8888 to XRGB8888 is trivial: ignore the alpha bits */
if (intel_rb_format(irb) == MESA_FORMAT_ARGB8888 &&
intelImage->base.Base.TexFormat == MESA_FORMAT_XRGB8888) {
copy_supported = true;
}
/* Converting XRGB8888 to ARGB8888 requires setting the alpha bits to 1.0 */
if (intel_rb_format(irb) == MESA_FORMAT_XRGB8888 &&
intelImage->base.Base.TexFormat == MESA_FORMAT_ARGB8888) {
copy_supported_with_alpha_override = true;
}
if (!copy_supported && !copy_supported_with_alpha_override) {
if (unlikely(INTEL_DEBUG & DEBUG_PERF))
fprintf(stderr, "%s mismatched formats %s, %s\n",
__FUNCTION__,
_mesa_get_format_name(intelImage->base.Base.TexFormat),
_mesa_get_format_name(intel_rb_format(irb)));
return false;
}
{
GLuint image_x, image_y;
GLshort src_pitch;
/* get dest x/y in destination texture */
intel_miptree_get_image_offset(intelImage->mt,
intelImage->base.Base.Level,
intelImage->base.Base.Face,
&image_x, &image_y);
/* The blitter can't handle Y-tiled buffers. */
if (intelImage->mt->region->tiling == I915_TILING_Y) {
return false;
}
if (_mesa_is_winsys_fbo(ctx->ReadBuffer)) {
/* Flip vertical orientation for system framebuffers */
y = ctx->ReadBuffer->Height - (y + height);
src_pitch = -region->pitch;
} else {
/* reading from a FBO, y is already oriented the way we like */
src_pitch = region->pitch;
}
/* blit from src buffer to texture */
if (!intelEmitCopyBlit(intel,
intelImage->mt->cpp,
src_pitch,
region->bo,
0,
region->tiling,
intelImage->mt->region->pitch,
intelImage->mt->region->bo,
0,
intelImage->mt->region->tiling,
irb->draw_x + x, irb->draw_y + y,
image_x + dstx, image_y + dsty,
width, height,
GL_COPY)) {
return false;
}
}
if (copy_supported_with_alpha_override)
intel_set_teximage_alpha_to_one(ctx, intelImage);
return true;
}
/* XXX: Do this for TexSubImage also:
*/
static bool
try_pbo_upload(struct gl_context *ctx,
struct gl_texture_image *image,
const struct gl_pixelstore_attrib *unpack,
GLenum format, GLenum type, const void *pixels)
{
struct intel_texture_image *intelImage = intel_texture_image(image);
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
GLuint src_offset, src_stride;
GLuint dst_x, dst_y;
drm_intel_bo *dst_buffer, *src_buffer;
if (!_mesa_is_bufferobj(unpack->BufferObj))
return false;
DBG("trying pbo upload\n");
if (intel->ctx._ImageTransferState ||
unpack->SkipPixels || unpack->SkipRows) {
DBG("%s: image transfer\n", __FUNCTION__);
return false;
}
if (!_mesa_format_matches_format_and_type(image->TexFormat,
format, type, false)) {
DBG("%s: format mismatch (upload to %s with format 0x%x, type 0x%x)\n",
__FUNCTION__, _mesa_get_format_name(image->TexFormat),
format, type);
return false;
}
ctx->Driver.AllocTextureImageBuffer(ctx, image);
if (!intelImage->mt) {
DBG("%s: no miptree\n", __FUNCTION__);
return false;
}
if (image->TexObject->Target == GL_TEXTURE_1D_ARRAY ||
image->TexObject->Target == GL_TEXTURE_2D_ARRAY) {
DBG("%s: no support for array textures\n", __FUNCTION__);
return false;
}
dst_buffer = intelImage->mt->region->bo;
src_buffer = intel_bufferobj_source(intel, pbo, 64, &src_offset);
/* note: potential 64-bit ptr to 32-bit int cast */
src_offset += (GLuint) (unsigned long) pixels;
if (unpack->RowLength > 0)
src_stride = unpack->RowLength;
else
src_stride = image->Width;
src_stride *= intelImage->mt->region->cpp;
intel_miptree_get_image_offset(intelImage->mt, intelImage->base.Base.Level,
intelImage->base.Base.Face,
&dst_x, &dst_y);
if (!intelEmitCopyBlit(intel,
intelImage->mt->cpp,
src_stride, src_buffer,
src_offset, false,
intelImage->mt->region->pitch, dst_buffer, 0,
intelImage->mt->region->tiling,
0, 0, dst_x, dst_y, image->Width, image->Height,
GL_COPY)) {
DBG("%s: blit failed\n", __FUNCTION__);
return false;
}
DBG("%s: success\n", __FUNCTION__);
return true;
}
static bool
copy_image_with_blitter(struct brw_context *brw,
struct intel_mipmap_tree *src_mt, int src_level,
int src_x, int src_y, int src_z,
struct intel_mipmap_tree *dst_mt, int dst_level,
int dst_x, int dst_y, int dst_z,
int src_width, int src_height)
{
GLuint bw, bh;
uint32_t src_image_x, src_image_y, dst_image_x, dst_image_y;
/* The blitter doesn't understand multisampling at all. */
if (src_mt->num_samples > 0 || dst_mt->num_samples > 0)
return false;
if (src_mt->format == MESA_FORMAT_S_UINT8)
return false;
/* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
* Data Size Limitations):
*
* The BLT engine is capable of transferring very large quantities of
* graphics data. Any graphics data read from and written to the
* destination is permitted to represent a number of pixels that
* occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
* at the destination. The maximum number of pixels that may be
* represented per scan line’s worth of graphics data depends on the
* color depth.
*
* Furthermore, intelEmitCopyBlit (which is called below) uses a signed
* 16-bit integer to represent buffer pitch, so it can only handle buffer
* pitches < 32k.
*
* As a result of these two limitations, we can only use the blitter to do
* this copy when the miptree's pitch is less than 32k.
*/
if (src_mt->pitch >= 32768 ||
dst_mt->pitch >= 32768) {
perf_debug("Falling back due to >=32k pitch\n");
return false;
}
intel_miptree_get_image_offset(src_mt, src_level, src_z,
&src_image_x, &src_image_y);
if (_mesa_is_format_compressed(src_mt->format)) {
_mesa_get_format_block_size(src_mt->format, &bw, &bh);
assert(src_x % bw == 0);
assert(src_y % bh == 0);
assert(src_width % bw == 0);
assert(src_height % bh == 0);
src_x /= (int)bw;
src_y /= (int)bh;
src_width /= (int)bw;
src_height /= (int)bh;
}
src_x += src_image_x;
src_y += src_image_y;
intel_miptree_get_image_offset(dst_mt, dst_level, dst_z,
&dst_image_x, &dst_image_y);
if (_mesa_is_format_compressed(dst_mt->format)) {
_mesa_get_format_block_size(dst_mt->format, &bw, &bh);
assert(dst_x % bw == 0);
assert(dst_y % bh == 0);
dst_x /= (int)bw;
dst_y /= (int)bh;
}
dst_x += dst_image_x;
dst_y += dst_image_y;
return intelEmitCopyBlit(brw,
src_mt->cpp,
src_mt->pitch,
src_mt->bo, src_mt->offset,
src_mt->tiling,
src_mt->tr_mode,
dst_mt->pitch,
dst_mt->bo, dst_mt->offset,
dst_mt->tiling,
dst_mt->tr_mode,
src_x, src_y,
dst_x, dst_y,
src_width, src_height,
GL_COPY);
}
static GLboolean
do_blit_readpixels(GLcontext * ctx,
GLint x, GLint y, GLsizei width, GLsizei height,
GLenum format, GLenum type,
const struct gl_pixelstore_attrib *pack, GLvoid * pixels)
{
struct intel_context *intel = intel_context(ctx);
struct intel_region *src = intel_readbuf_region(intel);
struct intel_buffer_object *dst = intel_buffer_object(pack->BufferObj);
GLuint dst_offset;
GLuint rowLength;
drm_intel_bo *dst_buffer;
GLboolean all;
GLint dst_x, dst_y;
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s\n", __FUNCTION__);
if (!src)
return GL_FALSE;
if (!_mesa_is_bufferobj(pack->BufferObj)) {
/* PBO only for now:
*/
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s - not PBO\n", __FUNCTION__);
return GL_FALSE;
}
if (ctx->_ImageTransferState ||
!intel_check_blit_format(src, format, type)) {
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s - bad format for blit\n", __FUNCTION__);
return GL_FALSE;
}
if (pack->Alignment != 1 || pack->SwapBytes || pack->LsbFirst) {
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s: bad packing params\n", __FUNCTION__);
return GL_FALSE;
}
if (pack->RowLength > 0)
rowLength = pack->RowLength;
else
rowLength = width;
if (pack->Invert) {
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s: MESA_PACK_INVERT not done yet\n", __FUNCTION__);
return GL_FALSE;
}
else {
if (ctx->ReadBuffer->Name == 0)
rowLength = -rowLength;
}
dst_offset = (GLintptr) _mesa_image_address(2, pack, pixels, width, height,
format, type, 0, 0, 0);
if (!_mesa_clip_copytexsubimage(ctx,
&dst_x, &dst_y,
&x, &y,
&width, &height)) {
return GL_TRUE;
}
intel_prepare_render(intel);
all = (width * height * src->cpp == dst->Base.Size &&
x == 0 && dst_offset == 0);
dst_x = 0;
dst_y = 0;
dst_buffer = intel_bufferobj_buffer(intel, dst,
all ? INTEL_WRITE_FULL :
INTEL_WRITE_PART);
if (ctx->ReadBuffer->Name == 0)
y = ctx->ReadBuffer->Height - (y + height);
if (!intelEmitCopyBlit(intel,
src->cpp,
src->pitch, src->buffer, 0, src->tiling,
rowLength, dst_buffer, dst_offset, GL_FALSE,
x, y,
dst_x, dst_y,
width, height,
GL_COPY)) {
return GL_FALSE;
}
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s - DONE\n", __FUNCTION__);
return GL_TRUE;
}
static GLboolean
do_copy_texsubimage(struct intel_context *intel,
struct intel_texture_image *intelImage,
GLenum internalFormat,
GLint dstx, GLint dsty,
GLint x, GLint y, GLsizei width, GLsizei height)
{
GLcontext *ctx = &intel->ctx;
const struct intel_region *src =
get_teximage_source(intel, internalFormat);
if (!intelImage->mt || !src) {
DBG("%s fail %p %p\n", __FUNCTION__, intelImage->mt, src);
return GL_FALSE;
}
intelFlush(ctx);
LOCK_HARDWARE(intel);
{
GLuint image_offset = intel_miptree_image_offset(intelImage->mt,
intelImage->face,
intelImage->level);
const GLint orig_x = x;
const GLint orig_y = y;
const struct gl_framebuffer *fb = ctx->DrawBuffer;
if (_mesa_clip_to_region(fb->_Xmin, fb->_Ymin, fb->_Xmax, fb->_Ymax,
&x, &y, &width, &height)) {
/* Update dst for clipped src. Need to also clip the source rect.
*/
dstx += x - orig_x;
dsty += y - orig_y;
if (ctx->ReadBuffer->Name == 0) {
/* reading from a window, adjust x, y */
__DRIdrawablePrivate *dPriv = intel->driDrawable;
GLuint window_y;
/* window_y = position of window on screen if y=0=bottom */
window_y = intel->intelScreen->height - (dPriv->y + dPriv->h);
y = window_y + y;
x += dPriv->x;
}
else {
/* reading from a FBO */
/* invert Y */
y = ctx->ReadBuffer->Height - y - 1;
}
/* A bit of fiddling to get the blitter to work with -ve
* pitches. But we get a nice inverted blit this way, so it's
* worth it:
*/
intelEmitCopyBlit(intel,
intelImage->mt->cpp,
-src->pitch,
src->buffer,
src->height * src->pitch * src->cpp,
intelImage->mt->pitch,
intelImage->mt->region->buffer,
image_offset,
x, y + height, dstx, dsty, width, height,
GL_COPY); /* ? */
intel_batchbuffer_flush(intel->batch);
}
}
UNLOCK_HARDWARE(intel);
#if 0
/* GL_SGIS_generate_mipmap -- this can be accelerated now.
* XXX Add a ctx->Driver.GenerateMipmaps() function?
*/
if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
intel_generate_mipmap(ctx, target,
&ctx->Texture.Unit[ctx->Texture.CurrentUnit],
texObj);
}
#endif
return GL_TRUE;
}
bool
intel_copy_texsubimage(struct intel_context *intel,
struct intel_texture_image *intelImage,
GLint dstx, GLint dsty,
struct intel_renderbuffer *irb,
GLint x, GLint y, GLsizei width, GLsizei height)
{
struct gl_context *ctx = &intel->ctx;
struct intel_region *region;
const GLenum internalFormat = intelImage->base.Base.InternalFormat;
bool copy_supported = false;
bool copy_supported_with_alpha_override = false;
intel_prepare_render(intel);
if (!intelImage->mt || !irb || !irb->mt) {
if (unlikely(INTEL_DEBUG & DEBUG_FALLBACKS))
fprintf(stderr, "%s fail %p %p (0x%08x)\n",
__FUNCTION__, intelImage->mt, irb, internalFormat);
return false;
} else {
region = irb->mt->region;
assert(region);
}
copy_supported = intelImage->base.Base.TexFormat == intel_rb_format(irb);
/* Converting ARGB8888 to XRGB8888 is trivial: ignore the alpha bits */
if (intel_rb_format(irb) == MESA_FORMAT_ARGB8888 &&
intelImage->base.Base.TexFormat == MESA_FORMAT_XRGB8888) {
copy_supported = true;
}
/* Converting XRGB8888 to ARGB8888 requires setting the alpha bits to 1.0 */
if (intel_rb_format(irb) == MESA_FORMAT_XRGB8888 &&
intelImage->base.Base.TexFormat == MESA_FORMAT_ARGB8888) {
copy_supported_with_alpha_override = true;
}
if (!copy_supported && !copy_supported_with_alpha_override) {
if (unlikely(INTEL_DEBUG & DEBUG_FALLBACKS))
fprintf(stderr, "%s mismatched formats %s, %s\n",
__FUNCTION__,
_mesa_get_format_name(intelImage->base.Base.TexFormat),
_mesa_get_format_name(intel_rb_format(irb)));
return false;
}
{
GLuint image_x, image_y;
GLshort src_pitch;
/* get dest x/y in destination texture */
intel_miptree_get_image_offset(intelImage->mt,
intelImage->base.Base.Level,
intelImage->base.Base.Face,
0,
&image_x, &image_y);
/* The blitter can't handle Y-tiled buffers. */
if (intelImage->mt->region->tiling == I915_TILING_Y) {
return false;
}
if (ctx->ReadBuffer->Name == 0) {
/* Flip vertical orientation for system framebuffers */
y = ctx->ReadBuffer->Height - (y + height);
src_pitch = -region->pitch;
} else {
/* reading from a FBO, y is already oriented the way we like */
src_pitch = region->pitch;
}
/* blit from src buffer to texture */
if (!intelEmitCopyBlit(intel,
intelImage->mt->cpp,
src_pitch,
region->bo,
0,
region->tiling,
intelImage->mt->region->pitch,
intelImage->mt->region->bo,
0,
intelImage->mt->region->tiling,
irb->draw_x + x, irb->draw_y + y,
image_x + dstx, image_y + dsty,
width, height,
GL_COPY)) {
return false;
}
}
if (copy_supported_with_alpha_override)
intel_set_teximage_alpha_to_one(ctx, intelImage);
return true;
}
/**
* Implements a rectangular block transfer (blit) of pixels between two
* miptrees.
*
* Our blitter can operate on 1, 2, or 4-byte-per-pixel data, with generous,
* but limited, pitches and sizes allowed.
*
* The src/dst coordinates are relative to the given level/slice of the
* miptree.
*
* If @src_flip or @dst_flip is set, then the rectangle within that miptree
* will be inverted (including scanline order) when copying. This is common
* in GL when copying between window system and user-created
* renderbuffers/textures.
*/
bool
intel_miptree_blit(struct brw_context *brw,
struct intel_mipmap_tree *src_mt,
int src_level, int src_slice,
uint32_t src_x, uint32_t src_y, bool src_flip,
struct intel_mipmap_tree *dst_mt,
int dst_level, int dst_slice,
uint32_t dst_x, uint32_t dst_y, bool dst_flip,
uint32_t width, uint32_t height,
GLenum logicop)
{
/* The blitter doesn't understand multisampling at all. */
if (src_mt->num_samples > 0 || dst_mt->num_samples > 0)
return false;
/* No sRGB decode or encode is done by the hardware blitter, which is
* consistent with what we want in the callers (glCopyTexSubImage(),
* glBlitFramebuffer(), texture validation, etc.).
*/
mesa_format src_format = _mesa_get_srgb_format_linear(src_mt->format);
mesa_format dst_format = _mesa_get_srgb_format_linear(dst_mt->format);
/* The blitter doesn't support doing any format conversions. We do also
* support blitting ARGB8888 to XRGB8888 (trivial, the values dropped into
* the X channel don't matter), and XRGB8888 to ARGB8888 by setting the A
* channel to 1.0 at the end.
*/
if (src_format != dst_format &&
((src_format != MESA_FORMAT_B8G8R8A8_UNORM &&
src_format != MESA_FORMAT_B8G8R8X8_UNORM) ||
(dst_format != MESA_FORMAT_B8G8R8A8_UNORM &&
dst_format != MESA_FORMAT_B8G8R8X8_UNORM))) {
perf_debug("%s: Can't use hardware blitter from %s to %s, "
"falling back.\n", __FUNCTION__,
_mesa_get_format_name(src_format),
_mesa_get_format_name(dst_format));
return false;
}
/* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
* Data Size Limitations):
*
* The BLT engine is capable of transferring very large quantities of
* graphics data. Any graphics data read from and written to the
* destination is permitted to represent a number of pixels that
* occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
* at the destination. The maximum number of pixels that may be
* represented per scan line’s worth of graphics data depends on the
* color depth.
*
* Furthermore, intelEmitCopyBlit (which is called below) uses a signed
* 16-bit integer to represent buffer pitch, so it can only handle buffer
* pitches < 32k.
*
* As a result of these two limitations, we can only use the blitter to do
* this copy when the miptree's pitch is less than 32k.
*/
if (src_mt->pitch >= 32768 ||
dst_mt->pitch >= 32768) {
perf_debug("Falling back due to >=32k pitch\n");
return false;
}
/* The blitter has no idea about HiZ or fast color clears, so we need to
* resolve the miptrees before we do anything.
*/
intel_miptree_slice_resolve_depth(brw, src_mt, src_level, src_slice);
intel_miptree_slice_resolve_depth(brw, dst_mt, dst_level, dst_slice);
intel_miptree_resolve_color(brw, src_mt);
intel_miptree_resolve_color(brw, dst_mt);
if (src_flip)
src_y = minify(src_mt->physical_height0, src_level - src_mt->first_level) - src_y - height;
if (dst_flip)
dst_y = minify(dst_mt->physical_height0, dst_level - dst_mt->first_level) - dst_y - height;
int src_pitch = src_mt->pitch;
if (src_flip != dst_flip)
src_pitch = -src_pitch;
uint32_t src_image_x, src_image_y;
intel_miptree_get_image_offset(src_mt, src_level, src_slice,
&src_image_x, &src_image_y);
src_x += src_image_x;
src_y += src_image_y;
/* The blitter interprets the 16-bit src x/y as a signed 16-bit value,
* where negative values are invalid. The values we're working with are
* unsigned, so make sure we don't overflow.
*/
if (src_x >= 32768 || src_y >= 32768) {
perf_debug("Falling back due to >=32k src offset (%d, %d)\n",
src_x, src_y);
return false;
}
uint32_t dst_image_x, dst_image_y;
intel_miptree_get_image_offset(dst_mt, dst_level, dst_slice,
&dst_image_x, &dst_image_y);
dst_x += dst_image_x;
dst_y += dst_image_y;
/* The blitter interprets the 16-bit destination x/y as a signed 16-bit
* value. The values we're working with are unsigned, so make sure we
* don't overflow.
*/
if (dst_x >= 32768 || dst_y >= 32768) {
perf_debug("Falling back due to >=32k dst offset (%d, %d)\n",
dst_x, dst_y);
return false;
}
if (!intelEmitCopyBlit(brw,
src_mt->cpp,
src_pitch,
src_mt->bo, src_mt->offset,
src_mt->tiling,
dst_mt->pitch,
dst_mt->bo, dst_mt->offset,
dst_mt->tiling,
src_x, src_y,
dst_x, dst_y,
width, height,
logicop)) {
return false;
}
if (src_mt->format == MESA_FORMAT_B8G8R8X8_UNORM &&
dst_mt->format == MESA_FORMAT_B8G8R8A8_UNORM) {
intel_miptree_set_alpha_to_one(brw, dst_mt,
dst_x, dst_y,
width, height);
}
return true;
}
/**
* CopyPixels with the blitter. Don't support zooming, pixel transfer, etc.
*/
static bool
do_blit_copypixels(struct gl_context * ctx,
GLint srcx, GLint srcy,
GLsizei width, GLsizei height,
GLint dstx, GLint dsty, GLenum type)
{
struct intel_context *intel = intel_context(ctx);
struct gl_framebuffer *fb = ctx->DrawBuffer;
struct gl_framebuffer *read_fb = ctx->ReadBuffer;
GLint orig_dstx;
GLint orig_dsty;
GLint orig_srcx;
GLint orig_srcy;
bool flip = false;
struct intel_renderbuffer *draw_irb = NULL;
struct intel_renderbuffer *read_irb = NULL;
gl_format read_format, draw_format;
/* Update draw buffer bounds */
_mesa_update_state(ctx);
switch (type) {
case GL_COLOR:
if (fb->_NumColorDrawBuffers != 1) {
perf_debug("glCopyPixels() fallback: MRT\n");
return false;
}
draw_irb = intel_renderbuffer(fb->_ColorDrawBuffers[0]);
read_irb = intel_renderbuffer(read_fb->_ColorReadBuffer);
break;
case GL_DEPTH_STENCIL_EXT:
draw_irb = intel_renderbuffer(fb->Attachment[BUFFER_DEPTH].Renderbuffer);
read_irb =
intel_renderbuffer(read_fb->Attachment[BUFFER_DEPTH].Renderbuffer);
break;
case GL_DEPTH:
perf_debug("glCopyPixels() fallback: GL_DEPTH\n");
return false;
case GL_STENCIL:
perf_debug("glCopyPixels() fallback: GL_STENCIL\n");
return false;
default:
perf_debug("glCopyPixels(): Unknown type\n");
return false;
}
if (!draw_irb) {
perf_debug("glCopyPixels() fallback: missing draw buffer\n");
return false;
}
if (!read_irb) {
perf_debug("glCopyPixels() fallback: missing read buffer\n");
return false;
}
read_format = intel_rb_format(read_irb);
draw_format = intel_rb_format(draw_irb);
if (draw_format != read_format &&
!(draw_format == MESA_FORMAT_XRGB8888 &&
read_format == MESA_FORMAT_ARGB8888)) {
perf_debug("glCopyPixels() fallback: mismatched formats (%s -> %s\n",
_mesa_get_format_name(read_format),
_mesa_get_format_name(draw_format));
return false;
}
/* Copypixels can be more than a straight copy. Ensure all the
* extra operations are disabled:
*/
if (!intel_check_copypixel_blit_fragment_ops(ctx) ||
ctx->Pixel.ZoomX != 1.0F || ctx->Pixel.ZoomY != 1.0F)
return false;
intel_prepare_render(intel);
intel_flush(&intel->ctx);
/* Clip to destination buffer. */
orig_dstx = dstx;
orig_dsty = dsty;
if (!_mesa_clip_to_region(fb->_Xmin, fb->_Ymin,
fb->_Xmax, fb->_Ymax,
&dstx, &dsty, &width, &height))
goto out;
/* Adjust src coords for our post-clipped destination origin */
srcx += dstx - orig_dstx;
srcy += dsty - orig_dsty;
/* Clip to source buffer. */
orig_srcx = srcx;
orig_srcy = srcy;
if (!_mesa_clip_to_region(0, 0,
read_fb->Width, read_fb->Height,
&srcx, &srcy, &width, &height))
goto out;
/* Adjust dst coords for our post-clipped source origin */
dstx += srcx - orig_srcx;
dsty += srcy - orig_srcy;
/* Flip dest Y if it's a window system framebuffer. */
if (_mesa_is_winsys_fbo(fb)) {
/* copypixels to a window system framebuffer */
dsty = fb->Height - dsty - height;
flip = !flip;
}
/* Flip source Y if it's a window system framebuffer. */
if (_mesa_is_winsys_fbo(read_fb)) {
srcy = read_fb->Height - srcy - height;
flip = !flip;
}
srcx += read_irb->draw_x;
srcy += read_irb->draw_y;
dstx += draw_irb->draw_x;
dsty += draw_irb->draw_y;
uint32_t src_pitch = read_irb->mt->region->pitch;
if (flip)
src_pitch = -src_pitch;
if (!intelEmitCopyBlit(intel,
draw_irb->mt->cpp,
src_pitch, read_irb->mt->region->bo,
0, read_irb->mt->region->tiling,
draw_irb->mt->region->pitch, draw_irb->mt->region->bo,
0, draw_irb->mt->region->tiling,
srcx, srcy,
dstx, dsty,
width, height,
ctx->Color.ColorLogicOpEnabled ?
ctx->Color.LogicOp : GL_COPY)) {
DBG("%s: blit failure\n", __FUNCTION__);
return false;
}
out:
intel_check_front_buffer_rendering(intel);
DBG("%s: success\n", __FUNCTION__);
return true;
}
/* Pros:
* - no waiting for idle before updating framebuffer.
*
* Cons:
* - if upload is by memcpy, this may actually be slower than fallback path.
* - uploads the whole image even if destination is clipped
*
* Need to benchmark.
*
* Given the questions about performance, implement for pbo's only.
* This path is definitely a win if the pbo is already in agp. If it
* turns out otherwise, we can add the code necessary to upload client
* data to agp space before performing the blit. (Though it may turn
* out to be better/simpler just to use the texture engine).
*/
static GLboolean
do_blit_drawpixels(GLcontext * ctx,
GLint x, GLint y,
GLsizei width, GLsizei height,
GLenum format, GLenum type,
const struct gl_pixelstore_attrib *unpack,
const GLvoid * pixels)
{
struct intel_context *intel = intel_context(ctx);
struct intel_region *dest = intel_drawbuf_region(intel);
struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
GLuint src_offset;
GLuint rowLength;
struct _DriFenceObject *fence = NULL;
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s\n", __FUNCTION__);
if (!dest) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - no dest\n", __FUNCTION__);
return GL_FALSE;
}
if (src) {
/* This validation should be done by core mesa:
*/
if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
format, type, pixels)) {
_mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
return GL_TRUE;
}
}
else {
/* PBO only for now:
*/
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - not PBO\n", __FUNCTION__);
return GL_FALSE;
}
if (!intel_check_blit_format(dest, format, type)) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad format for blit\n", __FUNCTION__);
return GL_FALSE;
}
if (!intel_check_blit_fragment_ops(ctx)) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad GL fragment state for blitter\n",
__FUNCTION__);
return GL_FALSE;
}
if (ctx->Pixel.ZoomX != 1.0F) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad PixelZoomX for blit\n", __FUNCTION__);
return GL_FALSE;
}
if (unpack->RowLength > 0)
rowLength = unpack->RowLength;
else
rowLength = width;
if (ctx->Pixel.ZoomY == -1.0F) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
return GL_FALSE; /* later */
y -= height;
}
else if (ctx->Pixel.ZoomY == 1.0F) {
rowLength = -rowLength;
}
else {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
return GL_FALSE;
}
src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height,
format, type, 0, 0, 0);
intelFlush(&intel->ctx);
LOCK_HARDWARE(intel);
if (intel->driDrawable->numClipRects) {
__DRIdrawablePrivate *dPriv = intel->driDrawable;
int nbox = dPriv->numClipRects;
drm_clip_rect_t *box = dPriv->pClipRects;
drm_clip_rect_t rect;
drm_clip_rect_t dest_rect;
struct _DriBufferObject *src_buffer =
intel_bufferobj_buffer(intel, src, INTEL_READ);
int i;
dest_rect.x1 = dPriv->x + x;
dest_rect.y1 = dPriv->y + dPriv->h - (y + height);
dest_rect.x2 = dest_rect.x1 + width;
dest_rect.y2 = dest_rect.y1 + height;
for (i = 0; i < nbox; i++) {
if (!intel_intersect_cliprects(&rect, &dest_rect, &box[i]))
continue;
intelEmitCopyBlit(intel,
dest->cpp,
rowLength,
src_buffer, src_offset,
dest->pitch,
dest->buffer, 0,
rect.x1 - dest_rect.x1,
rect.y2 - dest_rect.y2,
rect.x1,
rect.y1, rect.x2 - rect.x1, rect.y2 - rect.y1,
ctx->Color.ColorLogicOpEnabled ?
ctx->Color.LogicOp : GL_COPY);
}
fence = intel_batchbuffer_flush(intel->batch);
driFenceReference(fence);
}
UNLOCK_HARDWARE(intel);
if (fence) {
driFenceFinish(fence, DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW, GL_FALSE);
driFenceUnReference(fence);
}
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - DONE\n", __FUNCTION__);
return GL_TRUE;
}
