/**
* Normally, we'll only _write_ texel data to a texture when we map it.
* But if the user is providing depth or stencil values and the texture
* image is a combined depth/stencil format, we'll actually read from
* the texture buffer too (in order to insert the depth or stencil values.
* \param userFormat the user-provided image format
* \param texFormat the destination texture format
*/
static GLbitfield
get_read_write_mode(GLenum userFormat, mesa_format texFormat)
{
if ((userFormat == GL_STENCIL_INDEX || userFormat == GL_DEPTH_COMPONENT)
&& _mesa_get_format_base_format(texFormat) == GL_DEPTH_STENCIL)
return GL_MAP_READ_BIT | GL_MAP_WRITE_BIT;
else
return GL_MAP_WRITE_BIT | GL_MAP_INVALIDATE_RANGE_BIT;
}
/**
* Return true if the conversion L=R+G+B is needed.
*/
static GLboolean
need_rgb_to_luminance_conversion(mesa_format texFormat, GLenum format)
{
GLenum baseTexFormat = _mesa_get_format_base_format(texFormat);
return (baseTexFormat == GL_RG ||
baseTexFormat == GL_RGB ||
baseTexFormat == GL_RGBA) &&
(format == GL_LUMINANCE || format == GL_LUMINANCE_ALPHA);
}
GLboolean
_mesa_texstore_red_rgtc1(TEXSTORE_PARAMS)
{
GLubyte *dst;
const GLint texWidth = dstRowStride * 4 / 8; /* a bit of a hack */
const GLubyte *tempImage = NULL;
int i, j;
int numxpixels, numypixels;
const GLubyte *srcaddr;
GLubyte srcpixels[4][4];
GLubyte *blkaddr;
GLint dstRowDiff;
ASSERT(dstFormat == MESA_FORMAT_RED_RGTC1 ||
dstFormat == MESA_FORMAT_L_LATC1);
ASSERT(dstXoffset % 4 == 0);
ASSERT(dstYoffset % 4 == 0);
ASSERT(dstZoffset % 4 == 0);
(void) dstZoffset;
tempImage = _mesa_make_temp_ubyte_image(ctx, dims,
baseInternalFormat,
_mesa_get_format_base_format(dstFormat),
srcWidth, srcHeight, srcDepth,
srcFormat, srcType, srcAddr,
srcPacking);
if (!tempImage)
return GL_FALSE; /* out of memory */
dst = _mesa_compressed_image_address(dstXoffset, dstYoffset, 0,
dstFormat,
texWidth, dstSlices[0]);
blkaddr = dst;
dstRowDiff = dstRowStride >= (srcWidth * 2) ? dstRowStride - (((srcWidth + 3) & ~3) * 2) : 0;
for (j = 0; j < srcHeight; j+=4) {
if (srcHeight > j + 3) numypixels = 4;
else numypixels = srcHeight - j;
srcaddr = tempImage + j * srcWidth;
for (i = 0; i < srcWidth; i += 4) {
if (srcWidth > i + 3) numxpixels = 4;
else numxpixels = srcWidth - i;
extractsrc_u(srcpixels, srcaddr, srcWidth, numxpixels, numypixels, 1);
unsigned_encode_rgtc_ubyte(blkaddr, srcpixels, numxpixels, numypixels);
srcaddr += numxpixels;
blkaddr += 8;
}
blkaddr += dstRowDiff;
}
if (tempImage)
free((void *) tempImage);
return GL_TRUE;
}
GLboolean
_mesa_texstore_signed_rg_rgtc2(TEXSTORE_PARAMS)
{
GLbyte *dst;
const GLfloat *tempImage = NULL;
int i, j;
int numxpixels, numypixels;
const GLfloat *srcaddr;
GLbyte srcpixels[4][4];
GLbyte *blkaddr;
GLint dstRowDiff;
ASSERT(dstFormat == MESA_FORMAT_RG_RGTC2_SNORM ||
dstFormat == MESA_FORMAT_LA_LATC2_SNORM);
tempImage = _mesa_make_temp_float_image(ctx, dims,
baseInternalFormat,
_mesa_get_format_base_format(dstFormat),
srcWidth, srcHeight, srcDepth,
srcFormat, srcType, srcAddr,
srcPacking, 0x0);
if (!tempImage)
return GL_FALSE; /* out of memory */
dst = (GLbyte *) dstSlices[0];
blkaddr = dst;
dstRowDiff = dstRowStride >= (srcWidth * 4) ? dstRowStride - (((srcWidth + 3) & ~3) * 4) : 0;
for (j = 0; j < srcHeight; j += 4) {
if (srcHeight > j + 3) numypixels = 4;
else numypixels = srcHeight - j;
srcaddr = tempImage + j * srcWidth * 2;
for (i = 0; i < srcWidth; i += 4) {
if (srcWidth > i + 3) numxpixels = 4;
else numxpixels = srcWidth - i;
extractsrc_s(srcpixels, srcaddr, srcWidth, numxpixels, numypixels, 2);
signed_encode_rgtc_ubyte(blkaddr, srcpixels, numxpixels, numypixels);
blkaddr += 8;
extractsrc_s(srcpixels, srcaddr + 1, srcWidth, numxpixels, numypixels, 2);
signed_encode_rgtc_ubyte(blkaddr, srcpixels, numxpixels, numypixels);
blkaddr += 8;
srcaddr += numxpixels * 2;
}
blkaddr += dstRowDiff;
}
free((void *) tempImage);
return GL_TRUE;
}
/**
* Store user's image in rgb_dxt1 format.
*/
GLboolean
_mesa_texstore_rgb_dxt1(TEXSTORE_PARAMS)
{
const GLchan *pixels;
GLubyte *dst;
const GLint texWidth = dstRowStride * 4 / 8; /* a bit of a hack */
const GLchan *tempImage = NULL;
ASSERT(dstFormat == MESA_FORMAT_RGB_DXT1 ||
dstFormat == MESA_FORMAT_SRGB_DXT1);
ASSERT(dstXoffset % 4 == 0);
ASSERT(dstYoffset % 4 == 0);
ASSERT(dstZoffset % 4 == 0);
(void) dstZoffset;
(void) dstImageOffsets;
if (srcFormat != GL_RGB ||
srcType != CHAN_TYPE ||
ctx->_ImageTransferState ||
srcPacking->SwapBytes) {
/* convert image to RGB/GLchan */
tempImage = _mesa_make_temp_chan_image(ctx, dims,
baseInternalFormat,
_mesa_get_format_base_format(dstFormat),
srcWidth, srcHeight, srcDepth,
srcFormat, srcType, srcAddr,
srcPacking);
if (!tempImage)
return GL_FALSE; /* out of memory */
pixels = tempImage;
srcFormat = GL_RGB;
}
else {
pixels = (const GLchan *) srcAddr;
}
dst = _mesa_compressed_image_address(dstXoffset, dstYoffset, 0,
dstFormat,
texWidth, (GLubyte *) dstAddr);
if (ext_tx_compress_dxtn) {
(*ext_tx_compress_dxtn)(3, srcWidth, srcHeight, pixels,
GL_COMPRESSED_RGB_S3TC_DXT1_EXT,
dst, dstRowStride);
}
else {
_mesa_warning(ctx, "external dxt library not available: texstore_rgb_dxt1");
}
if (tempImage)
free((void *) tempImage);
return GL_TRUE;
}
static unsigned int
intel_horizontal_texture_alignment_unit(struct brw_context *brw,
gl_format format)
{
/**
* From the "Alignment Unit Size" section of various specs, namely:
* - Gen3 Spec: "Memory Data Formats" Volume, Section 1.20.1.4
* - i965 and G45 PRMs: Volume 1, Section 6.17.3.4.
* - Ironlake and Sandybridge PRMs: Volume 1, Part 1, Section 7.18.3.4
* - BSpec (for Ivybridge and slight variations in separate stencil)
*
* +----------------------------------------------------------------------+
* | | alignment unit width ("i") |
* | Surface Property |-----------------------------|
* | | 915 | 965 | ILK | SNB | IVB |
* +----------------------------------------------------------------------+
* | YUV 4:2:2 format | 8 | 4 | 4 | 4 | 4 |
* | BC1-5 compressed format (DXTn/S3TC) | 4 | 4 | 4 | 4 | 4 |
* | FXT1 compressed format | 8 | 8 | 8 | 8 | 8 |
* | Depth Buffer (16-bit) | 4 | 4 | 4 | 4 | 8 |
* | Depth Buffer (other) | 4 | 4 | 4 | 4 | 4 |
* | Separate Stencil Buffer | N/A | N/A | 8 | 8 | 8 |
* | All Others | 4 | 4 | 4 | 4 | 4 |
* +----------------------------------------------------------------------+
*
* On IVB+, non-special cases can be overridden by setting the SURFACE_STATE
* "Surface Horizontal Alignment" field to HALIGN_4 or HALIGN_8.
*/
if (_mesa_is_format_compressed(format)) {
/* The hardware alignment requirements for compressed textures
* happen to match the block boundaries.
*/
unsigned int i, j;
_mesa_get_format_block_size(format, &i, &j);
return i;
}
if (format == MESA_FORMAT_S8)
return 8;
/* The depth alignment requirements in the table above are for rendering to
* depth miplevels using the LOD control fields. We don't use LOD control
* fields, and instead use page offsets plus intra-tile x/y offsets, which
* require that the low 3 bits are zero. To reduce the number of x/y
* offset workaround blits we do, align the X to 8, which depth texturing
* can handle (sadly, it can't handle 8 in the Y direction).
*/
if (brw->gen >= 7 &&
_mesa_get_format_base_format(format) == GL_DEPTH_COMPONENT)
return 8;
return 4;
}
/**
* Return transfer op flags for this ReadPixels operation.
*/
GLbitfield
_mesa_get_readpixels_transfer_ops(const struct gl_context *ctx,
mesa_format texFormat,
GLenum format, GLenum type,
GLboolean uses_blit)
{
GLbitfield transferOps = ctx->_ImageTransferState;
GLenum srcBaseFormat = _mesa_get_format_base_format(texFormat);
GLenum dstBaseFormat = _mesa_unpack_format_to_base_format(format);
if (format == GL_DEPTH_COMPONENT ||
format == GL_DEPTH_STENCIL ||
format == GL_STENCIL_INDEX) {
return 0;
}
/* Pixel transfer ops (scale, bias, table lookup) do not apply
* to integer formats.
*/
if (_mesa_is_enum_format_integer(format)) {
return 0;
}
if (uses_blit) {
/* For blit-based ReadPixels packing, the clamping is done automatically
* unless the type is float. */
if (_mesa_get_clamp_read_color(ctx, ctx->ReadBuffer) &&
(type == GL_FLOAT || type == GL_HALF_FLOAT)) {
transferOps |= IMAGE_CLAMP_BIT;
}
}
else {
/* For CPU-based ReadPixels packing, the clamping must always be done
* for non-float types, */
if (_mesa_get_clamp_read_color(ctx, ctx->ReadBuffer) ||
(type != GL_FLOAT && type != GL_HALF_FLOAT)) {
transferOps |= IMAGE_CLAMP_BIT;
}
}
/* If the format is unsigned normalized, we can ignore clamping
* because the values are already in the range [0,1] so it won't
* have any effect anyway.
*/
if (_mesa_get_format_datatype(texFormat) == GL_UNSIGNED_NORMALIZED &&
!_mesa_need_rgb_to_luminance_conversion(srcBaseFormat, dstBaseFormat)) {
transferOps &= ~IMAGE_CLAMP_BIT;
}
return transferOps;
}
static void
intel_image_target_texture_2d(struct gl_context *ctx, GLenum target,
struct gl_texture_object *texObj,
struct gl_texture_image *texImage,
GLeglImageOES image_handle)
{
struct brw_context *brw = brw_context(ctx);
struct intel_mipmap_tree *mt;
__DRIscreen *dri_screen = brw->screen->driScrnPriv;
__DRIimage *image;
image = dri_screen->dri2.image->lookupEGLImage(dri_screen, image_handle,
dri_screen->loaderPrivate);
if (image == NULL)
return;
/* We support external textures only for EGLImages created with
* EGL_EXT_image_dma_buf_import. We may lift that restriction in the future.
*/
if (target == GL_TEXTURE_EXTERNAL_OES && !image->dma_buf_imported) {
_mesa_error(ctx, GL_INVALID_OPERATION,
"glEGLImageTargetTexture2DOES(external target is enabled only "
"for images created with EGL_EXT_image_dma_buf_import");
return;
}
/* Disallow depth/stencil textures: we don't have a way to pass the
* separate stencil miptree of a GL_DEPTH_STENCIL texture through.
*/
if (image->has_depthstencil) {
_mesa_error(ctx, GL_INVALID_OPERATION, __func__);
return;
}
if (image->planar_format && image->planar_format->nplanes > 0)
mt = create_mt_for_planar_dri_image(brw, target, image);
else
mt = create_mt_for_dri_image(brw, target, image);
if (mt == NULL)
return;
struct intel_texture_object *intel_texobj = intel_texture_object(texObj);
intel_texobj->planar_format = image->planar_format;
const GLenum internal_format =
image->internal_format != 0 ?
image->internal_format : _mesa_get_format_base_format(mt->format);
intel_set_texture_image_mt(brw, texImage, internal_format, mt);
intel_miptree_release(&mt);
}
/**
* Store user's image in rgba_dxt1 format.
*/
GLboolean
_mesa_texstore_rgba_dxt1(TEXSTORE_PARAMS)
{
const GLubyte *pixels;
GLubyte *dst;
const GLubyte *tempImage = NULL;
ASSERT(dstFormat == MESA_FORMAT_RGBA_DXT1 ||
dstFormat == MESA_FORMAT_SRGBA_DXT1);
if (srcFormat != GL_RGBA ||
srcType != GL_UNSIGNED_BYTE ||
ctx->_ImageTransferState ||
srcPacking->RowLength != srcWidth ||
srcPacking->SwapBytes) {
/* convert image to RGBA/GLubyte */
tempImage = _mesa_make_temp_ubyte_image(ctx, dims,
baseInternalFormat,
_mesa_get_format_base_format(dstFormat),
srcWidth, srcHeight, srcDepth,
srcFormat, srcType, srcAddr,
srcPacking);
if (!tempImage)
return GL_FALSE; /* out of memory */
pixels = tempImage;
srcFormat = GL_RGBA;
}
else {
pixels = _mesa_image_address2d(srcPacking, srcAddr, srcWidth, srcHeight,
srcFormat, srcType, 0, 0);
}
dst = dstSlices[0];
if (ext_tx_compress_dxtn) {
(*ext_tx_compress_dxtn)(4, srcWidth, srcHeight, pixels,
GL_COMPRESSED_RGBA_S3TC_DXT1_EXT,
dst, dstRowStride);
}
else {
_mesa_warning(ctx, "external dxt library not available: texstore_rgba_dxt1");
}
free((void*) tempImage);
return GL_TRUE;
}
static void
intel_image_target_renderbuffer_storage(struct gl_context *ctx,
struct gl_renderbuffer *rb,
void *image_handle)
{
struct intel_context *intel = intel_context(ctx);
struct intel_renderbuffer *irb;
__DRIscreen *screen;
__DRIimage *image;
screen = intel->intelScreen->driScrnPriv;
image = screen->dri2.image->lookupEGLImage(screen, image_handle,
screen->loaderPrivate);
if (image == NULL)
return;
/* __DRIimage is opaque to the core so it has to be checked here */
switch (image->format) {
case MESA_FORMAT_R8G8B8A8_UNORM:
_mesa_error(&intel->ctx, GL_INVALID_OPERATION,
"glEGLImageTargetRenderbufferStorage(unsupported image format");
return;
break;
default:
break;
}
irb = intel_renderbuffer(rb);
intel_miptree_release(&irb->mt);
irb->mt = intel_miptree_create_for_bo(intel,
image->region->bo,
image->format,
image->offset,
image->region->width,
image->region->height,
image->region->pitch,
image->region->tiling);
if (!irb->mt)
return;
rb->InternalFormat = image->internal_format;
rb->Width = image->region->width;
rb->Height = image->region->height;
rb->Format = image->format;
rb->_BaseFormat = _mesa_get_format_base_format(image->format);
rb->NeedsFinishRenderTexture = true;
}
static __DRIimage *
intel_allocate_image(int dri_format, void *loaderPrivate)
{
__DRIimage *image;
image = calloc(1, sizeof *image);
if (image == NULL)
return NULL;
image->dri_format = dri_format;
image->offset = 0;
switch (dri_format) {
case __DRI_IMAGE_FORMAT_RGB565:
image->format = MESA_FORMAT_RGB565;
break;
case __DRI_IMAGE_FORMAT_XRGB8888:
image->format = MESA_FORMAT_XRGB8888;
break;
case __DRI_IMAGE_FORMAT_ARGB8888:
image->format = MESA_FORMAT_ARGB8888;
break;
case __DRI_IMAGE_FORMAT_ABGR8888:
image->format = MESA_FORMAT_RGBA8888_REV;
break;
case __DRI_IMAGE_FORMAT_XBGR8888:
image->format = MESA_FORMAT_RGBX8888_REV;
break;
case __DRI_IMAGE_FORMAT_R8:
image->format = MESA_FORMAT_R8;
break;
case __DRI_IMAGE_FORMAT_GR88:
image->format = MESA_FORMAT_GR88;
break;
case __DRI_IMAGE_FORMAT_NONE:
image->format = MESA_FORMAT_NONE;
break;
default:
free(image);
return NULL;
}
image->internal_format = _mesa_get_format_base_format(image->format);
image->data = loaderPrivate;
return image;
}
/**
* Creates a new named renderbuffer that wraps the first slice
* of an existing miptree.
*
* Clobbers the current renderbuffer binding (ctx->CurrentRenderbuffer).
*/
GLuint
brw_get_rb_for_slice(struct brw_context *brw,
struct intel_mipmap_tree *mt,
unsigned level, unsigned layer, bool flat)
{
struct gl_context *ctx = &brw->ctx;
GLuint rbo;
struct gl_renderbuffer *rb;
struct intel_renderbuffer *irb;
/* This turns the GenRenderbuffers name into an actual struct
* intel_renderbuffer.
*/
_mesa_GenRenderbuffers(1, &rbo);
_mesa_BindRenderbuffer(GL_RENDERBUFFER, rbo);
rb = ctx->CurrentRenderbuffer;
irb = intel_renderbuffer(rb);
rb->Format = mt->format;
rb->_BaseFormat = _mesa_get_format_base_format(mt->format);
/* Program takes care of msaa and mip-level access manually for stencil.
* The surface is also treated as Y-tiled instead of as W-tiled calling for
* twice the width and half the height in dimensions.
*/
if (flat) {
const unsigned halign_stencil = 8;
rb->NumSamples = 0;
rb->Width = ALIGN(mt->total_width, halign_stencil) * 2;
rb->Height = (mt->total_height / mt->physical_depth0) / 2;
irb->mt_level = 0;
} else {
rb->NumSamples = mt->num_samples;
rb->Width = mt->logical_width0;
rb->Height = mt->logical_height0;
irb->mt_level = level;
}
irb->mt_layer = layer;
intel_miptree_reference(&irb->mt, mt);
return rbo;
}
static unsigned int
intel_vertical_texture_alignment_unit(struct brw_context *brw,
gl_format format, bool multisampled)
{
/**
* From the "Alignment Unit Size" section of various specs, namely:
* - Gen3 Spec: "Memory Data Formats" Volume, Section 1.20.1.4
* - i965 and G45 PRMs: Volume 1, Section 6.17.3.4.
* - Ironlake and Sandybridge PRMs: Volume 1, Part 1, Section 7.18.3.4
* - BSpec (for Ivybridge and slight variations in separate stencil)
*
* +----------------------------------------------------------------------+
* | | alignment unit height ("j") |
* | Surface Property |-----------------------------|
* | | 915 | 965 | ILK | SNB | IVB |
* +----------------------------------------------------------------------+
* | BC1-5 compressed format (DXTn/S3TC) | 4 | 4 | 4 | 4 | 4 |
* | FXT1 compressed format | 4 | 4 | 4 | 4 | 4 |
* | Depth Buffer | 2 | 2 | 2 | 4 | 4 |
* | Separate Stencil Buffer | N/A | N/A | N/A | 4 | 8 |
* | Multisampled (4x or 8x) render target | N/A | N/A | N/A | 4 | 4 |
* | All Others | 2 | 2 | 2 | 2 | 2 |
* +----------------------------------------------------------------------+
*
* On SNB+, non-special cases can be overridden by setting the SURFACE_STATE
* "Surface Vertical Alignment" field to VALIGN_2 or VALIGN_4.
*/
if (_mesa_is_format_compressed(format))
return 4;
if (format == MESA_FORMAT_S8)
return brw->gen >= 7 ? 8 : 4;
if (multisampled)
return 4;
GLenum base_format = _mesa_get_format_base_format(format);
if (brw->gen >= 6 &&
(base_format == GL_DEPTH_COMPONENT ||
base_format == GL_DEPTH_STENCIL)) {
return 4;
}
return 2;
}
unsigned
brw_miptree_get_vertical_slice_pitch(const struct brw_context *brw,
const struct intel_mipmap_tree *mt,
unsigned level)
{
if (brw->gen >= 9) {
/* ALL_SLICES_AT_EACH_LOD isn't supported on Gen8+ but this code will
* effectively end up with a packed qpitch anyway whenever
* mt->first_level == mt->last_level.
*/
assert(mt->array_layout != ALL_SLICES_AT_EACH_LOD);
/* On Gen9 we can pick whatever qpitch we like as long as it's aligned
* to the vertical alignment so we don't need to add any extra rows.
*/
unsigned qpitch = mt->total_height;
/* If the surface might be used as a stencil buffer or HiZ buffer then
* it needs to be a multiple of 8.
*/
const GLenum base_format = _mesa_get_format_base_format(mt->format);
if (_mesa_is_depth_or_stencil_format(base_format))
qpitch = ALIGN(qpitch, 8);
/* 3D textures need to be aligned to the tile height. At this point we
* don't know which tiling will be used so let's just align it to 32
*/
if (mt->target == GL_TEXTURE_3D)
qpitch = ALIGN(qpitch, 32);
return qpitch;
} else if (mt->target == GL_TEXTURE_3D ||
(brw->gen == 4 && mt->target == GL_TEXTURE_CUBE_MAP) ||
mt->array_layout == ALL_SLICES_AT_EACH_LOD) {
return ALIGN_NPOT(minify(mt->physical_height0, level), mt->align_h);
} else {
const unsigned h0 = ALIGN_NPOT(mt->physical_height0, mt->align_h);
const unsigned h1 = ALIGN_NPOT(minify(mt->physical_height0, 1), mt->align_h);
return h0 + h1 + (brw->gen >= 7 ? 12 : 11) * mt->align_h;
}
}
/**
* Update the renderbuffer wrapper for rendering to a texture.
* For example, update the width, height of the RB based on the texture size,
* update the internal format info, etc.
*/
static void
update_wrapper(struct gl_context *ctx, struct gl_renderbuffer_attachment *att)
{
struct gl_renderbuffer *rb = att->Renderbuffer;
struct swrast_renderbuffer *srb = swrast_renderbuffer(rb);
struct swrast_texture_image *swImage;
gl_format format;
GLuint zOffset;
(void) ctx;
swImage = swrast_texture_image(_mesa_get_attachment_teximage(att));
assert(swImage);
format = swImage->Base.TexFormat;
if (att->Texture->Target == GL_TEXTURE_1D_ARRAY_EXT) {
zOffset = 0;
}
else {
zOffset = att->Zoffset;
}
rb->Width = swImage->Base.Width;
rb->Height = swImage->Base.Height;
rb->InternalFormat = swImage->Base.InternalFormat;
rb->_BaseFormat = _mesa_get_format_base_format(format);
/* Want to store linear values, not sRGB */
rb->Format = _mesa_get_srgb_format_linear(format);
/* Set the gl_renderbuffer::Buffer field so that mapping the buffer
* succeeds.
*/
if (att->Texture->Target == GL_TEXTURE_3D ||
att->Texture->Target == GL_TEXTURE_2D_ARRAY_EXT) {
srb->Buffer = swImage->Buffer +
swImage->ImageOffsets[zOffset] * _mesa_get_format_bytes(format);
}
else {
srb->Buffer = swImage->Buffer;
}
}
static bool
use_linear_1d_layout(struct brw_context *brw,
struct intel_mipmap_tree *mt)
{
/* On Gen9+ the mipmap levels of a 1D surface are all laid out in a
* horizontal line. This isn't done for depth/stencil buffers however
* because those will be using a tiled layout
*/
if (brw->gen >= 9 &&
(mt->target == GL_TEXTURE_1D ||
mt->target == GL_TEXTURE_1D_ARRAY)) {
GLenum base_format = _mesa_get_format_base_format(mt->format);
if (base_format != GL_DEPTH_COMPONENT &&
base_format != GL_DEPTH_STENCIL &&
base_format != GL_STENCIL_INDEX)
return true;
}
return false;
}
/**
* Adaptor for fetching a float texel from a GLchan-valued texture.
*/
static void
fetch_texel_chan_to_float(const struct gl_texture_image *texImage,
GLint i, GLint j, GLint k, GLfloat *texelOut)
{
GLchan temp[4];
GLenum baseFormat = _mesa_get_format_base_format(texImage->TexFormat);
ASSERT(texImage->FetchTexelc);
texImage->FetchTexelc(texImage, i, j, k, temp);
if (baseFormat == GL_DEPTH_COMPONENT ||
baseFormat == GL_DEPTH_STENCIL_EXT) {
/* just one channel */
texelOut[0] = CHAN_TO_FLOAT(temp[0]);
}
else {
/* four channels */
texelOut[0] = CHAN_TO_FLOAT(temp[0]);
texelOut[1] = CHAN_TO_FLOAT(temp[1]);
texelOut[2] = CHAN_TO_FLOAT(temp[2]);
texelOut[3] = CHAN_TO_FLOAT(temp[3]);
}
}
GLboolean
_mesa_texstore_can_use_memcpy(struct gl_context *ctx,
GLenum baseInternalFormat, mesa_format dstFormat,
GLenum srcFormat, GLenum srcType,
const struct gl_pixelstore_attrib *srcPacking)
{
if (_mesa_texstore_needs_transfer_ops(ctx, baseInternalFormat, dstFormat)) {
return GL_FALSE;
}
/* The base internal format and the base Mesa format must match. */
if (baseInternalFormat != _mesa_get_format_base_format(dstFormat)) {
return GL_FALSE;
}
/* The Mesa format must match the input format and type. */
if (!_mesa_format_matches_format_and_type(dstFormat, srcFormat, srcType,
srcPacking->SwapBytes)) {
return GL_FALSE;
}
/* Depth texture data needs clamping in following cases:
* - Floating point dstFormat with signed srcType: clamp to [0.0, 1.0].
* - Fixed point dstFormat with signed srcType: clamp to [0, 2^n -1].
*
* All the cases except one (float dstFormat with float srcType) are ruled
* out by _mesa_format_matches_format_and_type() check above. Handle the
* remaining case here.
*/
if ((baseInternalFormat == GL_DEPTH_COMPONENT ||
baseInternalFormat == GL_DEPTH_STENCIL) &&
(srcType == GL_FLOAT ||
srcType == GL_FLOAT_32_UNSIGNED_INT_24_8_REV)) {
return GL_FALSE;
}
return GL_TRUE;
}
static __DRIimage *
intel_allocate_image(int dri_format, void *loaderPrivate)
{
__DRIimage *image;
image = calloc(1, sizeof *image);
if (image == NULL)
return NULL;
image->dri_format = dri_format;
image->offset = 0;
image->format = driImageFormatToGLFormat(dri_format);
if (dri_format != __DRI_IMAGE_FORMAT_NONE &&
image->format == MESA_FORMAT_NONE) {
free(image);
return NULL;
}
image->internal_format = _mesa_get_format_base_format(image->format);
image->data = loaderPrivate;
return image;
}
void
brw_miptree_layout(struct brw_context *brw, struct intel_mipmap_tree *mt)
{
bool multisampled = mt->num_samples > 1;
bool gen6_hiz_or_stencil = false;
if (brw->gen == 6 && mt->array_layout == ALL_SLICES_AT_EACH_LOD) {
const GLenum base_format = _mesa_get_format_base_format(mt->format);
gen6_hiz_or_stencil = _mesa_is_depth_or_stencil_format(base_format);
}
if (gen6_hiz_or_stencil) {
/* On gen6, we use ALL_SLICES_AT_EACH_LOD for stencil/hiz because the
* hardware doesn't support multiple mip levels on stencil/hiz.
*
* PRM Vol 2, Part 1, 7.5.3 Hierarchical Depth Buffer:
* "The hierarchical depth buffer does not support the LOD field"
*
* PRM Vol 2, Part 1, 7.5.4.1 Separate Stencil Buffer:
* "The stencil depth buffer does not support the LOD field"
*/
if (mt->format == MESA_FORMAT_S_UINT8) {
/* Stencil uses W tiling, so we force W tiling alignment for the
* ALL_SLICES_AT_EACH_LOD miptree layout.
*/
mt->align_w = 64;
mt->align_h = 64;
} else {
/* Depth uses Y tiling, so we force need Y tiling alignment for the
* ALL_SLICES_AT_EACH_LOD miptree layout.
*/
mt->align_w = 128 / mt->cpp;
mt->align_h = 32;
}
} else {
mt->align_w = intel_horizontal_texture_alignment_unit(brw, mt);
mt->align_h =
intel_vertical_texture_alignment_unit(brw, mt->format, multisampled);
}
switch (mt->target) {
case GL_TEXTURE_CUBE_MAP:
if (brw->gen == 4) {
/* Gen4 stores cube maps as 3D textures. */
assert(mt->physical_depth0 == 6);
brw_miptree_layout_texture_3d(brw, mt);
} else {
/* All other hardware stores cube maps as 2D arrays. */
brw_miptree_layout_texture_array(brw, mt);
}
break;
case GL_TEXTURE_3D:
if (brw->gen >= 9)
brw_miptree_layout_texture_array(brw, mt);
else
brw_miptree_layout_texture_3d(brw, mt);
break;
case GL_TEXTURE_1D_ARRAY:
case GL_TEXTURE_2D_ARRAY:
case GL_TEXTURE_2D_MULTISAMPLE_ARRAY:
case GL_TEXTURE_CUBE_MAP_ARRAY:
brw_miptree_layout_texture_array(brw, mt);
break;
default:
switch (mt->msaa_layout) {
case INTEL_MSAA_LAYOUT_UMS:
case INTEL_MSAA_LAYOUT_CMS:
brw_miptree_layout_texture_array(brw, mt);
break;
case INTEL_MSAA_LAYOUT_NONE:
case INTEL_MSAA_LAYOUT_IMS:
if (use_linear_1d_layout(brw, mt))
gen9_miptree_layout_1d(mt);
else
brw_miptree_layout_2d(mt);
break;
}
break;
}
DBG("%s: %dx%dx%d\n", __FUNCTION__,
mt->total_width, mt->total_height, mt->cpp);
}
/**
* This uses a blit to copy the read buffer to a texture format which matches
* the format and type combo and then a fast read-back is done using memcpy.
* We can do arbitrary X/Y/Z/W/0/1 swizzling here as long as there is
* a format which matches the swizzling.
*
* If such a format isn't available, we fall back to _mesa_readpixels.
*
* NOTE: Some drivers use a blit to convert between tiled and linear
* texture layouts during texture uploads/downloads, so the blit
* we do here should be free in such cases.
*/
static void
st_readpixels(struct gl_context *ctx, GLint x, GLint y,
GLsizei width, GLsizei height,
GLenum format, GLenum type,
const struct gl_pixelstore_attrib *pack,
GLvoid *pixels)
{
struct st_context *st = st_context(ctx);
struct gl_renderbuffer *rb =
_mesa_get_read_renderbuffer_for_format(ctx, format);
struct st_renderbuffer *strb = st_renderbuffer(rb);
struct pipe_context *pipe = st->pipe;
struct pipe_screen *screen = pipe->screen;
struct pipe_resource *src;
struct pipe_resource *dst = NULL;
struct pipe_resource dst_templ;
enum pipe_format dst_format, src_format;
struct pipe_blit_info blit;
unsigned bind = PIPE_BIND_TRANSFER_READ;
struct pipe_transfer *tex_xfer;
ubyte *map = NULL;
/* Validate state (to be sure we have up-to-date framebuffer surfaces)
* and flush the bitmap cache prior to reading. */
st_validate_state(st);
st_flush_bitmap_cache(st);
if (!st->prefer_blit_based_texture_transfer) {
goto fallback;
}
/* This must be done after state validation. */
src = strb->texture;
/* XXX Fallback for depth-stencil formats due to an incomplete
* stencil blit implementation in some drivers. */
if (format == GL_DEPTH_STENCIL) {
goto fallback;
}
/* We are creating a texture of the size of the region being read back.
* Need to check for NPOT texture support. */
if (!screen->get_param(screen, PIPE_CAP_NPOT_TEXTURES) &&
(!util_is_power_of_two(width) ||
!util_is_power_of_two(height))) {
goto fallback;
}
/* If the base internal format and the texture format don't match, we have
* to use the slow path. */
if (rb->_BaseFormat !=
_mesa_get_format_base_format(rb->Format)) {
goto fallback;
}
/* See if the texture format already matches the format and type,
* in which case the memcpy-based fast path will likely be used and
* we don't have to blit. */
if (_mesa_format_matches_format_and_type(rb->Format, format,
type, pack->SwapBytes)) {
goto fallback;
}
if (_mesa_readpixels_needs_slow_path(ctx, format, type, GL_TRUE)) {
goto fallback;
}
/* Convert the source format to what is expected by ReadPixels
* and see if it's supported. */
src_format = util_format_linear(src->format);
src_format = util_format_luminance_to_red(src_format);
src_format = util_format_intensity_to_red(src_format);
if (!src_format ||
!screen->is_format_supported(screen, src_format, src->target,
src->nr_samples,
PIPE_BIND_SAMPLER_VIEW)) {
goto fallback;
}
if (format == GL_DEPTH_COMPONENT || format == GL_DEPTH_STENCIL)
bind |= PIPE_BIND_DEPTH_STENCIL;
else
bind |= PIPE_BIND_RENDER_TARGET;
/* Choose the destination format by finding the best match
* for the format+type combo. */
dst_format = st_choose_matching_format(screen, bind, format, type,
pack->SwapBytes);
if (dst_format == PIPE_FORMAT_NONE) {
goto fallback;
}
/* create the destination texture */
memset(&dst_templ, 0, sizeof(dst_templ));
dst_templ.target = PIPE_TEXTURE_2D;
dst_templ.format = dst_format;
dst_templ.bind = bind;
dst_templ.usage = PIPE_USAGE_STAGING;
st_gl_texture_dims_to_pipe_dims(GL_TEXTURE_2D, width, height, 1,
&dst_templ.width0, &dst_templ.height0,
&dst_templ.depth0, &dst_templ.array_size);
dst = screen->resource_create(screen, &dst_templ);
if (!dst) {
goto fallback;
}
memset(&blit, 0, sizeof(blit));
blit.src.resource = src;
blit.src.level = strb->surface->u.tex.level;
blit.src.format = src_format;
blit.dst.resource = dst;
blit.dst.level = 0;
blit.dst.format = dst->format;
blit.src.box.x = x;
blit.dst.box.x = 0;
blit.src.box.y = y;
blit.dst.box.y = 0;
blit.src.box.z = strb->surface->u.tex.first_layer;
blit.dst.box.z = 0;
blit.src.box.width = blit.dst.box.width = width;
blit.src.box.height = blit.dst.box.height = height;
blit.src.box.depth = blit.dst.box.depth = 1;
blit.mask = st_get_blit_mask(rb->_BaseFormat, format);
blit.filter = PIPE_TEX_FILTER_NEAREST;
blit.scissor_enable = FALSE;
if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) {
blit.src.box.y = rb->Height - blit.src.box.y;
blit.src.box.height = -blit.src.box.height;
}
/* blit */
st->pipe->blit(st->pipe, &blit);
/* map resources */
pixels = _mesa_map_pbo_dest(ctx, pack, pixels);
map = pipe_transfer_map_3d(pipe, dst, 0, PIPE_TRANSFER_READ,
0, 0, 0, width, height, 1, &tex_xfer);
if (!map) {
_mesa_unmap_pbo_dest(ctx, pack);
pipe_resource_reference(&dst, NULL);
goto fallback;
}
/* memcpy data into a user buffer */
{
const uint bytesPerRow = width * util_format_get_blocksize(dst_format);
GLuint row;
for (row = 0; row < (unsigned) height; row++) {
GLvoid *dest = _mesa_image_address3d(pack, pixels,
width, height, format,
type, 0, row, 0);
memcpy(dest, map, bytesPerRow);
map += tex_xfer->stride;
}
}
pipe_transfer_unmap(pipe, tex_xfer);
_mesa_unmap_pbo_dest(ctx, pack);
pipe_resource_reference(&dst, NULL);
return;
fallback:
_mesa_readpixels(ctx, x, y, width, height, format, type, pack, pixels);
}
/**
* Allocate a renderbuffer for a an on-screen window (not a user-created
* renderbuffer). The window system code determines the format.
*/
struct gl_renderbuffer *
st_new_renderbuffer_fb(enum pipe_format format, int samples, boolean sw)
{
struct st_renderbuffer *strb;
strb = ST_CALLOC_STRUCT(st_renderbuffer);
if (!strb) {
_mesa_error(NULL, GL_OUT_OF_MEMORY, "creating renderbuffer");
return NULL;
}
_mesa_init_renderbuffer(&strb->Base, 0);
strb->Base.ClassID = 0x4242; /* just a unique value */
strb->Base.NumSamples = samples;
strb->Base.Format = st_pipe_format_to_mesa_format(format);
strb->Base._BaseFormat = _mesa_get_format_base_format(strb->Base.Format);
strb->software = sw;
switch (format) {
case PIPE_FORMAT_R8G8B8A8_UNORM:
case PIPE_FORMAT_B8G8R8A8_UNORM:
case PIPE_FORMAT_A8R8G8B8_UNORM:
strb->Base.InternalFormat = GL_RGBA8;
break;
case PIPE_FORMAT_R8G8B8X8_UNORM:
case PIPE_FORMAT_B8G8R8X8_UNORM:
case PIPE_FORMAT_X8R8G8B8_UNORM:
strb->Base.InternalFormat = GL_RGB8;
break;
case PIPE_FORMAT_R8G8B8A8_SRGB:
case PIPE_FORMAT_B8G8R8A8_SRGB:
case PIPE_FORMAT_A8R8G8B8_SRGB:
strb->Base.InternalFormat = GL_SRGB8_ALPHA8;
break;
case PIPE_FORMAT_R8G8B8X8_SRGB:
case PIPE_FORMAT_B8G8R8X8_SRGB:
case PIPE_FORMAT_X8R8G8B8_SRGB:
strb->Base.InternalFormat = GL_SRGB8;
break;
case PIPE_FORMAT_B5G5R5A1_UNORM:
strb->Base.InternalFormat = GL_RGB5_A1;
break;
case PIPE_FORMAT_B4G4R4A4_UNORM:
strb->Base.InternalFormat = GL_RGBA4;
break;
case PIPE_FORMAT_B5G6R5_UNORM:
strb->Base.InternalFormat = GL_RGB565;
break;
case PIPE_FORMAT_Z16_UNORM:
strb->Base.InternalFormat = GL_DEPTH_COMPONENT16;
break;
case PIPE_FORMAT_Z32_UNORM:
strb->Base.InternalFormat = GL_DEPTH_COMPONENT32;
break;
case PIPE_FORMAT_Z24_UNORM_S8_UINT:
case PIPE_FORMAT_S8_UINT_Z24_UNORM:
strb->Base.InternalFormat = GL_DEPTH24_STENCIL8_EXT;
break;
case PIPE_FORMAT_Z24X8_UNORM:
case PIPE_FORMAT_X8Z24_UNORM:
strb->Base.InternalFormat = GL_DEPTH_COMPONENT24;
break;
case PIPE_FORMAT_S8_UINT:
strb->Base.InternalFormat = GL_STENCIL_INDEX8_EXT;
break;
case PIPE_FORMAT_R16G16B16A16_SNORM:
/* accum buffer */
strb->Base.InternalFormat = GL_RGBA16_SNORM;
break;
case PIPE_FORMAT_R16G16B16A16_UNORM:
strb->Base.InternalFormat = GL_RGBA16;
break;
case PIPE_FORMAT_R8_UNORM:
strb->Base.InternalFormat = GL_R8;
break;
case PIPE_FORMAT_R8G8_UNORM:
strb->Base.InternalFormat = GL_RG8;
break;
case PIPE_FORMAT_R16_UNORM:
strb->Base.InternalFormat = GL_R16;
break;
case PIPE_FORMAT_R16G16_UNORM:
strb->Base.InternalFormat = GL_RG16;
break;
case PIPE_FORMAT_R32G32B32A32_FLOAT:
strb->Base.InternalFormat = GL_RGBA32F;
break;
case PIPE_FORMAT_R16G16B16A16_FLOAT:
strb->Base.InternalFormat = GL_RGBA16F;
break;
default:
_mesa_problem(NULL,
"Unexpected format %s in st_new_renderbuffer_fb",
util_format_name(format));
free(strb);
return NULL;
}
/* st-specific methods */
strb->Base.Delete = st_renderbuffer_delete;
strb->Base.AllocStorage = st_renderbuffer_alloc_storage;
/* surface is allocated in st_renderbuffer_alloc_storage() */
strb->surface = NULL;
return &strb->Base;
}
/**
* The glGet queries of the framebuffer red/green/blue size, stencil size,
* etc. are satisfied by the fields of ctx->DrawBuffer->Visual. These can
* change depending on the renderbuffer bindings. This function updates
* the given framebuffer's Visual from the current renderbuffer bindings.
*
* This may apply to user-created framebuffers or window system framebuffers.
*
* Also note: ctx->DrawBuffer->Visual.depthBits might not equal
* ctx->DrawBuffer->Attachment[BUFFER_DEPTH].Renderbuffer.DepthBits.
* The former one is used to convert floating point depth values into
* integer Z values.
*/
void
_mesa_update_framebuffer_visual(struct gl_context *ctx,
struct gl_framebuffer *fb)
{
GLuint i;
memset(&fb->Visual, 0, sizeof(fb->Visual));
fb->Visual.rgbMode = GL_TRUE; /* assume this */
/* find first RGB renderbuffer */
for (i = 0; i < BUFFER_COUNT; i++) {
if (fb->Attachment[i].Renderbuffer) {
const struct gl_renderbuffer *rb = fb->Attachment[i].Renderbuffer;
const GLenum baseFormat = _mesa_get_format_base_format(rb->Format);
const mesa_format fmt = rb->Format;
/* Grab samples and sampleBuffers from any attachment point (assuming
* the framebuffer is complete, we'll get the same answer from all
* attachments).
*/
fb->Visual.samples = rb->NumSamples;
fb->Visual.sampleBuffers = rb->NumSamples > 0 ? 1 : 0;
if (_mesa_is_legal_color_format(ctx, baseFormat)) {
fb->Visual.redBits = _mesa_get_format_bits(fmt, GL_RED_BITS);
fb->Visual.greenBits = _mesa_get_format_bits(fmt, GL_GREEN_BITS);
fb->Visual.blueBits = _mesa_get_format_bits(fmt, GL_BLUE_BITS);
fb->Visual.alphaBits = _mesa_get_format_bits(fmt, GL_ALPHA_BITS);
fb->Visual.rgbBits = fb->Visual.redBits
+ fb->Visual.greenBits + fb->Visual.blueBits;
if (_mesa_get_format_color_encoding(fmt) == GL_SRGB)
fb->Visual.sRGBCapable = ctx->Extensions.EXT_framebuffer_sRGB;
break;
}
}
}
fb->Visual.floatMode = GL_FALSE;
for (i = 0; i < BUFFER_COUNT; i++) {
if (fb->Attachment[i].Renderbuffer) {
const struct gl_renderbuffer *rb = fb->Attachment[i].Renderbuffer;
const mesa_format fmt = rb->Format;
if (_mesa_get_format_datatype(fmt) == GL_FLOAT) {
fb->Visual.floatMode = GL_TRUE;
break;
}
}
}
if (fb->Attachment[BUFFER_DEPTH].Renderbuffer) {
const struct gl_renderbuffer *rb =
fb->Attachment[BUFFER_DEPTH].Renderbuffer;
const mesa_format fmt = rb->Format;
fb->Visual.haveDepthBuffer = GL_TRUE;
fb->Visual.depthBits = _mesa_get_format_bits(fmt, GL_DEPTH_BITS);
}
if (fb->Attachment[BUFFER_STENCIL].Renderbuffer) {
const struct gl_renderbuffer *rb =
fb->Attachment[BUFFER_STENCIL].Renderbuffer;
const mesa_format fmt = rb->Format;
fb->Visual.haveStencilBuffer = GL_TRUE;
fb->Visual.stencilBits = _mesa_get_format_bits(fmt, GL_STENCIL_BITS);
}
if (fb->Attachment[BUFFER_ACCUM].Renderbuffer) {
const struct gl_renderbuffer *rb =
fb->Attachment[BUFFER_ACCUM].Renderbuffer;
const mesa_format fmt = rb->Format;
fb->Visual.haveAccumBuffer = GL_TRUE;
fb->Visual.accumRedBits = _mesa_get_format_bits(fmt, GL_RED_BITS);
fb->Visual.accumGreenBits = _mesa_get_format_bits(fmt, GL_GREEN_BITS);
fb->Visual.accumBlueBits = _mesa_get_format_bits(fmt, GL_BLUE_BITS);
fb->Visual.accumAlphaBits = _mesa_get_format_bits(fmt, GL_ALPHA_BITS);
}
compute_depth_max(fb);
}
/**
* Helper for _mesa_meta_CopyTexSubImage1/2/3D() functions.
* Have to be careful with locking and meta state for pixel transfer.
*/
static void
copy_tex_sub_image(struct gl_context *ctx,
GLuint dims,
struct gl_texture_image *texImage,
GLint xoffset, GLint yoffset, GLint zoffset,
struct gl_renderbuffer *rb,
GLint x, GLint y,
GLsizei width, GLsizei height)
{
struct gl_texture_object *texObj = texImage->TexObject;
const GLenum target = texObj->Target;
GLenum format, type;
GLint bpp;
void *buf;
/* Choose format/type for temporary image buffer */
format = _mesa_get_format_base_format(texImage->TexFormat);
if (format == GL_LUMINANCE ||
format == GL_LUMINANCE_ALPHA ||
format == GL_INTENSITY) {
/* We don't want to use GL_LUMINANCE, GL_INTENSITY, etc. for the
* temp image buffer because glReadPixels will do L=R+G+B which is
* not what we want (should be L=R).
*/
format = GL_RGBA;
}
if (_mesa_is_format_integer_color(texImage->TexFormat)) {
_mesa_problem(ctx, "unsupported integer color copyteximage");
return;
}
type = get_temp_image_type(ctx, format);
bpp = _mesa_bytes_per_pixel(format, type);
if (bpp <= 0) {
_mesa_problem(ctx, "Bad bpp in meta copy_tex_sub_image()");
return;
}
/*
* Alloc image buffer (XXX could use a PBO)
*/
buf = malloc(width * height * bpp);
if (!buf) {
_mesa_error(ctx, GL_OUT_OF_MEMORY, "glCopyTexSubImage%uD", dims);
return;
}
_mesa_unlock_texture(ctx, texObj); /* need to unlock first */
/*
* Read image from framebuffer (disable pixel transfer ops)
*/
_mesa_meta_begin(ctx, MESA_META_PIXEL_STORE | MESA_META_PIXEL_TRANSFER);
ctx->Driver.ReadPixels(ctx, x, y, width, height,
format, type, &ctx->Pack, buf);
_mesa_meta_end(ctx);
_mesa_update_state(ctx); /* to update pixel transfer state */
/*
* Store texture data (with pixel transfer ops)
*/
_mesa_meta_begin(ctx, MESA_META_PIXEL_STORE);
if (target == GL_TEXTURE_1D) {
ctx->Driver.TexSubImage1D(ctx, texImage,
xoffset, width,
format, type, buf, &ctx->Unpack);
}
else {
ctx->Driver.TexSubImage2D(ctx, texImage,
xoffset, yoffset, width, height,
format, type, buf, &ctx->Unpack);
}
_mesa_meta_end(ctx);
_mesa_lock_texture(ctx, texObj); /* re-lock */
free(buf);
}
/*
* Read R, G, B, A, RGB, L, or LA pixels.
*/
static void
read_rgba_pixels( struct gl_context *ctx,
GLint x, GLint y,
GLsizei width, GLsizei height,
GLenum format, GLenum type, GLvoid *pixels,
const struct gl_pixelstore_attrib *packing )
{
GLbitfield transferOps;
bool dst_is_integer, dst_is_luminance, needs_rebase;
int dst_stride, src_stride, rb_stride;
uint32_t dst_format, src_format;
GLubyte *dst, *map;
mesa_format rb_format;
bool needs_rgba;
void *rgba, *src;
bool src_is_uint = false;
uint8_t rebase_swizzle[4];
struct gl_framebuffer *fb = ctx->ReadBuffer;
struct gl_renderbuffer *rb = fb->_ColorReadBuffer;
if (!rb)
return;
transferOps = get_readpixels_transfer_ops(ctx, rb->Format, format, type,
GL_FALSE);
/* Describe the dst format */
dst_is_integer = _mesa_is_enum_format_integer(format);
dst_stride = _mesa_image_row_stride(packing, width, format, type);
dst_format = _mesa_format_from_format_and_type(format, type);
dst_is_luminance = format == GL_LUMINANCE ||
format == GL_LUMINANCE_ALPHA ||
format == GL_LUMINANCE_INTEGER_EXT ||
format == GL_LUMINANCE_ALPHA_INTEGER_EXT;
dst = (GLubyte *) _mesa_image_address2d(packing, pixels, width, height,
format, type, 0, 0);
/* Map the source render buffer */
ctx->Driver.MapRenderbuffer(ctx, rb, x, y, width, height, GL_MAP_READ_BIT,
&map, &rb_stride);
if (!map) {
_mesa_error(ctx, GL_OUT_OF_MEMORY, "glReadPixels");
return;
}
rb_format = _mesa_get_srgb_format_linear(rb->Format);
/*
* Depending on the base formats involved in the conversion we might need to
* rebase some values, so for these formats we compute a rebase swizzle.
*/
if (rb->_BaseFormat == GL_LUMINANCE || rb->_BaseFormat == GL_INTENSITY) {
needs_rebase = true;
rebase_swizzle[0] = MESA_FORMAT_SWIZZLE_X;
rebase_swizzle[1] = MESA_FORMAT_SWIZZLE_ZERO;
rebase_swizzle[2] = MESA_FORMAT_SWIZZLE_ZERO;
rebase_swizzle[3] = MESA_FORMAT_SWIZZLE_ONE;
} else if (rb->_BaseFormat == GL_LUMINANCE_ALPHA) {
needs_rebase = true;
rebase_swizzle[0] = MESA_FORMAT_SWIZZLE_X;
rebase_swizzle[1] = MESA_FORMAT_SWIZZLE_ZERO;
rebase_swizzle[2] = MESA_FORMAT_SWIZZLE_ZERO;
rebase_swizzle[3] = MESA_FORMAT_SWIZZLE_W;
} else if (_mesa_get_format_base_format(rb_format) != rb->_BaseFormat) {
needs_rebase =
_mesa_compute_rgba2base2rgba_component_mapping(rb->_BaseFormat,
rebase_swizzle);
} else {
needs_rebase = false;
}
/* Since _mesa_format_convert does not handle transferOps we need to handle
* them before we call the function. This requires to convert to RGBA float
* first so we can call _mesa_apply_rgba_transfer_ops. If the dst format is
* integer transferOps do not apply.
*
* Converting to luminance also requires converting to RGBA first, so we can
* then compute luminance values as L=R+G+B. Notice that this is different
* from GetTexImage, where we compute L=R.
*/
assert(!transferOps || (transferOps && !dst_is_integer));
needs_rgba = transferOps || dst_is_luminance;
rgba = NULL;
if (needs_rgba) {
uint32_t rgba_format;
int rgba_stride;
bool need_convert;
/* Convert to RGBA float or int/uint depending on the type of the src */
if (dst_is_integer) {
src_is_uint = _mesa_is_format_unsigned(rb_format);
if (src_is_uint) {
rgba_format = RGBA32_UINT;
rgba_stride = width * 4 * sizeof(GLuint);
} else {
rgba_format = RGBA32_INT;
rgba_stride = width * 4 * sizeof(GLint);
}
} else {
rgba_format = RGBA32_FLOAT;
rgba_stride = width * 4 * sizeof(GLfloat);
}
/* If we are lucky and the dst format matches the RGBA format we need to
* convert to, then we can convert directly into the dst buffer and avoid
* the final conversion/copy from the rgba buffer to the dst buffer.
*/
if (dst_format == rgba_format) {
need_convert = false;
rgba = dst;
} else {
need_convert = true;
rgba = malloc(height * rgba_stride);
if (!rgba) {
_mesa_error(ctx, GL_OUT_OF_MEMORY, "glReadPixels");
goto done_unmap;
}
}
/* Convert to RGBA now */
_mesa_format_convert(rgba, rgba_format, rgba_stride,
map, rb_format, rb_stride,
width, height,
needs_rebase ? rebase_swizzle : NULL);
/* Handle transfer ops if necessary */
if (transferOps)
_mesa_apply_rgba_transfer_ops(ctx, transferOps, width * height, rgba);
/* If we had to rebase, we have already taken care of that */
needs_rebase = false;
/* If we were lucky and our RGBA conversion matches the dst format, then
* we are done.
*/
if (!need_convert)
goto done_swap;
/* Otherwise, we need to convert from RGBA to dst next */
src = rgba;
src_format = rgba_format;
src_stride = rgba_stride;
} else {
/* No RGBA conversion needed, convert directly to dst */
src = map;
src_format = rb_format;
src_stride = rb_stride;
}
/* Do the conversion.
*
* If the dst format is Luminance, we need to do the conversion by computing
* L=R+G+B values.
*/
if (!dst_is_luminance) {
_mesa_format_convert(dst, dst_format, dst_stride,
src, src_format, src_stride,
width, height,
needs_rebase ? rebase_swizzle : NULL);
} else if (!dst_is_integer) {
/* Compute float Luminance values from RGBA float */
int luminance_stride, luminance_bytes;
void *luminance;
uint32_t luminance_format;
luminance_stride = width * sizeof(GL_FLOAT);
if (format == GL_LUMINANCE_ALPHA)
luminance_stride *= 2;
luminance_bytes = height * luminance_stride;
luminance = malloc(luminance_bytes);
if (!luminance) {
_mesa_error(ctx, GL_OUT_OF_MEMORY, "glReadPixels");
free(rgba);
goto done_unmap;
}
_mesa_pack_luminance_from_rgba_float(width * height, src,
luminance, format, transferOps);
/* Convert from Luminance float to dst (this will hadle type conversion
* from float to the type of dst if necessary)
*/
luminance_format = _mesa_format_from_format_and_type(format, GL_FLOAT);
_mesa_format_convert(dst, dst_format, dst_stride,
luminance, luminance_format, luminance_stride,
width, height, NULL);
} else {
_mesa_pack_luminance_from_rgba_integer(width * height, src, !src_is_uint,
dst, format, type);
}
if (rgba)
free(rgba);
done_swap:
/* Handle byte swapping if required */
if (packing->SwapBytes) {
int components = _mesa_components_in_format(format);
GLint swapSize = _mesa_sizeof_packed_type(type);
if (swapSize == 2)
_mesa_swap2((GLushort *) dst, width * height * components);
else if (swapSize == 4)
_mesa_swap4((GLuint *) dst, width * height * components);
}
done_unmap:
ctx->Driver.UnmapRenderbuffer(ctx, rb);
}
/** A partial implementation of glCopyImageSubData
*
* This is a partial implementation of glCopyImageSubData that works only
* if both textures are uncompressed and the destination texture is
* renderable. It uses a slight abuse of a texture view (see make_view) to
* turn the source texture into the destination texture type and then uses
* _mesa_meta_BlitFramebuffers to do the copy.
*/
bool
_mesa_meta_CopyImageSubData_uncompressed(struct gl_context *ctx,
struct gl_texture_image *src_tex_image,
struct gl_renderbuffer *src_renderbuffer,
int src_x, int src_y, int src_z,
struct gl_texture_image *dst_tex_image,
struct gl_renderbuffer *dst_renderbuffer,
int dst_x, int dst_y, int dst_z,
int src_width, int src_height)
{
mesa_format src_format, dst_format;
GLint src_internal_format, dst_internal_format;
GLuint src_view_texture = 0;
struct gl_texture_image *src_view_tex_image;
struct gl_framebuffer *readFb;
struct gl_framebuffer *drawFb = NULL;
bool success = false;
GLbitfield mask;
GLenum status, attachment;
if (src_renderbuffer) {
src_format = src_renderbuffer->Format;
src_internal_format = src_renderbuffer->InternalFormat;
} else {
assert(src_tex_image);
src_format = src_tex_image->TexFormat;
src_internal_format = src_tex_image->InternalFormat;
}
if (dst_renderbuffer) {
dst_format = dst_renderbuffer->Format;
dst_internal_format = dst_renderbuffer->InternalFormat;
} else {
assert(dst_tex_image);
dst_format = dst_tex_image->TexFormat;
dst_internal_format = dst_tex_image->InternalFormat;
}
if (_mesa_is_format_compressed(src_format))
return false;
if (_mesa_is_format_compressed(dst_format))
return false;
if (src_internal_format == dst_internal_format) {
src_view_tex_image = src_tex_image;
} else {
if (src_renderbuffer) {
assert(src_tex_image == NULL);
src_tex_image = wrap_renderbuffer(ctx, src_renderbuffer);
}
if (!make_view(ctx, src_tex_image, &src_view_tex_image, &src_view_texture,
dst_internal_format))
goto cleanup;
}
/* We really only need to stash the bound framebuffers and scissor. */
_mesa_meta_begin(ctx, MESA_META_SCISSOR);
readFb = ctx->Driver.NewFramebuffer(ctx, 0xDEADBEEF);
if (readFb == NULL)
goto meta_end;
drawFb = ctx->Driver.NewFramebuffer(ctx, 0xDEADBEEF);
if (drawFb == NULL)
goto meta_end;
_mesa_bind_framebuffers(ctx, drawFb, readFb);
switch (_mesa_get_format_base_format(src_format)) {
case GL_DEPTH_COMPONENT:
attachment = GL_DEPTH_ATTACHMENT;
mask = GL_DEPTH_BUFFER_BIT;
break;
case GL_DEPTH_STENCIL:
attachment = GL_DEPTH_STENCIL_ATTACHMENT;
mask = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
break;
case GL_STENCIL_INDEX:
attachment = GL_STENCIL_ATTACHMENT;
mask = GL_STENCIL_BUFFER_BIT;
break;
default:
attachment = GL_COLOR_ATTACHMENT0;
mask = GL_COLOR_BUFFER_BIT;
_mesa_DrawBuffer(GL_COLOR_ATTACHMENT0);
_mesa_ReadBuffer(GL_COLOR_ATTACHMENT0);
}
if (src_view_tex_image) {
/* Prefer the tex image because, even if we have a renderbuffer, we may
* have had to wrap it in a texture view.
*/
_mesa_meta_framebuffer_texture_image(ctx, ctx->ReadBuffer, attachment,
src_view_tex_image, src_z);
} else {
_mesa_framebuffer_renderbuffer(ctx, ctx->ReadBuffer, attachment,
src_renderbuffer);
}
status = _mesa_check_framebuffer_status(ctx, ctx->ReadBuffer);
if (status != GL_FRAMEBUFFER_COMPLETE)
goto meta_end;
if (dst_renderbuffer) {
_mesa_framebuffer_renderbuffer(ctx, ctx->DrawBuffer, attachment,
dst_renderbuffer);
} else {
_mesa_meta_framebuffer_texture_image(ctx, ctx->DrawBuffer, attachment,
dst_tex_image, dst_z);
}
status = _mesa_check_framebuffer_status(ctx, ctx->DrawBuffer);
if (status != GL_FRAMEBUFFER_COMPLETE)
goto meta_end;
/* Explicitly disable sRGB encoding */
ctx->DrawBuffer->Visual.sRGBCapable = false;
/* Since we've bound a new draw framebuffer, we need to update its
* derived state -- _Xmin, etc -- for BlitFramebuffer's clipping to
* be correct.
*/
_mesa_update_state(ctx);
/* We skip the core BlitFramebuffer checks for format consistency.
* We have already created views to ensure that the texture formats
* match.
*/
ctx->Driver.BlitFramebuffer(ctx, ctx->ReadBuffer, ctx->DrawBuffer,
src_x, src_y,
src_x + src_width, src_y + src_height,
dst_x, dst_y,
dst_x + src_width, dst_y + src_height,
mask, GL_NEAREST);
success = true;
meta_end:
_mesa_reference_framebuffer(&readFb, NULL);
_mesa_reference_framebuffer(&drawFb, NULL);
_mesa_meta_end(ctx);
cleanup:
_mesa_DeleteTextures(1, &src_view_texture);
/* If we got a renderbuffer source, delete the temporary texture */
if (src_renderbuffer && src_tex_image)
ctx->Driver.DeleteTexture(ctx, src_tex_image->TexObject);
return success;
}
/**
* Allocate a new driRenderbuffer object.
* Individual drivers are free to implement different versions of
* this function.
*
* At this time, this function can only be used for window-system
* renderbuffers, not user-created RBOs.
*
* \param format Either GL_RGBA, GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT24,
* GL_DEPTH_COMPONENT32, or GL_STENCIL_INDEX8_EXT (for now).
* \param addr address in main memory of the buffer. Probably a memory
* mapped region.
* \param cpp chars or bytes per pixel
* \param offset start of renderbuffer with respect to start of framebuffer
* \param pitch pixels per row
*/
driRenderbuffer *
driNewRenderbuffer(gl_format format, GLvoid *addr,
GLint cpp, GLint offset, GLint pitch,
__DRIdrawablePrivate *dPriv)
{
driRenderbuffer *drb;
assert(format == GL_RGBA ||
format == GL_RGB5 ||
format == GL_RGBA8 ||
format == GL_DEPTH_COMPONENT16 ||
format == GL_DEPTH_COMPONENT24 ||
format == GL_DEPTH_COMPONENT32 ||
format == GL_STENCIL_INDEX8_EXT);
assert(cpp > 0);
assert(pitch > 0);
drb = _mesa_calloc(sizeof(driRenderbuffer));
if (drb) {
const GLuint name = 0;
_mesa_init_renderbuffer(&drb->Base, name);
/* Make sure we're using a null-valued GetPointer routine */
assert(drb->Base.GetPointer(NULL, &drb->Base, 0, 0) == NULL);
switch (format) {
case MESA_FORMAT_ARGB8888:
if (cpp == 2) {
/* override format */
format = MESA_FORMAT_RGB565;
}
drb->Base.DataType = GL_UNSIGNED_BYTE;
break;
case MESA_FORMAT_Z16:
/* Depth */
/* we always Get/Put 32-bit Z values */
drb->Base.DataType = GL_UNSIGNED_INT;
assert(cpp == 2);
break;
case MESA_FORMAT_Z32:
/* Depth */
/* we always Get/Put 32-bit Z values */
drb->Base.DataType = GL_UNSIGNED_INT;
assert(cpp == 4);
break;
case MESA_FORMAT_Z24_S8:
drb->Base.DataType = GL_UNSIGNED_INT_24_8_EXT;
assert(cpp == 4);
break;
case MESA_FORMAT_S8_Z24:
drb->Base.DataType = GL_UNSIGNED_INT_24_8_EXT;
assert(cpp == 4);
break;
case MESA_FORMAT_S8:
/* Stencil */
drb->Base.DataType = GL_UNSIGNED_BYTE;
break;
default:
_mesa_problem(NULL, "Bad format 0x%x in driNewRenderbuffer", format);
return NULL;
}
drb->Base.Format = format;
drb->Base.InternalFormat =
drb->Base._BaseFormat = _mesa_get_format_base_format(format);
drb->Base.AllocStorage = driRenderbufferStorage;
drb->Base.Delete = driDeleteRenderbuffer;
drb->Base.Data = addr;
/* DRI renderbuffer-specific fields: */
drb->dPriv = dPriv;
drb->offset = offset;
drb->pitch = pitch;
drb->cpp = cpp;
/* may be changed if page flipping is active: */
drb->flippedOffset = offset;
drb->flippedPitch = pitch;
drb->flippedData = addr;
}
return drb;
}
static struct gl_texture_image *
create_texture_for_pbo(struct gl_context *ctx, bool create_pbo,
GLenum pbo_target, int width, int height,
GLenum format, GLenum type, const void *pixels,
const struct gl_pixelstore_attrib *packing,
GLuint *tmp_pbo, GLuint *tmp_tex)
{
uint32_t pbo_format;
GLenum internal_format;
unsigned row_stride;
struct gl_buffer_object *buffer_obj;
struct gl_texture_object *tex_obj;
struct gl_texture_image *tex_image;
bool read_only;
if (packing->SwapBytes ||
packing->LsbFirst ||
packing->Invert)
return NULL;
pbo_format = _mesa_format_from_format_and_type(format, type);
if (_mesa_format_is_mesa_array_format(pbo_format))
pbo_format = _mesa_format_from_array_format(pbo_format);
if (!pbo_format || !ctx->TextureFormatSupported[pbo_format])
return NULL;
/* Account for SKIP_PIXELS, SKIP_ROWS, ALIGNMENT, and SKIP_IMAGES */
pixels = _mesa_image_address3d(packing, pixels,
width, height, format, type, 0, 0, 0);
row_stride = _mesa_image_row_stride(packing, width, format, type);
if (_mesa_is_bufferobj(packing->BufferObj)) {
*tmp_pbo = 0;
buffer_obj = packing->BufferObj;
} else {
bool is_pixel_pack = pbo_target == GL_PIXEL_PACK_BUFFER;
assert(create_pbo);
_mesa_GenBuffers(1, tmp_pbo);
/* We are not doing this inside meta_begin/end. However, we know the
* client doesn't have the given target bound, so we can go ahead and
* squash it. We'll set it back when we're done.
*/
_mesa_BindBuffer(pbo_target, *tmp_pbo);
/* In case of GL_PIXEL_PACK_BUFFER, pass null pointer for the pixel
* data to avoid unnecessary data copying in _mesa_BufferData().
*/
if (is_pixel_pack)
_mesa_BufferData(pbo_target, row_stride * height, NULL,
GL_STREAM_READ);
else
_mesa_BufferData(pbo_target, row_stride * height, pixels,
GL_STREAM_DRAW);
buffer_obj = packing->BufferObj;
pixels = NULL;
_mesa_BindBuffer(pbo_target, 0);
}
_mesa_GenTextures(1, tmp_tex);
tex_obj = _mesa_lookup_texture(ctx, *tmp_tex);
_mesa_initialize_texture_object(ctx, tex_obj, *tmp_tex, GL_TEXTURE_2D);
/* This must be set after _mesa_initialize_texture_object, not before. */
tex_obj->Immutable = GL_TRUE;
/* This is required for interactions with ARB_texture_view. */
tex_obj->NumLayers = 1;
internal_format = _mesa_get_format_base_format(pbo_format);
tex_image = _mesa_get_tex_image(ctx, tex_obj, tex_obj->Target, 0);
_mesa_init_teximage_fields(ctx, tex_image, width, height, 1,
0, internal_format, pbo_format);
read_only = pbo_target == GL_PIXEL_UNPACK_BUFFER;
if (!ctx->Driver.SetTextureStorageForBufferObject(ctx, tex_obj,
buffer_obj,
(intptr_t)pixels,
row_stride,
read_only)) {
_mesa_DeleteTextures(1, tmp_tex);
_mesa_DeleteBuffers(1, tmp_pbo);
return NULL;
}
return tex_image;
}
static unsigned int
intel_vertical_texture_alignment_unit(struct brw_context *brw,
mesa_format format, bool multisampled)
{
/**
* From the "Alignment Unit Size" section of various specs, namely:
* - Gen3 Spec: "Memory Data Formats" Volume, Section 1.20.1.4
* - i965 and G45 PRMs: Volume 1, Section 6.17.3.4.
* - Ironlake and Sandybridge PRMs: Volume 1, Part 1, Section 7.18.3.4
* - BSpec (for Ivybridge and slight variations in separate stencil)
*
* +----------------------------------------------------------------------+
* | | alignment unit height ("j") |
* | Surface Property |-----------------------------|
* | | 915 | 965 | ILK | SNB | IVB |
* +----------------------------------------------------------------------+
* | BC1-5 compressed format (DXTn/S3TC) | 4 | 4 | 4 | 4 | 4 |
* | FXT1 compressed format | 4 | 4 | 4 | 4 | 4 |
* | Depth Buffer | 2 | 2 | 2 | 4 | 4 |
* | Separate Stencil Buffer | N/A | N/A | N/A | 4 | 8 |
* | Multisampled (4x or 8x) render target | N/A | N/A | N/A | 4 | 4 |
* | All Others | 2 | 2 | 2 | * | * |
* +----------------------------------------------------------------------+
*
* Where "*" means either VALIGN_2 or VALIGN_4 depending on the setting of
* the SURFACE_STATE "Surface Vertical Alignment" field.
*/
if (_mesa_is_format_compressed(format))
return 4;
if (format == MESA_FORMAT_S_UINT8)
return brw->gen >= 7 ? 8 : 4;
/* Broadwell only supports VALIGN of 4, 8, and 16. The BSpec says 4
* should always be used, except for stencil buffers, which should be 8.
*/
if (brw->gen >= 8)
return 4;
if (multisampled)
return 4;
GLenum base_format = _mesa_get_format_base_format(format);
if (brw->gen >= 6 &&
(base_format == GL_DEPTH_COMPONENT ||
base_format == GL_DEPTH_STENCIL)) {
return 4;
}
if (brw->gen == 7) {
/* On Gen7, we prefer a vertical alignment of 4 when possible, because
* that allows Y tiled render targets.
*
* From the Ivy Bridge PRM, Vol4 Part1 2.12.2.1 (SURFACE_STATE for most
* messages), on p64, under the heading "Surface Vertical Alignment":
*
* Value of 1 [VALIGN_4] is not supported for format YCRCB_NORMAL
* (0x182), YCRCB_SWAPUVY (0x183), YCRCB_SWAPUV (0x18f), YCRCB_SWAPY
* (0x190)
*
* VALIGN_4 is not supported for surface format R32G32B32_FLOAT.
*/
if (base_format == GL_YCBCR_MESA || format == MESA_FORMAT_RGB_FLOAT32)
return 2;
return 4;
}
return 2;
}
void
brw_workaround_depthstencil_alignment(struct brw_context *brw,
GLbitfield clear_mask)
{
struct gl_context *ctx = &brw->ctx;
struct gl_framebuffer *fb = ctx->DrawBuffer;
bool rebase_depth = false;
bool rebase_stencil = false;
struct intel_renderbuffer *depth_irb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
struct intel_renderbuffer *stencil_irb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
struct intel_mipmap_tree *depth_mt = NULL;
struct intel_mipmap_tree *stencil_mt = get_stencil_miptree(stencil_irb);
uint32_t tile_x = 0, tile_y = 0, stencil_tile_x = 0, stencil_tile_y = 0;
uint32_t stencil_draw_x = 0, stencil_draw_y = 0;
bool invalidate_depth = clear_mask & BUFFER_BIT_DEPTH;
bool invalidate_stencil = clear_mask & BUFFER_BIT_STENCIL;
if (depth_irb)
depth_mt = depth_irb->mt;
/* Initialize brw->depthstencil to 'nop' workaround state.
*/
brw->depthstencil.tile_x = 0;
brw->depthstencil.tile_y = 0;
brw->depthstencil.depth_offset = 0;
brw->depthstencil.stencil_offset = 0;
brw->depthstencil.hiz_offset = 0;
brw->depthstencil.depth_mt = NULL;
brw->depthstencil.stencil_mt = NULL;
if (depth_irb)
brw->depthstencil.depth_mt = depth_mt;
if (stencil_irb)
brw->depthstencil.stencil_mt = get_stencil_miptree(stencil_irb);
/* Gen7+ doesn't require the workarounds, since we always program the
* surface state at the start of the whole surface.
*/
if (brw->gen >= 7)
return;
/* Check if depth buffer is in depth/stencil format. If so, then it's only
* safe to invalidate it if we're also clearing stencil, and both depth_irb
* and stencil_irb point to the same miptree.
*
* Note: it's not sufficient to check for the case where
* _mesa_get_format_base_format(depth_mt->format) == GL_DEPTH_STENCIL,
* because this fails to catch depth/stencil buffers on hardware that uses
* separate stencil. To catch that case, we check whether
* depth_mt->stencil_mt is non-NULL.
*/
if (depth_irb && invalidate_depth &&
(_mesa_get_format_base_format(depth_mt->format) == GL_DEPTH_STENCIL ||
depth_mt->stencil_mt)) {
invalidate_depth = invalidate_stencil && depth_irb && stencil_irb
&& depth_irb->mt == stencil_irb->mt;
}
uint32_t tile_mask_x, tile_mask_y;
brw_get_depthstencil_tile_masks(depth_mt,
depth_mt ? depth_irb->mt_level : 0,
depth_mt ? depth_irb->mt_layer : 0,
stencil_mt,
&tile_mask_x, &tile_mask_y);
if (depth_irb) {
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
/* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
* (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
* Coordinate Offset X/Y":
*
* "The 3 LSBs of both offsets must be zero to ensure correct
* alignment"
*/
if (tile_x & 7 || tile_y & 7)
rebase_depth = true;
/* We didn't even have intra-tile offsets before g45. */
if (!brw->has_surface_tile_offset) {
if (tile_x || tile_y)
rebase_depth = true;
}
if (rebase_depth) {
perf_debug("HW workaround: blitting depth level %d to a temporary "
"to fix alignment (depth tile offset %d,%d)\n",
depth_irb->mt_level, tile_x, tile_y);
intel_renderbuffer_move_to_temp(brw, depth_irb, invalidate_depth);
/* In the case of stencil_irb being the same packed depth/stencil
* texture but not the same rb, make it point at our rebased mt, too.
*/
if (stencil_irb &&
stencil_irb != depth_irb &&
stencil_irb->mt == depth_mt) {
intel_miptree_reference(&stencil_irb->mt, depth_irb->mt);
intel_renderbuffer_set_draw_offset(stencil_irb);
}
stencil_mt = get_stencil_miptree(stencil_irb);
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
}
if (stencil_irb) {
stencil_mt = get_stencil_miptree(stencil_irb);
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
int stencil_tile_x = stencil_draw_x & tile_mask_x;
int stencil_tile_y = stencil_draw_y & tile_mask_y;
/* If stencil doesn't match depth, then we'll need to rebase stencil
* as well. (if we hadn't decided to rebase stencil before, the
* post-stencil depth test will also rebase depth to try to match it
* up).
*/
if (tile_x != stencil_tile_x ||
tile_y != stencil_tile_y) {
rebase_stencil = true;
}
}
}
/* If we have (just) stencil, check it for ignored low bits as well */
if (stencil_irb) {
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
stencil_tile_x = stencil_draw_x & tile_mask_x;
stencil_tile_y = stencil_draw_y & tile_mask_y;
if (stencil_tile_x & 7 || stencil_tile_y & 7)
rebase_stencil = true;
if (!brw->has_surface_tile_offset) {
if (stencil_tile_x || stencil_tile_y)
rebase_stencil = true;
}
}
if (rebase_stencil) {
perf_debug("HW workaround: blitting stencil level %d to a temporary "
"to fix alignment (stencil tile offset %d,%d)\n",
stencil_irb->mt_level, stencil_tile_x, stencil_tile_y);
intel_renderbuffer_move_to_temp(brw, stencil_irb, invalidate_stencil);
stencil_mt = get_stencil_miptree(stencil_irb);
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
stencil_tile_x = stencil_draw_x & tile_mask_x;
stencil_tile_y = stencil_draw_y & tile_mask_y;
if (depth_irb && depth_irb->mt == stencil_irb->mt) {
intel_miptree_reference(&depth_irb->mt, stencil_irb->mt);
intel_renderbuffer_set_draw_offset(depth_irb);
} else if (depth_irb && !rebase_depth) {
if (tile_x != stencil_tile_x ||
tile_y != stencil_tile_y) {
perf_debug("HW workaround: blitting depth level %d to a temporary "
"to match stencil level %d alignment (depth tile offset "
"%d,%d, stencil offset %d,%d)\n",
depth_irb->mt_level,
stencil_irb->mt_level,
tile_x, tile_y,
stencil_tile_x, stencil_tile_y);
intel_renderbuffer_move_to_temp(brw, depth_irb, invalidate_depth);
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
if (stencil_irb && stencil_irb->mt == depth_mt) {
intel_miptree_reference(&stencil_irb->mt, depth_irb->mt);
intel_renderbuffer_set_draw_offset(stencil_irb);
}
WARN_ONCE(stencil_tile_x != tile_x ||
stencil_tile_y != tile_y,
"Rebased stencil tile offset (%d,%d) doesn't match depth "
"tile offset (%d,%d).\n",
stencil_tile_x, stencil_tile_y,
tile_x, tile_y);
}
}
}
if (!depth_irb) {
tile_x = stencil_tile_x;
tile_y = stencil_tile_y;
}
/* While we just tried to get everything aligned, we may have failed to do
* so in the case of rendering to array or 3D textures, where nonzero faces
* will still have an offset post-rebase. At least give an informative
* warning.
*/
WARN_ONCE((tile_x & 7) || (tile_y & 7),
"Depth/stencil buffer needs alignment to 8-pixel boundaries.\n"
"Truncating offset, bad rendering may occur.\n");
tile_x &= ~7;
tile_y &= ~7;
/* Now, after rebasing, save off the new dephtstencil state so the hardware
* packets can just dereference that without re-calculating tile offsets.
*/
brw->depthstencil.tile_x = tile_x;
brw->depthstencil.tile_y = tile_y;
if (depth_irb) {
depth_mt = depth_irb->mt;
brw->depthstencil.depth_mt = depth_mt;
brw->depthstencil.depth_offset =
intel_region_get_aligned_offset(depth_mt->region,
depth_irb->draw_x & ~tile_mask_x,
depth_irb->draw_y & ~tile_mask_y,
false);
if (intel_renderbuffer_has_hiz(depth_irb)) {
brw->depthstencil.hiz_offset =
intel_region_get_aligned_offset(depth_mt->region,
depth_irb->draw_x & ~tile_mask_x,
(depth_irb->draw_y & ~tile_mask_y) /
2,
false);
}
}
if (stencil_irb) {
stencil_mt = get_stencil_miptree(stencil_irb);
brw->depthstencil.stencil_mt = stencil_mt;
if (stencil_mt->format == MESA_FORMAT_S_UINT8) {
/* Note: we can't compute the stencil offset using
* intel_region_get_aligned_offset(), because stencil_region claims
* that the region is untiled even though it's W tiled.
*/
brw->depthstencil.stencil_offset =
(stencil_draw_y & ~tile_mask_y) * stencil_mt->region->pitch +
(stencil_draw_x & ~tile_mask_x) * 64;
}
}
}
