*/
#include "main/teximage.h"
+#include "main/fbobject.h"
+#include "main/renderbuffer.h"
#include "glsl/ralloc.h"
#include "brw_eu.h"
#include "brw_state.h"
+#define FILE_DEBUG_FLAG DEBUG_BLORP
/**
* Helper function for handling mirror image blits.
* If coord0 > coord1, swap them and invert the "mirror" boolean.
*/
static inline void
-fixup_mirroring(bool &mirror, GLint &coord0, GLint &coord1)
+fixup_mirroring(bool &mirror, GLfloat &coord0, GLfloat &coord1)
{
if (coord0 > coord1) {
mirror = !mirror;
- GLint tmp = coord0;
+ GLfloat tmp = coord0;
coord0 = coord1;
coord1 = tmp;
}
* For clarity, the nomenclature of this function assumes we are clipping and
* scissoring the X coordinate; the exact same logic applies for Y
* coordinates.
+ *
+ * Note: this function may also be used to account for clipping of source
+ * coordinates, by swapping the roles of src and dst.
*/
static inline bool
-clip_or_scissor(bool mirror, GLint &src_x0, GLint &src_x1, GLint &dst_x0,
- GLint &dst_x1, GLint fb_xmin, GLint fb_xmax)
+clip_or_scissor(bool mirror, GLfloat &src_x0, GLfloat &src_x1, GLfloat &dst_x0,
+ GLfloat &dst_x1, GLfloat fb_xmin, GLfloat fb_xmax)
{
+ float scale = (float) (src_x1 - src_x0) / (dst_x1 - dst_x0);
/* If we are going to scissor everything away, stop. */
if (!(fb_xmin < fb_xmax &&
dst_x0 < fb_xmax &&
/* Adjust the source rectangle to remove the pixels corresponding to those
* that were clipped/scissored out of the destination rectangle.
*/
- src_x0 += pixels_clipped_left;
- src_x1 -= pixels_clipped_right;
+ src_x0 += pixels_clipped_left * scale;
+ src_x1 -= pixels_clipped_right * scale;
return true;
}
+static struct intel_mipmap_tree *
+find_miptree(GLbitfield buffer_bit, struct intel_renderbuffer *irb)
+{
+ struct intel_mipmap_tree *mt = irb->mt;
+ if (buffer_bit == GL_STENCIL_BUFFER_BIT && mt->stencil_mt)
+ mt = mt->stencil_mt;
+ return mt;
+}
+
+void
+brw_blorp_blit_miptrees(struct brw_context *brw,
+ struct intel_mipmap_tree *src_mt,
+ unsigned src_level, unsigned src_layer,
+ struct intel_mipmap_tree *dst_mt,
+ unsigned dst_level, unsigned dst_layer,
+ float src_x0, float src_y0,
+ float src_x1, float src_y1,
+ float dst_x0, float dst_y0,
+ float dst_x1, float dst_y1,
+ GLenum filter, bool mirror_x, bool mirror_y)
+{
+ /* Get ready to blit. This includes depth resolving the src and dst
+ * buffers if necessary. Note: it's not necessary to do a color resolve on
+ * the destination buffer because we use the standard render path to render
+ * to destination color buffers, and the standard render path is
+ * fast-color-aware.
+ */
+ intel_miptree_resolve_color(brw, src_mt);
+ intel_miptree_slice_resolve_depth(brw, src_mt, src_level, src_layer);
+ intel_miptree_slice_resolve_depth(brw, dst_mt, dst_level, dst_layer);
+
+ DBG("%s from %s mt %p %d %d (%f,%f) (%f,%f)"
+ "to %s mt %p %d %d (%f,%f) (%f,%f) (flip %d,%d)\n",
+ __FUNCTION__,
+ _mesa_get_format_name(src_mt->format), src_mt,
+ src_level, src_layer, src_x0, src_y0, src_x1, src_y1,
+ _mesa_get_format_name(dst_mt->format), dst_mt,
+ dst_level, dst_layer, dst_x0, dst_y0, dst_x1, dst_y1,
+ mirror_x, mirror_y);
+
+ brw_blorp_blit_params params(brw,
+ src_mt, src_level, src_layer,
+ dst_mt, dst_level, dst_layer,
+ src_x0, src_y0,
+ src_x1, src_y1,
+ dst_x0, dst_y0,
+ dst_x1, dst_y1,
+ filter, mirror_x, mirror_y);
+ brw_blorp_exec(brw, ¶ms);
+
+ intel_miptree_slice_set_needs_hiz_resolve(dst_mt, dst_level, dst_layer);
+}
+
+static void
+do_blorp_blit(struct brw_context *brw, GLbitfield buffer_bit,
+ struct intel_renderbuffer *src_irb,
+ struct intel_renderbuffer *dst_irb,
+ GLfloat srcX0, GLfloat srcY0, GLfloat srcX1, GLfloat srcY1,
+ GLfloat dstX0, GLfloat dstY0, GLfloat dstX1, GLfloat dstY1,
+ GLenum filter, bool mirror_x, bool mirror_y)
+{
+ /* Find source/dst miptrees */
+ struct intel_mipmap_tree *src_mt = find_miptree(buffer_bit, src_irb);
+ struct intel_mipmap_tree *dst_mt = find_miptree(buffer_bit, dst_irb);
+
+ /* Do the blit */
+ brw_blorp_blit_miptrees(brw,
+ src_mt, src_irb->mt_level, src_irb->mt_layer,
+ dst_mt, dst_irb->mt_level, dst_irb->mt_layer,
+ srcX0, srcY0, srcX1, srcY1,
+ dstX0, dstY0, dstX1, dstY1,
+ filter, mirror_x, mirror_y);
+
+ intel_renderbuffer_set_needs_downsample(dst_irb);
+}
+
static bool
-try_blorp_blit(struct intel_context *intel,
- GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
- GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
+color_formats_match(gl_format src_format, gl_format dst_format)
+{
+ gl_format linear_src_format = _mesa_get_srgb_format_linear(src_format);
+ gl_format linear_dst_format = _mesa_get_srgb_format_linear(dst_format);
+
+ /* Normally, we require the formats to be equal. However, we also support
+ * blitting from ARGB to XRGB (discarding alpha), and from XRGB to ARGB
+ * (overriding alpha to 1.0 via blending).
+ */
+ return linear_src_format == linear_dst_format ||
+ (linear_src_format == MESA_FORMAT_XRGB8888 &&
+ linear_dst_format == MESA_FORMAT_ARGB8888) ||
+ (linear_src_format == MESA_FORMAT_ARGB8888 &&
+ linear_dst_format == MESA_FORMAT_XRGB8888);
+}
+
+static bool
+formats_match(GLbitfield buffer_bit, struct intel_renderbuffer *src_irb,
+ struct intel_renderbuffer *dst_irb)
+{
+ /* Note: don't just check gl_renderbuffer::Format, because in some cases
+ * multiple gl_formats resolve to the same native type in the miptree (for
+ * example MESA_FORMAT_X8_Z24 and MESA_FORMAT_S8_Z24), and we can blit
+ * between those formats.
+ */
+ gl_format src_format = find_miptree(buffer_bit, src_irb)->format;
+ gl_format dst_format = find_miptree(buffer_bit, dst_irb)->format;
+
+ return color_formats_match(src_format, dst_format);
+}
+
+static bool
+try_blorp_blit(struct brw_context *brw,
+ GLfloat srcX0, GLfloat srcY0, GLfloat srcX1, GLfloat srcY1,
+ GLfloat dstX0, GLfloat dstY0, GLfloat dstX1, GLfloat dstY1,
GLenum filter, GLbitfield buffer_bit)
{
- struct gl_context *ctx = &intel->ctx;
+ struct gl_context *ctx = &brw->ctx;
/* Sync up the state of window system buffers. We need to do this before
* we go looking for the buffers.
*/
- intel_prepare_render(intel);
+ intel_prepare_render(brw);
- /* Find buffers */
const struct gl_framebuffer *read_fb = ctx->ReadBuffer;
const struct gl_framebuffer *draw_fb = ctx->DrawBuffer;
- struct gl_renderbuffer *src_rb;
- struct gl_renderbuffer *dst_rb;
+
+ /* Detect if the blit needs to be mirrored */
+ bool mirror_x = false, mirror_y = false;
+ fixup_mirroring(mirror_x, srcX0, srcX1);
+ fixup_mirroring(mirror_x, dstX0, dstX1);
+ fixup_mirroring(mirror_y, srcY0, srcY1);
+ fixup_mirroring(mirror_y, dstY0, dstY1);
+
+ /* If the destination rectangle needs to be clipped or scissored, do so.
+ */
+ if (!(clip_or_scissor(mirror_x, srcX0, srcX1, dstX0, dstX1,
+ draw_fb->_Xmin, draw_fb->_Xmax) &&
+ clip_or_scissor(mirror_y, srcY0, srcY1, dstY0, dstY1,
+ draw_fb->_Ymin, draw_fb->_Ymax))) {
+ /* Everything got clipped/scissored away, so the blit was successful. */
+ return true;
+ }
+
+ /* If the source rectangle needs to be clipped or scissored, do so. */
+ if (!(clip_or_scissor(mirror_x, dstX0, dstX1, srcX0, srcX1,
+ 0, read_fb->Width) &&
+ clip_or_scissor(mirror_y, dstY0, dstY1, srcY0, srcY1,
+ 0, read_fb->Height))) {
+ /* Everything got clipped/scissored away, so the blit was successful. */
+ return true;
+ }
+
+ /* Account for the fact that in the system framebuffer, the origin is at
+ * the lower left.
+ */
+ if (_mesa_is_winsys_fbo(read_fb)) {
+ GLint tmp = read_fb->Height - srcY0;
+ srcY0 = read_fb->Height - srcY1;
+ srcY1 = tmp;
+ mirror_y = !mirror_y;
+ }
+ if (_mesa_is_winsys_fbo(draw_fb)) {
+ GLint tmp = draw_fb->Height - dstY0;
+ dstY0 = draw_fb->Height - dstY1;
+ dstY1 = tmp;
+ mirror_y = !mirror_y;
+ }
+
+ /* Find buffers */
+ struct intel_renderbuffer *src_irb;
+ struct intel_renderbuffer *dst_irb;
switch (buffer_bit) {
case GL_COLOR_BUFFER_BIT:
- src_rb = read_fb->_ColorReadBuffer;
- dst_rb =
- draw_fb->Attachment[
- draw_fb->_ColorDrawBufferIndexes[0]].Renderbuffer;
+ src_irb = intel_renderbuffer(read_fb->_ColorReadBuffer);
+ for (unsigned i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; ++i) {
+ dst_irb = intel_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[i]);
+ if (dst_irb && !formats_match(buffer_bit, src_irb, dst_irb))
+ return false;
+ }
+ for (unsigned i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; ++i) {
+ dst_irb = intel_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[i]);
+ if (dst_irb)
+ do_blorp_blit(brw, buffer_bit, src_irb, dst_irb, srcX0, srcY0,
+ srcX1, srcY1, dstX0, dstY0, dstX1, dstY1,
+ filter, mirror_x, mirror_y);
+ }
break;
case GL_DEPTH_BUFFER_BIT:
- src_rb = read_fb->Attachment[BUFFER_DEPTH].Renderbuffer;
- dst_rb = draw_fb->Attachment[BUFFER_DEPTH].Renderbuffer;
+ src_irb =
+ intel_renderbuffer(read_fb->Attachment[BUFFER_DEPTH].Renderbuffer);
+ dst_irb =
+ intel_renderbuffer(draw_fb->Attachment[BUFFER_DEPTH].Renderbuffer);
+ if (!formats_match(buffer_bit, src_irb, dst_irb))
+ return false;
+ do_blorp_blit(brw, buffer_bit, src_irb, dst_irb, srcX0, srcY0,
+ srcX1, srcY1, dstX0, dstY0, dstX1, dstY1,
+ filter, mirror_x, mirror_y);
break;
case GL_STENCIL_BUFFER_BIT:
- src_rb = read_fb->Attachment[BUFFER_STENCIL].Renderbuffer;
- dst_rb = draw_fb->Attachment[BUFFER_STENCIL].Renderbuffer;
+ src_irb =
+ intel_renderbuffer(read_fb->Attachment[BUFFER_STENCIL].Renderbuffer);
+ dst_irb =
+ intel_renderbuffer(draw_fb->Attachment[BUFFER_STENCIL].Renderbuffer);
+ if (!formats_match(buffer_bit, src_irb, dst_irb))
+ return false;
+ do_blorp_blit(brw, buffer_bit, src_irb, dst_irb, srcX0, srcY0,
+ srcX1, srcY1, dstX0, dstY0, dstX1, dstY1,
+ filter, mirror_x, mirror_y);
break;
default:
assert(false);
}
- /* Validate source */
- if (!src_rb) return false;
+ return true;
+}
+
+bool
+brw_blorp_copytexsubimage(struct brw_context *brw,
+ struct gl_renderbuffer *src_rb,
+ struct gl_texture_image *dst_image,
+ int slice,
+ int srcX0, int srcY0,
+ int dstX0, int dstY0,
+ int width, int height)
+{
+ struct gl_context *ctx = &brw->ctx;
struct intel_renderbuffer *src_irb = intel_renderbuffer(src_rb);
+ struct intel_texture_image *intel_image = intel_texture_image(dst_image);
+
+ /* Sync up the state of window system buffers. We need to do this before
+ * we go looking at the src renderbuffer's miptree.
+ */
+ intel_prepare_render(brw);
+
struct intel_mipmap_tree *src_mt = src_irb->mt;
- if (!src_mt) return false;
- if (buffer_bit == GL_STENCIL_BUFFER_BIT && src_mt->stencil_mt)
- src_mt = src_mt->stencil_mt;
- switch (src_mt->format) {
- case MESA_FORMAT_ARGB8888:
- case MESA_FORMAT_X8_Z24:
- case MESA_FORMAT_S8:
- break; /* Supported */
- default:
- /* Unsupported format.
- *
- * TODO: need to support all formats that are allowed as multisample
- * render targets.
- */
+ struct intel_mipmap_tree *dst_mt = intel_image->mt;
+
+ /* BLORP is not supported before Gen6. */
+ if (brw->gen < 6)
return false;
- }
- /* Validate destination */
- if (!dst_rb) return false;
- struct intel_renderbuffer *dst_irb = intel_renderbuffer(dst_rb);
- struct intel_mipmap_tree *dst_mt = dst_irb->mt;
- if (!dst_mt) return false;
- if (buffer_bit == GL_STENCIL_BUFFER_BIT && dst_mt->stencil_mt)
- dst_mt = dst_mt->stencil_mt;
- switch (dst_mt->format) {
- case MESA_FORMAT_ARGB8888:
- case MESA_FORMAT_X8_Z24:
- case MESA_FORMAT_S8:
- break; /* Supported */
- default:
- /* Unsupported format.
- *
- * TODO: need to support all formats that are allowed as multisample
- * render targets.
- */
+ if (!color_formats_match(src_mt->format, dst_mt->format)) {
return false;
}
- /* Account for the fact that in the system framebuffer, the origin is at
- * the lower left.
+ /* Source clipping shouldn't be necessary, since copytexsubimage (in
+ * src/mesa/main/teximage.c) calls _mesa_clip_copytexsubimage() which
+ * takes care of it.
+ *
+ * Destination clipping shouldn't be necessary since the restrictions on
+ * glCopyTexSubImage prevent the user from specifying a destination rectangle
+ * that falls outside the bounds of the destination texture.
+ * See error_check_subtexture_dimensions().
*/
- if (read_fb->Name == 0) {
- srcY0 = read_fb->Height - srcY0;
- srcY1 = read_fb->Height - srcY1;
- }
- if (draw_fb->Name == 0) {
- dstY0 = draw_fb->Height - dstY0;
- dstY1 = draw_fb->Height - dstY1;
- }
- /* Detect if the blit needs to be mirrored */
- bool mirror_x = false, mirror_y = false;
- fixup_mirroring(mirror_x, srcX0, srcX1);
- fixup_mirroring(mirror_x, dstX0, dstX1);
- fixup_mirroring(mirror_y, srcY0, srcY1);
- fixup_mirroring(mirror_y, dstY0, dstY1);
+ int srcY1 = srcY0 + height;
+ int srcX1 = srcX0 + width;
+ int dstX1 = dstX0 + width;
+ int dstY1 = dstY0 + height;
- /* Make sure width and height match */
- GLsizei width = srcX1 - srcX0;
- GLsizei height = srcY1 - srcY0;
- if (width != dstX1 - dstX0) return false;
- if (height != dstY1 - dstY0) return false;
-
- /* If the destination rectangle needs to be clipped or scissored, do so.
+ /* Account for the fact that in the system framebuffer, the origin is at
+ * the lower left.
*/
- if (!(clip_or_scissor(mirror_x, srcX0, srcX1, dstX0, dstX1,
- draw_fb->_Xmin, draw_fb->_Xmax) &&
- clip_or_scissor(mirror_y, srcY0, srcY1, dstY0, dstY1,
- draw_fb->_Ymin, draw_fb->_Ymax))) {
- /* Everything got clipped/scissored away, so the blit was successful. */
- return true;
+ bool mirror_y = false;
+ if (_mesa_is_winsys_fbo(ctx->ReadBuffer)) {
+ GLint tmp = src_rb->Height - srcY0;
+ srcY0 = src_rb->Height - srcY1;
+ srcY1 = tmp;
+ mirror_y = true;
}
- /* TODO: Clipping the source rectangle is not yet implemented. */
- if (srcX0 < 0 || (GLuint) srcX1 > read_fb->Width) return false;
- if (srcY0 < 0 || (GLuint) srcY1 > read_fb->Height) return false;
+ brw_blorp_blit_miptrees(brw,
+ src_mt, src_irb->mt_level, src_irb->mt_layer,
+ dst_mt, dst_image->Level, dst_image->Face + slice,
+ srcX0, srcY0, srcX1, srcY1,
+ dstX0, dstY0, dstX1, dstY1,
+ GL_NEAREST, false, mirror_y);
- /* Get ready to blit. This includes depth resolving the src and dst
- * buffers if necessary.
+ /* If we're copying to a packed depth stencil texture and the source
+ * framebuffer has separate stencil, we need to also copy the stencil data
+ * over.
*/
- intel_renderbuffer_resolve_depth(intel, src_irb);
- intel_renderbuffer_resolve_depth(intel, dst_irb);
-
- /* Do the blit */
- brw_blorp_blit_params params(brw_context(ctx), src_mt, dst_mt,
- srcX0, srcY0, dstX0, dstY0, dstX1, dstY1,
- mirror_x, mirror_y);
- brw_blorp_exec(intel, ¶ms);
-
- /* Mark the dst buffer as needing a HiZ resolve if necessary. */
- intel_renderbuffer_set_needs_hiz_resolve(dst_irb);
+ src_rb = ctx->ReadBuffer->Attachment[BUFFER_STENCIL].Renderbuffer;
+ if (_mesa_get_format_bits(dst_image->TexFormat, GL_STENCIL_BITS) > 0 &&
+ src_rb != NULL) {
+ src_irb = intel_renderbuffer(src_rb);
+ src_mt = src_irb->mt;
+
+ if (src_mt->stencil_mt)
+ src_mt = src_mt->stencil_mt;
+ if (dst_mt->stencil_mt)
+ dst_mt = dst_mt->stencil_mt;
+
+ if (src_mt != dst_mt) {
+ brw_blorp_blit_miptrees(brw,
+ src_mt, src_irb->mt_level, src_irb->mt_layer,
+ dst_mt, dst_image->Level,
+ dst_image->Face + slice,
+ srcX0, srcY0, srcX1, srcY1,
+ dstX0, dstY0, dstX1, dstY1,
+ GL_NEAREST, false, mirror_y);
+ }
+ }
return true;
}
+
GLbitfield
-brw_blorp_framebuffer(struct intel_context *intel,
+brw_blorp_framebuffer(struct brw_context *brw,
GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
GLbitfield mask, GLenum filter)
{
/* BLORP is not supported before Gen6. */
- if (intel->gen < 6)
+ if (brw->gen < 6)
return mask;
static GLbitfield buffer_bits[] = {
for (unsigned int i = 0; i < ARRAY_SIZE(buffer_bits); ++i) {
if ((mask & buffer_bits[i]) &&
- try_blorp_blit(intel,
+ try_blorp_blit(brw,
srcX0, srcY0, srcX1, srcY1,
dstX0, dstY0, dstX1, dstY1,
filter, buffer_bits[i])) {
SAMPLER_MESSAGE_ARG_U_INT,
SAMPLER_MESSAGE_ARG_V_INT,
SAMPLER_MESSAGE_ARG_SI_INT,
+ SAMPLER_MESSAGE_ARG_MCS_INT,
SAMPLER_MESSAGE_ARG_ZERO_INT,
};
* offset = tile(tiling_format, encode_msaa(num_samples, layout, X, Y, S))
* (X, Y, S) = decode_msaa(num_samples, layout, detile(tiling_format, offset))
*
- * For a single-sampled surface, or for a multisampled surface that stores
- * each sample in a different array slice, encode_msaa() and decode_msaa are
- * the identity function:
+ * For a single-sampled surface, or for a multisampled surface using
+ * INTEL_MSAA_LAYOUT_UMS, encode_msaa() and decode_msaa are the identity
+ * function:
*
- * encode_msaa(1, N/A, X, Y, 0) = (X, Y, 0)
- * decode_msaa(1, N/A, X, Y, 0) = (X, Y, 0)
- * encode_msaa(n, sliced, X, Y, S) = (X, Y, S)
- * decode_msaa(n, sliced, X, Y, S) = (X, Y, S)
+ * encode_msaa(1, NONE, X, Y, 0) = (X, Y, 0)
+ * decode_msaa(1, NONE, X, Y, 0) = (X, Y, 0)
+ * encode_msaa(n, UMS, X, Y, S) = (X, Y, S)
+ * decode_msaa(n, UMS, X, Y, S) = (X, Y, S)
*
- * For a 4x interleaved multisampled surface, encode_msaa() embeds the sample
- * number into bit 1 of the X and Y coordinates:
+ * For a 4x multisampled surface using INTEL_MSAA_LAYOUT_IMS, encode_msaa()
+ * embeds the sample number into bit 1 of the X and Y coordinates:
*
- * encode_msaa(4, interleaved, X, Y, S) = (X', Y', 0)
+ * encode_msaa(4, IMS, X, Y, S) = (X', Y', 0)
* where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1)
* Y' = (Y & ~0b1 ) << 1 | (S & 0b10) | (Y & 0b1)
- * decode_msaa(4, interleaved, X, Y, 0) = (X', Y', S)
+ * decode_msaa(4, IMS, X, Y, 0) = (X', Y', S)
* where X' = (X & ~0b11) >> 1 | (X & 0b1)
* Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
* S = (Y & 0b10) | (X & 0b10) >> 1
*
+ * For an 8x multisampled surface using INTEL_MSAA_LAYOUT_IMS, encode_msaa()
+ * embeds the sample number into bits 1 and 2 of the X coordinate and bit 1 of
+ * the Y coordinate:
+ *
+ * encode_msaa(8, IMS, X, Y, S) = (X', Y', 0)
+ * where X' = (X & ~0b1) << 2 | (S & 0b100) | (S & 0b1) << 1 | (X & 0b1)
+ * Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1)
+ * decode_msaa(8, IMS, X, Y, 0) = (X', Y', S)
+ * where X' = (X & ~0b111) >> 2 | (X & 0b1)
+ * Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
+ * S = (X & 0b100) | (Y & 0b10) | (X & 0b10) >> 1
+ *
* For X tiling, tile() combines together the low-order bits of the X and Y
* coordinates in the pattern 0byyyxxxxxxxxx, creating 4k tiles that are 512
* bytes wide and 8 rows high:
void alloc_push_const_regs(int base_reg);
void compute_frag_coords();
void translate_tiling(bool old_tiled_w, bool new_tiled_w);
- void encode_msaa(unsigned num_samples, bool interleaved);
- void decode_msaa(unsigned num_samples, bool interleaved);
+ void encode_msaa(unsigned num_samples, intel_msaa_layout layout);
+ void decode_msaa(unsigned num_samples, intel_msaa_layout layout);
void kill_if_outside_dst_rect();
void translate_dst_to_src();
+ void clamp_tex_coords(struct brw_reg regX, struct brw_reg regY,
+ struct brw_reg clampX0, struct brw_reg clampY0,
+ struct brw_reg clampX1, struct brw_reg clampY1);
void single_to_blend();
- void manual_blend();
+ void manual_blend_average(unsigned num_samples);
+ void manual_blend_bilinear(unsigned num_samples);
void sample(struct brw_reg dst);
void texel_fetch(struct brw_reg dst);
- void expand_to_32_bits(struct brw_reg src, struct brw_reg dst);
+ void mcs_fetch();
void texture_lookup(struct brw_reg dst, GLuint msg_type,
const sampler_message_arg *args, int num_args);
void render_target_write();
+ /**
+ * Base-2 logarithm of the maximum number of samples that can be blended.
+ */
+ static const unsigned LOG2_MAX_BLEND_SAMPLES = 3;
+
void *mem_ctx;
struct brw_context *brw;
const brw_blorp_blit_prog_key *key;
struct brw_reg dst_x1;
struct brw_reg dst_y0;
struct brw_reg dst_y1;
+ /* Top right coordinates of the rectangular grid used for scaled blitting */
+ struct brw_reg rect_grid_x1;
+ struct brw_reg rect_grid_y1;
struct {
struct brw_reg multiplier;
struct brw_reg offset;
} x_transform, y_transform;
- /* Data to be written to render target (4 vec16's) */
- struct brw_reg result;
+ /* Data read from texture (4 vec16's per array element) */
+ struct brw_reg texture_data[LOG2_MAX_BLEND_SAMPLES + 1];
- /* Auxiliary storage for data returned by a sampling operation when
- * blending (4 vec16's)
+ /* Auxiliary storage for the contents of the MCS surface.
+ *
+ * Since the sampler always returns 8 registers worth of data, this is 8
+ * registers wide, even though we only use the first 2 registers of it.
*/
- struct brw_reg texture_data;
+ struct brw_reg mcs_data;
/* X coordinates. We have two of them so that we can perform coordinate
* transformations easily.
*/
struct brw_reg y_coords[2];
+ /* X, Y coordinates of the pixel from which we need to fetch the specific
+ * sample. These are used for multisample scaled blitting.
+ */
+ struct brw_reg x_sample_coords;
+ struct brw_reg y_sample_coords;
+
+ /* Fractional parts of the x and y coordinates, used as bilinear interpolation coefficients */
+ struct brw_reg x_frac;
+ struct brw_reg y_frac;
+
/* Which element of x_coords and y_coords is currently in use.
*/
int xy_coord_index;
brw_blorp_blit_program::compile(struct brw_context *brw,
GLuint *program_size)
{
- /* Since blorp uses color textures and render targets to do all its work
- * (even when blitting stencil and depth data), we always have to configure
- * the Gen7 GPU to use sliced layout on Gen7. On Gen6, the MSAA layout is
- * always interleaved.
- */
- const bool rt_interleaved = key->rt_samples > 0 && brw->intel.gen == 6;
- const bool tex_interleaved = key->tex_samples > 0 && brw->intel.gen == 6;
-
/* Sanity checks */
if (key->dst_tiled_w && key->rt_samples > 0) {
/* If the destination image is W tiled and multisampled, then the thread
*/
assert(!key->src_tiled_w);
assert(key->tex_samples == key->src_samples);
- assert(tex_interleaved == key->src_interleaved);
+ assert(key->tex_layout == key->src_layout);
assert(key->tex_samples > 0);
}
assert(key->rt_samples > 0);
}
- /* Interleaved only makes sense on MSAA surfaces */
- if (tex_interleaved) assert(key->tex_samples > 0);
- if (key->src_interleaved) assert(key->src_samples > 0);
- if (key->dst_interleaved) assert(key->dst_samples > 0);
+ /* Make sure layout is consistent with sample count */
+ assert((key->tex_layout == INTEL_MSAA_LAYOUT_NONE) ==
+ (key->tex_samples == 0));
+ assert((key->rt_layout == INTEL_MSAA_LAYOUT_NONE) ==
+ (key->rt_samples == 0));
+ assert((key->src_layout == INTEL_MSAA_LAYOUT_NONE) ==
+ (key->src_samples == 0));
+ assert((key->dst_layout == INTEL_MSAA_LAYOUT_NONE) ==
+ (key->dst_samples == 0));
/* Set up prog_data */
memset(&prog_data, 0, sizeof(prog_data));
*/
if (rt_tiled_w != key->dst_tiled_w ||
key->rt_samples != key->dst_samples ||
- rt_interleaved != key->dst_interleaved) {
- encode_msaa(key->rt_samples, rt_interleaved);
+ key->rt_layout != key->dst_layout) {
+ encode_msaa(key->rt_samples, key->rt_layout);
/* Now (X, Y, S) = detile(rt_tiling, offset) */
translate_tiling(rt_tiled_w, key->dst_tiled_w);
/* Now (X, Y, S) = detile(dst_tiling, offset) */
- decode_msaa(key->dst_samples, key->dst_interleaved);
+ decode_msaa(key->dst_samples, key->dst_layout);
}
/* Now (X, Y, S) = decode_msaa(dst_samples, detile(dst_tiling, offset)).
* that we want to texture from. Exception: if we are blending, then S is
* irrelevant, because we are going to fetch all samples.
*/
- if (key->blend) {
- if (brw->intel.gen == 6) {
+ if (key->blend && !key->blit_scaled) {
+ if (brw->gen == 6) {
/* Gen6 hardware an automatically blend using the SAMPLE message */
single_to_blend();
- sample(result);
+ sample(texture_data[0]);
} else {
/* Gen7+ hardware doesn't automaticaly blend. */
- manual_blend();
+ manual_blend_average(key->src_samples);
}
+ } else if(key->blend && key->blit_scaled) {
+ manual_blend_bilinear(key->src_samples);
} else {
/* We aren't blending, which means we just want to fetch a single sample
* from the source surface. The address that we want to fetch from is
* the same as the configuration of the texture, then we need to adjust
* the coordinates to compensate for the difference.
*/
- if (tex_tiled_w != key->src_tiled_w ||
- key->tex_samples != key->src_samples ||
- tex_interleaved != key->src_interleaved) {
- encode_msaa(key->src_samples, key->src_interleaved);
+ if ((tex_tiled_w != key->src_tiled_w ||
+ key->tex_samples != key->src_samples ||
+ key->tex_layout != key->src_layout) &&
+ !key->bilinear_filter) {
+ encode_msaa(key->src_samples, key->src_layout);
/* Now (X, Y, S) = detile(src_tiling, offset) */
translate_tiling(key->src_tiled_w, tex_tiled_w);
/* Now (X, Y, S) = detile(tex_tiling, offset) */
- decode_msaa(key->tex_samples, tex_interleaved);
+ decode_msaa(key->tex_samples, key->tex_layout);
}
- /* Now (X, Y, S) = decode_msaa(tex_samples, detile(tex_tiling, offset)).
- *
- * In other words: X, Y, and S now contain values which, when passed to
- * the texturing unit, will cause data to be read from the correct
- * memory location. So we can fetch the texel now.
- */
- texel_fetch(result);
+ if (key->bilinear_filter) {
+ sample(texture_data[0]);
+ }
+ else {
+ /* Now (X, Y, S) = decode_msaa(tex_samples, detile(tex_tiling, offset)).
+ *
+ * In other words: X, Y, and S now contain values which, when passed to
+ * the texturing unit, will cause data to be read from the correct
+ * memory location. So we can fetch the texel now.
+ */
+ if (key->tex_layout == INTEL_MSAA_LAYOUT_CMS)
+ mcs_fetch();
+ texel_fetch(texture_data[0]);
+ }
}
/* Finally, write the fetched (or blended) value to the render target and
* terminate the thread.
*/
render_target_write();
+
+ if (unlikely(INTEL_DEBUG & DEBUG_BLORP)) {
+ printf("Native code for BLORP blit:\n");
+ brw_dump_compile(&func, stdout, 0, func.next_insn_offset);
+ printf("\n");
+ }
return brw_get_program(&func, program_size);
}
brw_blorp_blit_program::alloc_push_const_regs(int base_reg)
{
#define CONST_LOC(name) offsetof(brw_blorp_wm_push_constants, name)
-#define ALLOC_REG(name) \
- this->name = \
- brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, base_reg, CONST_LOC(name) / 2)
-
- ALLOC_REG(dst_x0);
- ALLOC_REG(dst_x1);
- ALLOC_REG(dst_y0);
- ALLOC_REG(dst_y1);
- ALLOC_REG(x_transform.multiplier);
- ALLOC_REG(x_transform.offset);
- ALLOC_REG(y_transform.multiplier);
- ALLOC_REG(y_transform.offset);
+#define ALLOC_REG(name, type) \
+ this->name = \
+ retype(brw_vec1_reg(BRW_GENERAL_REGISTER_FILE, \
+ base_reg + CONST_LOC(name) / 32, \
+ (CONST_LOC(name) % 32) / 4), type)
+
+ ALLOC_REG(dst_x0, BRW_REGISTER_TYPE_UD);
+ ALLOC_REG(dst_x1, BRW_REGISTER_TYPE_UD);
+ ALLOC_REG(dst_y0, BRW_REGISTER_TYPE_UD);
+ ALLOC_REG(dst_y1, BRW_REGISTER_TYPE_UD);
+ ALLOC_REG(rect_grid_x1, BRW_REGISTER_TYPE_F);
+ ALLOC_REG(rect_grid_y1, BRW_REGISTER_TYPE_F);
+ ALLOC_REG(x_transform.multiplier, BRW_REGISTER_TYPE_F);
+ ALLOC_REG(x_transform.offset, BRW_REGISTER_TYPE_F);
+ ALLOC_REG(y_transform.multiplier, BRW_REGISTER_TYPE_F);
+ ALLOC_REG(y_transform.offset, BRW_REGISTER_TYPE_F);
#undef CONST_LOC
#undef ALLOC_REG
}
prog_data.first_curbe_grf = reg;
alloc_push_const_regs(reg);
reg += BRW_BLORP_NUM_PUSH_CONST_REGS;
- this->result = vec16(brw_vec8_grf(reg, 0)); reg += 8;
- this->texture_data = vec16(brw_vec8_grf(reg, 0)); reg += 8;
+ for (unsigned i = 0; i < ARRAY_SIZE(texture_data); ++i) {
+ this->texture_data[i] =
+ retype(vec16(brw_vec8_grf(reg, 0)), key->texture_data_type);
+ reg += 8;
+ }
+ this->mcs_data =
+ retype(brw_vec8_grf(reg, 0), BRW_REGISTER_TYPE_UD); reg += 8;
+
for (int i = 0; i < 2; ++i) {
this->x_coords[i]
- = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW));
+ = retype(brw_vec8_grf(reg, 0), BRW_REGISTER_TYPE_UD);
+ reg += 2;
this->y_coords[i]
- = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW));
+ = retype(brw_vec8_grf(reg, 0), BRW_REGISTER_TYPE_UD);
+ reg += 2;
}
+
+ if (key->blit_scaled && key->blend) {
+ this->x_sample_coords = brw_vec8_grf(reg, 0);
+ reg += 2;
+ this->y_sample_coords = brw_vec8_grf(reg, 0);
+ reg += 2;
+ this->x_frac = brw_vec8_grf(reg, 0);
+ reg += 2;
+ this->y_frac = brw_vec8_grf(reg, 0);
+ reg += 2;
+ }
+
this->xy_coord_index = 0;
this->sample_index
- = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW));
- this->t1 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW));
- this->t2 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW));
+ = retype(brw_vec8_grf(reg, 0), BRW_REGISTER_TYPE_UD);
+ reg += 2;
+ this->t1 = retype(brw_vec8_grf(reg, 0), BRW_REGISTER_TYPE_UD);
+ reg += 2;
+ this->t2 = retype(brw_vec8_grf(reg, 0), BRW_REGISTER_TYPE_UD);
+ reg += 2;
+
+ /* Make sure we didn't run out of registers */
+ assert(reg <= GEN7_MRF_HACK_START);
int mrf = 2;
this->base_mrf = mrf;
* Then, we need to add the repeating sequence (0, 1, 0, 1, ...) to the
* result, since pixels n+1 and n+3 are in the right half of the subspan.
*/
- brw_ADD(&func, X, stride(suboffset(R1, 4), 2, 4, 0), brw_imm_v(0x10101010));
+ brw_ADD(&func, vec16(retype(X, BRW_REGISTER_TYPE_UW)),
+ stride(suboffset(R1, 4), 2, 4, 0), brw_imm_v(0x10101010));
/* Similarly, Y coordinates for subspans come from R1.2[31:16] through
* R1.5[31:16], so to get pixel Y coordinates we need to start at the 5th
* And we need to add the repeating sequence (0, 0, 1, 1, ...), since
* pixels n+2 and n+3 are in the bottom half of the subspan.
*/
- brw_ADD(&func, Y, stride(suboffset(R1, 5), 2, 4, 0), brw_imm_v(0x11001100));
+ brw_ADD(&func, vec16(retype(Y, BRW_REGISTER_TYPE_UW)),
+ stride(suboffset(R1, 5), 2, 4, 0), brw_imm_v(0x11001100));
+
+ /* Move the coordinates to UD registers. */
+ brw_MOV(&func, vec16(Xp), retype(X, BRW_REGISTER_TYPE_UW));
+ brw_MOV(&func, vec16(Yp), retype(Y, BRW_REGISTER_TYPE_UW));
+ SWAP_XY_AND_XPYP();
if (key->persample_msaa_dispatch) {
- /* The WM will be run in MSDISPMODE_PERSAMPLE with num_samples > 0.
- * Therefore, subspan 0 will represent sample 0, subspan 1 will
- * represent sample 1, and so on.
- *
- * So we need to populate S with the sequence (0, 0, 0, 0, 1, 1, 1, 1,
- * 2, 2, 2, 2, 3, 3, 3, 3). The easiest way to do this is to populate a
- * temporary variable with the sequence (0, 1, 2, 3), and then copy from
- * it using vstride=1, width=4, hstride=0.
- *
- * TODO: implement the necessary calculation for 8x multisampling.
- */
- brw_MOV(&func, t1, brw_imm_v(0x3210));
- brw_MOV(&func, S, stride(t1, 1, 4, 0));
+ switch (key->rt_samples) {
+ case 4: {
+ /* The WM will be run in MSDISPMODE_PERSAMPLE with num_samples == 4.
+ * Therefore, subspan 0 will represent sample 0, subspan 1 will
+ * represent sample 1, and so on.
+ *
+ * So we need to populate S with the sequence (0, 0, 0, 0, 1, 1, 1,
+ * 1, 2, 2, 2, 2, 3, 3, 3, 3). The easiest way to do this is to
+ * populate a temporary variable with the sequence (0, 1, 2, 3), and
+ * then copy from it using vstride=1, width=4, hstride=0.
+ */
+ struct brw_reg t1_uw1 = retype(t1, BRW_REGISTER_TYPE_UW);
+ brw_MOV(&func, vec16(t1_uw1), brw_imm_v(0x3210));
+ /* Move to UD sample_index register. */
+ brw_MOV(&func, S, stride(t1_uw1, 1, 4, 0));
+ brw_MOV(&func, offset(S, 1), suboffset(stride(t1_uw1, 1, 4, 0), 2));
+ break;
+ }
+ case 8: {
+ /* The WM will be run in MSDISPMODE_PERSAMPLE with num_samples == 8.
+ * Therefore, subspan 0 will represent sample N (where N is 0 or 4),
+ * subspan 1 will represent sample 1, and so on. We can find the
+ * value of N by looking at R0.0 bits 7:6 ("Starting Sample Pair
+ * Index") and multiplying by two (since samples are always delivered
+ * in pairs). That is, we compute 2*((R0.0 & 0xc0) >> 6) == (R0.0 &
+ * 0xc0) >> 5.
+ *
+ * Then we need to add N to the sequence (0, 0, 0, 0, 1, 1, 1, 1, 2,
+ * 2, 2, 2, 3, 3, 3, 3), which we compute by populating a temporary
+ * variable with the sequence (0, 1, 2, 3), and then reading from it
+ * using vstride=1, width=4, hstride=0.
+ */
+ struct brw_reg t1_ud1 = vec1(retype(t1, BRW_REGISTER_TYPE_UD));
+ struct brw_reg t2_uw1 = retype(t2, BRW_REGISTER_TYPE_UW);
+ struct brw_reg r0_ud1 = vec1(retype(R0, BRW_REGISTER_TYPE_UD));
+ brw_AND(&func, t1_ud1, r0_ud1, brw_imm_ud(0xc0));
+ brw_SHR(&func, t1_ud1, t1_ud1, brw_imm_ud(5));
+ brw_MOV(&func, vec16(t2_uw1), brw_imm_v(0x3210));
+ brw_ADD(&func, vec16(S), retype(t1_ud1, BRW_REGISTER_TYPE_UW),
+ stride(t2_uw1, 1, 4, 0));
+ brw_ADD(&func, offset(S, 1),
+ retype(t1_ud1, BRW_REGISTER_TYPE_UW),
+ suboffset(stride(t2_uw1, 1, 4, 0), 2));
+ break;
+ }
+ default:
+ assert(!"Unrecognized sample count in "
+ "brw_blorp_blit_program::compute_frag_coords()");
+ break;
+ }
s_is_zero = false;
} else {
/* Either the destination surface is single-sampled, or the WM will be
return;
/* In the code that follows, we can safely assume that S = 0, because W
- * tiling formats always use interleaved encoding.
+ * tiling formats always use IMS layout.
*/
assert(s_is_zero);
+ brw_set_compression_control(&func, BRW_COMPRESSION_COMPRESSED);
if (new_tiled_w) {
/* Given X and Y coordinates that describe an address using Y tiling,
* translate to the X and Y coordinates that describe the same address
brw_OR(&func, Yp, t1, t2);
SWAP_XY_AND_XPYP();
}
+ brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
}
/**
*
* This code modifies the X and Y coordinates according to the formula:
*
- * (X', Y', S') = encode_msaa_4x(X, Y, S)
+ * (X', Y', S') = encode_msaa(num_samples, IMS, X, Y, S)
*
* (See brw_blorp_blit_program).
*/
void
-brw_blorp_blit_program::encode_msaa(unsigned num_samples, bool interleaved)
+brw_blorp_blit_program::encode_msaa(unsigned num_samples,
+ intel_msaa_layout layout)
{
- if (num_samples == 0) {
+ brw_set_compression_control(&func, BRW_COMPRESSION_COMPRESSED);
+ switch (layout) {
+ case INTEL_MSAA_LAYOUT_NONE:
/* No translation necessary, and S should already be zero. */
assert(s_is_zero);
- } else if (!interleaved) {
- /* No translation necessary. */
- } else {
- /* encode_msaa(4, interleaved, X, Y, S) = (X', Y', 0)
- * where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1)
- * Y' = (Y & ~0b1 ) << 1 | (S & 0b10) | (Y & 0b1)
+ break;
+ case INTEL_MSAA_LAYOUT_CMS:
+ /* We can't compensate for compressed layout since at this point in the
+ * program we haven't read from the MCS buffer.
*/
- brw_AND(&func, t1, X, brw_imm_uw(0xfffe)); /* X & ~0b1 */
- if (!s_is_zero) {
- brw_AND(&func, t2, S, brw_imm_uw(1)); /* S & 0b1 */
- brw_OR(&func, t1, t1, t2); /* (X & ~0b1) | (S & 0b1) */
- }
- brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b1) << 1
- | (S & 0b1) << 1 */
- brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */
- brw_OR(&func, Xp, t1, t2);
- brw_AND(&func, t1, Y, brw_imm_uw(0xfffe)); /* Y & ~0b1 */
- brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b1) << 1 */
- if (!s_is_zero) {
- brw_AND(&func, t2, S, brw_imm_uw(2)); /* S & 0b10 */
- brw_OR(&func, t1, t1, t2); /* (Y & ~0b1) << 1 | (S & 0b10) */
+ assert(!"Bad layout in encode_msaa");
+ break;
+ case INTEL_MSAA_LAYOUT_UMS:
+ /* No translation necessary. */
+ break;
+ case INTEL_MSAA_LAYOUT_IMS:
+ switch (num_samples) {
+ case 4:
+ /* encode_msaa(4, IMS, X, Y, S) = (X', Y', 0)
+ * where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1)
+ * Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1)
+ */
+ brw_AND(&func, t1, X, brw_imm_uw(0xfffe)); /* X & ~0b1 */
+ if (!s_is_zero) {
+ brw_AND(&func, t2, S, brw_imm_uw(1)); /* S & 0b1 */
+ brw_OR(&func, t1, t1, t2); /* (X & ~0b1) | (S & 0b1) */
+ }
+ brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b1) << 1
+ | (S & 0b1) << 1 */
+ brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */
+ brw_OR(&func, Xp, t1, t2);
+ brw_AND(&func, t1, Y, brw_imm_uw(0xfffe)); /* Y & ~0b1 */
+ brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b1) << 1 */
+ if (!s_is_zero) {
+ brw_AND(&func, t2, S, brw_imm_uw(2)); /* S & 0b10 */
+ brw_OR(&func, t1, t1, t2); /* (Y & ~0b1) << 1 | (S & 0b10) */
+ }
+ brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */
+ brw_OR(&func, Yp, t1, t2);
+ break;
+ case 8:
+ /* encode_msaa(8, IMS, X, Y, S) = (X', Y', 0)
+ * where X' = (X & ~0b1) << 2 | (S & 0b100) | (S & 0b1) << 1
+ * | (X & 0b1)
+ * Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1)
+ */
+ brw_AND(&func, t1, X, brw_imm_uw(0xfffe)); /* X & ~0b1 */
+ brw_SHL(&func, t1, t1, brw_imm_uw(2)); /* (X & ~0b1) << 2 */
+ if (!s_is_zero) {
+ brw_AND(&func, t2, S, brw_imm_uw(4)); /* S & 0b100 */
+ brw_OR(&func, t1, t1, t2); /* (X & ~0b1) << 2 | (S & 0b100) */
+ brw_AND(&func, t2, S, brw_imm_uw(1)); /* S & 0b1 */
+ brw_SHL(&func, t2, t2, brw_imm_uw(1)); /* (S & 0b1) << 1 */
+ brw_OR(&func, t1, t1, t2); /* (X & ~0b1) << 2 | (S & 0b100)
+ | (S & 0b1) << 1 */
+ }
+ brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */
+ brw_OR(&func, Xp, t1, t2);
+ brw_AND(&func, t1, Y, brw_imm_uw(0xfffe)); /* Y & ~0b1 */
+ brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b1) << 1 */
+ if (!s_is_zero) {
+ brw_AND(&func, t2, S, brw_imm_uw(2)); /* S & 0b10 */
+ brw_OR(&func, t1, t1, t2); /* (Y & ~0b1) << 1 | (S & 0b10) */
+ }
+ brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */
+ brw_OR(&func, Yp, t1, t2);
+ break;
}
- brw_AND(&func, t2, Y, brw_imm_uw(1));
- brw_OR(&func, Yp, t1, t2);
SWAP_XY_AND_XPYP();
s_is_zero = true;
+ break;
}
+ brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
}
/**
*
* This code modifies the X and Y coordinates according to the formula:
*
- * (X', Y', S) = decode_msaa(num_samples, X, Y, S)
+ * (X', Y', S) = decode_msaa(num_samples, IMS, X, Y, S)
*
* (See brw_blorp_blit_program).
*/
void
-brw_blorp_blit_program::decode_msaa(unsigned num_samples, bool interleaved)
+brw_blorp_blit_program::decode_msaa(unsigned num_samples,
+ intel_msaa_layout layout)
{
- if (num_samples == 0) {
+ brw_set_compression_control(&func, BRW_COMPRESSION_COMPRESSED);
+ switch (layout) {
+ case INTEL_MSAA_LAYOUT_NONE:
/* No translation necessary, and S should already be zero. */
assert(s_is_zero);
- } else if (!interleaved) {
- /* No translation necessary. */
- } else {
- /* decode_msaa(4, interleaved, X, Y, 0) = (X', Y', S)
- * where X' = (X & ~0b11) >> 1 | (X & 0b1)
- * Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
- * S = (Y & 0b10) | (X & 0b10) >> 1
+ break;
+ case INTEL_MSAA_LAYOUT_CMS:
+ /* We can't compensate for compressed layout since at this point in the
+ * program we don't have access to the MCS buffer.
*/
+ assert(!"Bad layout in encode_msaa");
+ break;
+ case INTEL_MSAA_LAYOUT_UMS:
+ /* No translation necessary. */
+ break;
+ case INTEL_MSAA_LAYOUT_IMS:
assert(s_is_zero);
- brw_AND(&func, t1, X, brw_imm_uw(0xfffc)); /* X & ~0b11 */
- brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b11) >> 1 */
- brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */
- brw_OR(&func, Xp, t1, t2);
- brw_AND(&func, t1, Y, brw_imm_uw(0xfffc)); /* Y & ~0b11 */
- brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b11) >> 1 */
- brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */
- brw_OR(&func, Yp, t1, t2);
- brw_AND(&func, t1, Y, brw_imm_uw(2)); /* Y & 0b10 */
- brw_AND(&func, t2, X, brw_imm_uw(2)); /* X & 0b10 */
- brw_SHR(&func, t2, t2, brw_imm_uw(1)); /* (X & 0b10) >> 1 */
- brw_OR(&func, S, t1, t2);
+ switch (num_samples) {
+ case 4:
+ /* decode_msaa(4, IMS, X, Y, 0) = (X', Y', S)
+ * where X' = (X & ~0b11) >> 1 | (X & 0b1)
+ * Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
+ * S = (Y & 0b10) | (X & 0b10) >> 1
+ */
+ brw_AND(&func, t1, X, brw_imm_uw(0xfffc)); /* X & ~0b11 */
+ brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b11) >> 1 */
+ brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */
+ brw_OR(&func, Xp, t1, t2);
+ brw_AND(&func, t1, Y, brw_imm_uw(0xfffc)); /* Y & ~0b11 */
+ brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b11) >> 1 */
+ brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */
+ brw_OR(&func, Yp, t1, t2);
+ brw_AND(&func, t1, Y, brw_imm_uw(2)); /* Y & 0b10 */
+ brw_AND(&func, t2, X, brw_imm_uw(2)); /* X & 0b10 */
+ brw_SHR(&func, t2, t2, brw_imm_uw(1)); /* (X & 0b10) >> 1 */
+ brw_OR(&func, S, t1, t2);
+ break;
+ case 8:
+ /* decode_msaa(8, IMS, X, Y, 0) = (X', Y', S)
+ * where X' = (X & ~0b111) >> 2 | (X & 0b1)
+ * Y' = (Y & ~0b11) >> 1 | (Y & 0b1)
+ * S = (X & 0b100) | (Y & 0b10) | (X & 0b10) >> 1
+ */
+ brw_AND(&func, t1, X, brw_imm_uw(0xfff8)); /* X & ~0b111 */
+ brw_SHR(&func, t1, t1, brw_imm_uw(2)); /* (X & ~0b111) >> 2 */
+ brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */
+ brw_OR(&func, Xp, t1, t2);
+ brw_AND(&func, t1, Y, brw_imm_uw(0xfffc)); /* Y & ~0b11 */
+ brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b11) >> 1 */
+ brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */
+ brw_OR(&func, Yp, t1, t2);
+ brw_AND(&func, t1, X, brw_imm_uw(4)); /* X & 0b100 */
+ brw_AND(&func, t2, Y, brw_imm_uw(2)); /* Y & 0b10 */
+ brw_OR(&func, t1, t1, t2); /* (X & 0b100) | (Y & 0b10) */
+ brw_AND(&func, t2, X, brw_imm_uw(2)); /* X & 0b10 */
+ brw_SHR(&func, t2, t2, brw_imm_uw(1)); /* (X & 0b10) >> 1 */
+ brw_OR(&func, S, t1, t2);
+ break;
+ }
s_is_zero = false;
SWAP_XY_AND_XPYP();
+ break;
}
+ brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
}
/**
void
brw_blorp_blit_program::kill_if_outside_dst_rect()
{
- struct brw_reg f0 = brw_flag_reg();
+ struct brw_reg f0 = brw_flag_reg(0, 0);
struct brw_reg g1 = retype(brw_vec1_grf(1, 7), BRW_REGISTER_TYPE_UW);
- struct brw_reg null16 = vec16(retype(brw_null_reg(), BRW_REGISTER_TYPE_UW));
+ struct brw_reg null32 = vec16(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
- brw_CMP(&func, null16, BRW_CONDITIONAL_GE, X, dst_x0);
- brw_CMP(&func, null16, BRW_CONDITIONAL_GE, Y, dst_y0);
- brw_CMP(&func, null16, BRW_CONDITIONAL_L, X, dst_x1);
- brw_CMP(&func, null16, BRW_CONDITIONAL_L, Y, dst_y1);
+ brw_CMP(&func, null32, BRW_CONDITIONAL_GE, X, dst_x0);
+ brw_CMP(&func, null32, BRW_CONDITIONAL_GE, Y, dst_y0);
+ brw_CMP(&func, null32, BRW_CONDITIONAL_L, X, dst_x1);
+ brw_CMP(&func, null32, BRW_CONDITIONAL_L, Y, dst_y1);
brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
brw_push_insn_state(&func);
brw_pop_insn_state(&func);
}
+#define X_f retype(X, BRW_REGISTER_TYPE_F)
+#define Y_f retype(Y, BRW_REGISTER_TYPE_F)
+#define Xp_f retype(Xp, BRW_REGISTER_TYPE_F)
+#define Yp_f retype(Yp, BRW_REGISTER_TYPE_F)
/**
* Emit code to translate from destination (X, Y) coordinates to source (X, Y)
* coordinates.
void
brw_blorp_blit_program::translate_dst_to_src()
{
- brw_MUL(&func, Xp, X, x_transform.multiplier);
- brw_MUL(&func, Yp, Y, y_transform.multiplier);
- brw_ADD(&func, Xp, Xp, x_transform.offset);
- brw_ADD(&func, Yp, Yp, y_transform.offset);
- SWAP_XY_AND_XPYP();
+ brw_set_compression_control(&func, BRW_COMPRESSION_COMPRESSED);
+ /* Move the UD coordinates to float registers. */
+ brw_MOV(&func, Xp_f, X);
+ brw_MOV(&func, Yp_f, Y);
+ /* Scale and offset */
+ brw_MUL(&func, X_f, Xp_f, x_transform.multiplier);
+ brw_MUL(&func, Y_f, Yp_f, y_transform.multiplier);
+ brw_ADD(&func, X_f, X_f, x_transform.offset);
+ brw_ADD(&func, Y_f, Y_f, y_transform.offset);
+ if (key->blit_scaled && key->blend) {
+ /* Translate coordinates to lay out the samples in a rectangular grid
+ * roughly corresponding to sample locations.
+ */
+ brw_MUL(&func, X_f, X_f, brw_imm_f(key->x_scale));
+ brw_MUL(&func, Y_f, Y_f, brw_imm_f(key->y_scale));
+ /* Adjust coordinates so that integers represent pixel centers rather
+ * than pixel edges.
+ */
+ brw_ADD(&func, X_f, X_f, brw_imm_f(-0.5));
+ brw_ADD(&func, Y_f, Y_f, brw_imm_f(-0.5));
+
+ /* Clamp the X, Y texture coordinates to properly handle the sampling of
+ * texels on texture edges.
+ */
+ clamp_tex_coords(X_f, Y_f,
+ brw_imm_f(0.0), brw_imm_f(0.0),
+ rect_grid_x1, rect_grid_y1);
+
+ /* Store the fractional parts to be used as bilinear interpolation
+ * coefficients.
+ */
+ brw_FRC(&func, x_frac, X_f);
+ brw_FRC(&func, y_frac, Y_f);
+
+ /* Round the float coordinates down to nearest integer */
+ brw_RNDD(&func, Xp_f, X_f);
+ brw_RNDD(&func, Yp_f, Y_f);
+ brw_MUL(&func, X_f, Xp_f, brw_imm_f(1 / key->x_scale));
+ brw_MUL(&func, Y_f, Yp_f, brw_imm_f(1 / key->y_scale));
+ SWAP_XY_AND_XPYP();
+ } else if (!key->bilinear_filter) {
+ /* Round the float coordinates down to nearest integer by moving to
+ * UD registers.
+ */
+ brw_MOV(&func, Xp, X_f);
+ brw_MOV(&func, Yp, Y_f);
+ SWAP_XY_AND_XPYP();
+ }
+ brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
}
+void
+brw_blorp_blit_program::clamp_tex_coords(struct brw_reg regX,
+ struct brw_reg regY,
+ struct brw_reg clampX0,
+ struct brw_reg clampY0,
+ struct brw_reg clampX1,
+ struct brw_reg clampY1)
+{
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_L, regX, clampX0);
+ brw_MOV(&func, regX, clampX0);
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_G, regX, clampX1);
+ brw_MOV(&func, regX, clampX1);
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_L, regY, clampY0);
+ brw_MOV(&func, regY, clampY0);
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_G, regY, clampY1);
+ brw_MOV(&func, regY, clampY1);
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+}
+#undef X_f
+#undef Y_f
+#undef Xp_f
+#undef Yp_f
+
/**
* Emit code to transform the X and Y coordinates as needed for blending
* together the different samples in an MSAA texture.
* that maxe up a pixel). So we need to multiply our X and Y coordinates
* each by 2 and then add 1.
*/
+ brw_set_compression_control(&func, BRW_COMPRESSION_COMPRESSED);
brw_SHL(&func, t1, X, brw_imm_w(1));
brw_SHL(&func, t2, Y, brw_imm_w(1));
brw_ADD(&func, Xp, t1, brw_imm_w(1));
brw_ADD(&func, Yp, t2, brw_imm_w(1));
+ brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
SWAP_XY_AND_XPYP();
}
+
+/**
+ * Count the number of trailing 1 bits in the given value. For example:
+ *
+ * count_trailing_one_bits(0) == 0
+ * count_trailing_one_bits(7) == 3
+ * count_trailing_one_bits(11) == 2
+ */
+inline int count_trailing_one_bits(unsigned value)
+{
+#if defined(__GNUC__) && ((__GNUC__ * 100 + __GNUC_MINOR__) >= 304) /* gcc 3.4 or later */
+ return __builtin_ctz(~value);
+#else
+ return _mesa_bitcount(value & ~(value + 1));
+#endif
+}
+
+
void
-brw_blorp_blit_program::manual_blend()
+brw_blorp_blit_program::manual_blend_average(unsigned num_samples)
{
- /* TODO: support num_samples != 4 */
- const int num_samples = 4;
+ if (key->tex_layout == INTEL_MSAA_LAYOUT_CMS)
+ mcs_fetch();
+
+ /* We add together samples using a binary tree structure, e.g. for 4x MSAA:
+ *
+ * result = ((sample[0] + sample[1]) + (sample[2] + sample[3])) / 4
+ *
+ * This ensures that when all samples have the same value, no numerical
+ * precision is lost, since each addition operation always adds two equal
+ * values, and summing two equal floating point values does not lose
+ * precision.
+ *
+ * We perform this computation by treating the texture_data array as a
+ * stack and performing the following operations:
+ *
+ * - push sample 0 onto stack
+ * - push sample 1 onto stack
+ * - add top two stack entries
+ * - push sample 2 onto stack
+ * - push sample 3 onto stack
+ * - add top two stack entries
+ * - add top two stack entries
+ * - divide top stack entry by 4
+ *
+ * Note that after pushing sample i onto the stack, the number of add
+ * operations we do is equal to the number of trailing 1 bits in i. This
+ * works provided the total number of samples is a power of two, which it
+ * always is for i965.
+ *
+ * For integer formats, we replace the add operations with average
+ * operations and skip the final division.
+ */
+ typedef struct brw_instruction *(*brw_op2_ptr)(struct brw_compile *,
+ struct brw_reg,
+ struct brw_reg,
+ struct brw_reg);
+ brw_op2_ptr combine_op =
+ key->texture_data_type == BRW_REGISTER_TYPE_F ? brw_ADD : brw_AVG;
+ unsigned stack_depth = 0;
+ for (unsigned i = 0; i < num_samples; ++i) {
+ assert(stack_depth == _mesa_bitcount(i)); /* Loop invariant */
+
+ /* Push sample i onto the stack */
+ assert(stack_depth < ARRAY_SIZE(texture_data));
+ if (i == 0) {
+ s_is_zero = true;
+ } else {
+ s_is_zero = false;
+ brw_MOV(&func, vec16(S), brw_imm_ud(i));
+ }
+ texel_fetch(texture_data[stack_depth++]);
+
+ if (i == 0 && key->tex_layout == INTEL_MSAA_LAYOUT_CMS) {
+ /* The Ivy Bridge PRM, Vol4 Part1 p27 (Multisample Control Surface)
+ * suggests an optimization:
+ *
+ * "A simple optimization with probable large return in
+ * performance is to compare the MCS value to zero (indicating
+ * all samples are on sample slice 0), and sample only from
+ * sample slice 0 using ld2dss if MCS is zero."
+ *
+ * Note that in the case where the MCS value is zero, sampling from
+ * sample slice 0 using ld2dss and sampling from sample 0 using
+ * ld2dms are equivalent (since all samples are on sample slice 0).
+ * Since we have already sampled from sample 0, all we need to do is
+ * skip the remaining fetches and averaging if MCS is zero.
+ */
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_NZ,
+ mcs_data, brw_imm_ud(0));
+ brw_IF(&func, BRW_EXECUTE_16);
+ }
- /* Gather sample 0 data first */
- s_is_zero = true;
- texel_fetch(result);
+ /* Do count_trailing_one_bits(i) times */
+ for (int j = count_trailing_one_bits(i); j-- > 0; ) {
+ assert(stack_depth >= 2);
+ --stack_depth;
+
+ /* TODO: should use a smaller loop bound for non_RGBA formats */
+ for (int k = 0; k < 4; ++k) {
+ combine_op(&func, offset(texture_data[stack_depth - 1], 2*k),
+ offset(vec8(texture_data[stack_depth - 1]), 2*k),
+ offset(vec8(texture_data[stack_depth]), 2*k));
+ }
+ }
+ }
- /* Gather data for remaining samples and accumulate it into result. */
- s_is_zero = false;
- for (int i = 1; i < num_samples; ++i) {
- brw_MOV(&func, S, brw_imm_uw(i));
- texel_fetch(texture_data);
+ /* We should have just 1 sample on the stack now. */
+ assert(stack_depth == 1);
+ if (key->texture_data_type == BRW_REGISTER_TYPE_F) {
+ /* Scale the result down by a factor of num_samples */
/* TODO: should use a smaller loop bound for non-RGBA formats */
for (int j = 0; j < 4; ++j) {
- brw_ADD(&func, offset(result, 2*j), offset(vec8(result), 2*j),
- offset(vec8(texture_data), 2*j));
+ brw_MUL(&func, offset(texture_data[0], 2*j),
+ offset(vec8(texture_data[0]), 2*j),
+ brw_imm_f(1.0/num_samples));
}
}
- /* Scale the result down by a factor of num_samples */
- /* TODO: should use a smaller loop bound for non-RGBA formats */
- for (int j = 0; j < 4; ++j) {
- brw_MUL(&func, offset(result, 2*j), offset(vec8(result), 2*j),
- brw_imm_f(1.0/num_samples));
+ if (key->tex_layout == INTEL_MSAA_LAYOUT_CMS)
+ brw_ENDIF(&func);
+}
+
+void
+brw_blorp_blit_program::manual_blend_bilinear(unsigned num_samples)
+{
+ /* We do this computation by performing the following operations:
+ *
+ * In case of 4x, 8x MSAA:
+ * - Compute the pixel coordinates and sample numbers (a, b, c, d)
+ * which are later used for interpolation
+ * - linearly interpolate samples a and b in X
+ * - linearly interpolate samples c and d in X
+ * - linearly interpolate the results of last two operations in Y
+ *
+ * result = lrp(lrp(a + b) + lrp(c + d))
+ */
+ struct brw_reg Xp_f = retype(Xp, BRW_REGISTER_TYPE_F);
+ struct brw_reg Yp_f = retype(Yp, BRW_REGISTER_TYPE_F);
+ struct brw_reg t1_f = retype(t1, BRW_REGISTER_TYPE_F);
+ struct brw_reg t2_f = retype(t2, BRW_REGISTER_TYPE_F);
+
+ for (unsigned i = 0; i < 4; ++i) {
+ assert(i < ARRAY_SIZE(texture_data));
+ s_is_zero = false;
+
+ /* Compute pixel coordinates */
+ brw_ADD(&func, vec16(x_sample_coords), Xp_f,
+ brw_imm_f((float)(i & 0x1) * (1.0 / key->x_scale)));
+ brw_ADD(&func, vec16(y_sample_coords), Yp_f,
+ brw_imm_f((float)((i >> 1) & 0x1) * (1.0 / key->y_scale)));
+ brw_MOV(&func, vec16(X), x_sample_coords);
+ brw_MOV(&func, vec16(Y), y_sample_coords);
+
+ /* The MCS value we fetch has to match up with the pixel that we're
+ * sampling from. Since we sample from different pixels in each
+ * iteration of this "for" loop, the call to mcs_fetch() should be
+ * here inside the loop after computing the pixel coordinates.
+ */
+ if (key->tex_layout == INTEL_MSAA_LAYOUT_CMS)
+ mcs_fetch();
+
+ /* Compute sample index and map the sample index to a sample number.
+ * Sample index layout shows the numbering of slots in a rectangular
+ * grid of samples with in a pixel. Sample number layout shows the
+ * rectangular grid of samples roughly corresponding to the real sample
+ * locations with in a pixel.
+ * In case of 4x MSAA, layout of sample indices matches the layout of
+ * sample numbers:
+ * ---------
+ * | 0 | 1 |
+ * ---------
+ * | 2 | 3 |
+ * ---------
+ *
+ * In case of 8x MSAA the two layouts don't match.
+ * sample index layout : --------- sample number layout : ---------
+ * | 0 | 1 | | 5 | 2 |
+ * --------- ---------
+ * | 2 | 3 | | 4 | 6 |
+ * --------- ---------
+ * | 4 | 5 | | 0 | 3 |
+ * --------- ---------
+ * | 6 | 7 | | 7 | 1 |
+ * --------- ---------
+ */
+ brw_FRC(&func, vec16(t1_f), x_sample_coords);
+ brw_FRC(&func, vec16(t2_f), y_sample_coords);
+ brw_MUL(&func, vec16(t1_f), t1_f, brw_imm_f(key->x_scale));
+ brw_MUL(&func, vec16(t2_f), t2_f, brw_imm_f(key->x_scale * key->y_scale));
+ brw_ADD(&func, vec16(t1_f), t1_f, t2_f);
+ brw_MOV(&func, vec16(S), t1_f);
+
+ if (num_samples == 8) {
+ /* Map the sample index to a sample number */
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_L,
+ S, brw_imm_d(4));
+ brw_IF(&func, BRW_EXECUTE_16);
+ {
+ brw_MOV(&func, vec16(t2), brw_imm_d(5));
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_EQ,
+ S, brw_imm_d(1));
+ brw_MOV(&func, vec16(t2), brw_imm_d(2));
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_EQ,
+ S, brw_imm_d(2));
+ brw_MOV(&func, vec16(t2), brw_imm_d(4));
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_EQ,
+ S, brw_imm_d(3));
+ brw_MOV(&func, vec16(t2), brw_imm_d(6));
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+ }
+ brw_ELSE(&func);
+ {
+ brw_MOV(&func, vec16(t2), brw_imm_d(0));
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_EQ,
+ S, brw_imm_d(5));
+ brw_MOV(&func, vec16(t2), brw_imm_d(3));
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_EQ,
+ S, brw_imm_d(6));
+ brw_MOV(&func, vec16(t2), brw_imm_d(7));
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+ brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_EQ,
+ S, brw_imm_d(7));
+ brw_MOV(&func, vec16(t2), brw_imm_d(1));
+ brw_set_predicate_control(&func, BRW_PREDICATE_NONE);
+ }
+ brw_ENDIF(&func);
+ brw_MOV(&func, vec16(S), t2);
+ }
+ texel_fetch(texture_data[i]);
}
+
+#define SAMPLE(x, y) offset(texture_data[x], y)
+ brw_set_access_mode(&func, BRW_ALIGN_16);
+ for (int index = 3; index > 0; ) {
+ /* Since we're doing SIMD16, 4 color channels fits in to 8 registers.
+ * Counter value of 8 in 'for' loop below is used to interpolate all
+ * the color components.
+ */
+ for (int k = 0; k < 8; ++k)
+ brw_LRP(&func,
+ vec8(SAMPLE(index - 1, k)),
+ offset(x_frac, k & 1),
+ SAMPLE(index, k),
+ SAMPLE(index - 1, k));
+ index -= 2;
+ }
+ for (int k = 0; k < 8; ++k)
+ brw_LRP(&func,
+ vec8(SAMPLE(0, k)),
+ offset(y_frac, k & 1),
+ vec8(SAMPLE(2, k)),
+ vec8(SAMPLE(0, k)));
+ brw_set_access_mode(&func, BRW_ALIGN_1);
+#undef SAMPLE
}
/**
SAMPLER_MESSAGE_ARG_V_FLOAT
};
- texture_lookup(dst, GEN5_SAMPLER_MESSAGE_SAMPLE, args, ARRAY_SIZE(args));
+ texture_lookup(dst, GEN5_SAMPLER_MESSAGE_SAMPLE, args,
+ ARRAY_SIZE(args));
}
/**
SAMPLER_MESSAGE_ARG_U_INT,
SAMPLER_MESSAGE_ARG_V_INT
};
+ static const sampler_message_arg gen7_ld2dms_args[4] = {
+ SAMPLER_MESSAGE_ARG_SI_INT,
+ SAMPLER_MESSAGE_ARG_MCS_INT,
+ SAMPLER_MESSAGE_ARG_U_INT,
+ SAMPLER_MESSAGE_ARG_V_INT
+ };
- switch (brw->intel.gen) {
+ switch (brw->gen) {
case 6:
texture_lookup(dst, GEN5_SAMPLER_MESSAGE_SAMPLE_LD, gen6_args,
s_is_zero ? 2 : 5);
break;
case 7:
- if (key->tex_samples > 0) {
+ switch (key->tex_layout) {
+ case INTEL_MSAA_LAYOUT_IMS:
+ /* From the Ivy Bridge PRM, Vol4 Part1 p72 (Multisampled Surface Storage
+ * Format):
+ *
+ * If this field is MSFMT_DEPTH_STENCIL
+ * [a.k.a. INTEL_MSAA_LAYOUT_IMS], the only sampling engine
+ * messages allowed are "ld2dms", "resinfo", and "sampleinfo".
+ *
+ * So fall through to emit the same message as we use for
+ * INTEL_MSAA_LAYOUT_CMS.
+ */
+ case INTEL_MSAA_LAYOUT_CMS:
+ texture_lookup(dst, GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS,
+ gen7_ld2dms_args, ARRAY_SIZE(gen7_ld2dms_args));
+ break;
+ case INTEL_MSAA_LAYOUT_UMS:
texture_lookup(dst, GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DSS,
gen7_ld2dss_args, ARRAY_SIZE(gen7_ld2dss_args));
- } else {
+ break;
+ case INTEL_MSAA_LAYOUT_NONE:
assert(s_is_zero);
texture_lookup(dst, GEN5_SAMPLER_MESSAGE_SAMPLE_LD, gen7_ld_args,
ARRAY_SIZE(gen7_ld_args));
+ break;
}
break;
default:
}
void
-brw_blorp_blit_program::expand_to_32_bits(struct brw_reg src,
- struct brw_reg dst)
+brw_blorp_blit_program::mcs_fetch()
{
- brw_MOV(&func, vec8(dst), vec8(src));
- brw_set_compression_control(&func, BRW_COMPRESSION_2NDHALF);
- brw_MOV(&func, offset(vec8(dst), 1), suboffset(vec8(src), 8));
- brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
+ static const sampler_message_arg gen7_ld_mcs_args[2] = {
+ SAMPLER_MESSAGE_ARG_U_INT,
+ SAMPLER_MESSAGE_ARG_V_INT
+ };
+ texture_lookup(vec16(mcs_data), GEN7_SAMPLER_MESSAGE_SAMPLE_LD_MCS,
+ gen7_ld_mcs_args, ARRAY_SIZE(gen7_ld_mcs_args));
}
void
for (int arg = 0; arg < num_args; ++arg) {
switch (args[arg]) {
case SAMPLER_MESSAGE_ARG_U_FLOAT:
- expand_to_32_bits(X, retype(mrf, BRW_REGISTER_TYPE_F));
+ if (key->bilinear_filter)
+ brw_MOV(&func, retype(mrf, BRW_REGISTER_TYPE_F),
+ retype(X, BRW_REGISTER_TYPE_F));
+ else
+ brw_MOV(&func, retype(mrf, BRW_REGISTER_TYPE_F), X);
break;
case SAMPLER_MESSAGE_ARG_V_FLOAT:
- expand_to_32_bits(Y, retype(mrf, BRW_REGISTER_TYPE_F));
+ if (key->bilinear_filter)
+ brw_MOV(&func, retype(mrf, BRW_REGISTER_TYPE_F),
+ retype(Y, BRW_REGISTER_TYPE_F));
+ else
+ brw_MOV(&func, retype(mrf, BRW_REGISTER_TYPE_F), Y);
break;
case SAMPLER_MESSAGE_ARG_U_INT:
- expand_to_32_bits(X, mrf);
+ brw_MOV(&func, mrf, X);
break;
case SAMPLER_MESSAGE_ARG_V_INT:
- expand_to_32_bits(Y, mrf);
+ brw_MOV(&func, mrf, Y);
break;
case SAMPLER_MESSAGE_ARG_SI_INT:
/* Note: on Gen7, this code may be reached with s_is_zero==true
if (s_is_zero)
brw_MOV(&func, mrf, brw_imm_ud(0));
else
- expand_to_32_bits(S, mrf);
+ brw_MOV(&func, mrf, S);
+ break;
+ case SAMPLER_MESSAGE_ARG_MCS_INT:
+ switch (key->tex_layout) {
+ case INTEL_MSAA_LAYOUT_CMS:
+ brw_MOV(&func, mrf, mcs_data);
+ break;
+ case INTEL_MSAA_LAYOUT_IMS:
+ /* When sampling from an IMS surface, MCS data is not relevant,
+ * and the hardware ignores it. So don't bother populating it.
+ */
+ break;
+ default:
+ /* We shouldn't be trying to send MCS data with any other
+ * layouts.
+ */
+ assert (!"Unsupported layout for MCS data");
+ break;
+ }
break;
case SAMPLER_MESSAGE_ARG_ZERO_INT:
brw_MOV(&func, mrf, brw_imm_ud(0));
brw_message_reg(base_mrf) /* src0 */,
BRW_BLORP_TEXTURE_BINDING_TABLE_INDEX,
0 /* sampler */,
- WRITEMASK_XYZW,
msg_type,
8 /* response_length. TODO: should be smaller for non-RGBA formats? */,
mrf.nr - base_mrf /* msg_length */,
void
brw_blorp_blit_program::render_target_write()
{
- struct brw_reg mrf_rt_write = vec16(brw_message_reg(base_mrf));
+ struct brw_reg mrf_rt_write =
+ retype(vec16(brw_message_reg(base_mrf)), key->texture_data_type);
int mrf_offset = 0;
/* If we may have killed pixels, then we need to send R0 and R1 in a header
for (int i = 0; i < 4; ++i) {
/* E.g. mov(16) m2.0<1>:f r2.0<8;8,1>:f { Align1, H1 } */
brw_MOV(&func, offset(mrf_rt_write, mrf_offset),
- offset(vec8(result), 2*i));
+ offset(vec8(texture_data[0]), 2*i));
mrf_offset += 2;
}
void
-brw_blorp_coord_transform_params::setup(GLuint src0, GLuint dst0, GLuint dst1,
+brw_blorp_coord_transform_params::setup(GLfloat src0, GLfloat src1,
+ GLfloat dst0, GLfloat dst1,
bool mirror)
{
+ float scale = (src1 - src0) / (dst1 - dst0);
if (!mirror) {
/* When not mirroring a coordinate (say, X), we need:
- * x' - src_x0 = x - dst_x0
+ * src_x - src_x0 = (dst_x - dst_x0 + 0.5) * scale
* Therefore:
- * x' = 1*x + (src_x0 - dst_x0)
+ * src_x = src_x0 + (dst_x - dst_x0 + 0.5) * scale
+ *
+ * blorp program uses "round toward zero" to convert the
+ * transformed floating point coordinates to integer coordinates,
+ * whereas the behaviour we actually want is "round to nearest",
+ * so 0.5 provides the necessary correction.
*/
- multiplier = 1;
- offset = src0 - dst0;
+ multiplier = scale;
+ offset = src0 + (-dst0 + 0.5) * scale;
} else {
/* When mirroring X we need:
- * x' - src_x0 = dst_x1 - x - 1
+ * src_x - src_x0 = dst_x1 - dst_x - 0.5
* Therefore:
- * x' = -1*x + (src_x0 + dst_x1 - 1)
+ * src_x = src_x0 + (dst_x1 -dst_x - 0.5) * scale
*/
- multiplier = -1;
- offset = src0 + dst1 - 1;
+ multiplier = -scale;
+ offset = src0 + (dst1 - 0.5) * scale;
}
}
+/**
+ * Determine which MSAA layout the GPU pipeline should be configured for,
+ * based on the chip generation, the number of samples, and the true layout of
+ * the image in memory.
+ */
+inline intel_msaa_layout
+compute_msaa_layout_for_pipeline(struct brw_context *brw, unsigned num_samples,
+ intel_msaa_layout true_layout)
+{
+ if (num_samples <= 1) {
+ /* When configuring the GPU for non-MSAA, we can still accommodate IMS
+ * format buffers, by transforming coordinates appropriately.
+ */
+ assert(true_layout == INTEL_MSAA_LAYOUT_NONE ||
+ true_layout == INTEL_MSAA_LAYOUT_IMS);
+ return INTEL_MSAA_LAYOUT_NONE;
+ } else {
+ assert(true_layout != INTEL_MSAA_LAYOUT_NONE);
+ }
+
+ /* Prior to Gen7, all MSAA surfaces use IMS layout. */
+ if (brw->gen == 6) {
+ assert(true_layout == INTEL_MSAA_LAYOUT_IMS);
+ }
+
+ return true_layout;
+}
+
+
brw_blorp_blit_params::brw_blorp_blit_params(struct brw_context *brw,
struct intel_mipmap_tree *src_mt,
+ unsigned src_level, unsigned src_layer,
struct intel_mipmap_tree *dst_mt,
- GLuint src_x0, GLuint src_y0,
- GLuint dst_x0, GLuint dst_y0,
- GLuint dst_x1, GLuint dst_y1,
+ unsigned dst_level, unsigned dst_layer,
+ GLfloat src_x0, GLfloat src_y0,
+ GLfloat src_x1, GLfloat src_y1,
+ GLfloat dst_x0, GLfloat dst_y0,
+ GLfloat dst_x1, GLfloat dst_y1,
+ GLenum filter,
bool mirror_x, bool mirror_y)
{
- src.set(src_mt, 0, 0);
- dst.set(dst_mt, 0, 0);
+ struct gl_context *ctx = &brw->ctx;
+ const struct gl_framebuffer *read_fb = ctx->ReadBuffer;
+
+ src.set(brw, src_mt, src_level, src_layer, false);
+ dst.set(brw, dst_mt, dst_level, dst_layer, true);
+
+ /* Even though we do multisample resolves at the time of the blit, OpenGL
+ * specification defines them as if they happen at the time of rendering,
+ * which means that the type of averaging we do during the resolve should
+ * only depend on the source format; the destination format should be
+ * ignored. But, specification doesn't seem to be strict about it.
+ *
+ * It has been observed that mulitisample resolves produce slightly better
+ * looking images when averaging is done using destination format. NVIDIA's
+ * proprietary OpenGL driver also follow this approach. So, we choose to
+ * follow it in our driver.
+ *
+ * Following if-else block takes care of this exception made for
+ * multisampled resolves.
+ */
+ if (src.num_samples > 1)
+ src.brw_surfaceformat = dst.brw_surfaceformat;
+ else
+ dst.brw_surfaceformat = src.brw_surfaceformat;
use_wm_prog = true;
memset(&wm_prog_key, 0, sizeof(wm_prog_key));
- if (brw->intel.gen > 6) {
- /* Gen7 only supports interleaved MSAA surfaces for texturing with the
- * ld2dms instruction (which blorp doesn't use). So if the source is
- * interleaved MSAA, we'll have to map it as a single-sampled texture
- * and de-interleave the samples ourselves.
- */
- if (src.num_samples > 0 && src_mt->msaa_is_interleaved)
- src.num_samples = 0;
+ /* texture_data_type indicates the register type that should be used to
+ * manipulate texture data.
+ */
+ switch (_mesa_get_format_datatype(src_mt->format)) {
+ case GL_UNSIGNED_NORMALIZED:
+ case GL_SIGNED_NORMALIZED:
+ case GL_FLOAT:
+ wm_prog_key.texture_data_type = BRW_REGISTER_TYPE_F;
+ break;
+ case GL_UNSIGNED_INT:
+ if (src_mt->format == MESA_FORMAT_S8) {
+ /* We process stencil as though it's an unsigned normalized color */
+ wm_prog_key.texture_data_type = BRW_REGISTER_TYPE_F;
+ } else {
+ wm_prog_key.texture_data_type = BRW_REGISTER_TYPE_UD;
+ }
+ break;
+ case GL_INT:
+ wm_prog_key.texture_data_type = BRW_REGISTER_TYPE_D;
+ break;
+ default:
+ assert(!"Unrecognized blorp format");
+ break;
+ }
- /* Similarly, Gen7 only supports interleaved MSAA surfaces for depth and
+ if (brw->gen > 6) {
+ /* Gen7's rendering hardware only supports the IMS layout for depth and
* stencil render targets. Blorp always maps its destination surface as
* a color render target (even if it's actually a depth or stencil
- * buffer). So if the destination is interleaved MSAA, we'll have to
- * map it as a single-sampled texture and interleave the samples
- * ourselves.
+ * buffer). So if the destination is IMS, we'll have to map it as a
+ * single-sampled texture and interleave the samples ourselves.
*/
- if (dst.num_samples > 0 && dst_mt->msaa_is_interleaved)
+ if (dst_mt->msaa_layout == INTEL_MSAA_LAYOUT_IMS)
dst.num_samples = 0;
}
- if (dst.map_stencil_as_y_tiled && dst.num_samples > 0) {
+ if (dst.map_stencil_as_y_tiled && dst.num_samples > 1) {
/* If the destination surface is a W-tiled multisampled stencil buffer
* that we're mapping as Y tiled, then we need to arrange for the WM
* program to run once per sample rather than once per pixel, because
wm_prog_key.persample_msaa_dispatch = true;
}
- if (src.num_samples > 0 && dst.num_samples > 0) {
+ if (src.num_samples > 0 && dst.num_samples > 1) {
/* We are blitting from a multisample buffer to a multisample buffer, so
* we must preserve samples within a pixel. This means we have to
* arrange for the WM program to run once per sample rather than once
wm_prog_key.persample_msaa_dispatch = true;
}
+ /* Scaled blitting or not. */
+ wm_prog_key.blit_scaled =
+ ((dst_x1 - dst_x0) == (src_x1 - src_x0) &&
+ (dst_y1 - dst_y0) == (src_y1 - src_y0)) ? false : true;
+
+ /* Scaling factors used for bilinear filtering in multisample scaled
+ * blits.
+ */
+ wm_prog_key.x_scale = 2.0;
+ wm_prog_key.y_scale = src_mt->num_samples / 2.0;
+
+ if (filter == GL_LINEAR)
+ wm_prog_key.bilinear_filter = true;
+
/* The render path must be configured to use the same number of samples as
* the destination buffer.
*/
GLenum base_format = _mesa_get_format_base_format(src_mt->format);
if (base_format != GL_DEPTH_COMPONENT && /* TODO: what about depth/stencil? */
base_format != GL_STENCIL_INDEX &&
- src_mt->num_samples > 0 && dst_mt->num_samples == 0) {
+ src_mt->num_samples > 1 && dst_mt->num_samples <= 1) {
/* We are downsampling a color buffer, so blend. */
wm_prog_key.blend = true;
}
wm_prog_key.tex_samples = src.num_samples;
wm_prog_key.rt_samples = dst.num_samples;
- /* src_interleaved and dst_interleaved indicate whether src and dst are
- * truly interleaved.
+ /* tex_layout and rt_layout indicate the MSAA layout the GPU pipeline will
+ * use to access the source and destination surfaces.
+ */
+ wm_prog_key.tex_layout =
+ compute_msaa_layout_for_pipeline(brw, src.num_samples, src.msaa_layout);
+ wm_prog_key.rt_layout =
+ compute_msaa_layout_for_pipeline(brw, dst.num_samples, dst.msaa_layout);
+
+ /* src_layout and dst_layout indicate the true MSAA layout used by src and
+ * dst.
*/
- wm_prog_key.src_interleaved = src_mt->msaa_is_interleaved;
- wm_prog_key.dst_interleaved = dst_mt->msaa_is_interleaved;
+ wm_prog_key.src_layout = src_mt->msaa_layout;
+ wm_prog_key.dst_layout = dst_mt->msaa_layout;
wm_prog_key.src_tiled_w = src.map_stencil_as_y_tiled;
wm_prog_key.dst_tiled_w = dst.map_stencil_as_y_tiled;
y0 = wm_push_consts.dst_y0 = dst_y0;
x1 = wm_push_consts.dst_x1 = dst_x1;
y1 = wm_push_consts.dst_y1 = dst_y1;
- wm_push_consts.x_transform.setup(src_x0, dst_x0, dst_x1, mirror_x);
- wm_push_consts.y_transform.setup(src_y0, dst_y0, dst_y1, mirror_y);
+ wm_push_consts.rect_grid_x1 = read_fb->Width * wm_prog_key.x_scale - 1.0;
+ wm_push_consts.rect_grid_y1 = read_fb->Height * wm_prog_key.y_scale - 1.0;
- if (dst.num_samples == 0 && dst_mt->num_samples > 0) {
+ wm_push_consts.x_transform.setup(src_x0, src_x1, dst_x0, dst_x1, mirror_x);
+ wm_push_consts.y_transform.setup(src_y0, src_y1, dst_y0, dst_y1, mirror_y);
+
+ if (dst.num_samples <= 1 && dst_mt->num_samples > 1) {
/* We must expand the rectangle we send through the rendering pipeline,
* to account for the fact that we are mapping the destination region as
* single-sampled when it is in fact multisampled. We must also align
* will mean that pixels are scrambled within the multisampling pattern.
* TODO: what if this makes the coordinates too large?
*
- * Note: this only works if the destination surface's MSAA layout is
- * interleaved. If it's sliced, then we have no choice but to set up
- * the rendering pipeline as multisampled.
+ * Note: this only works if the destination surface uses the IMS layout.
+ * If it's UMS, then we have no choice but to set up the rendering
+ * pipeline as multisampled.
*/
- assert(dst_mt->msaa_is_interleaved);
- x0 = (x0 * 2) & ~3;
- y0 = (y0 * 2) & ~3;
- x1 = ALIGN(x1 * 2, 4);
- y1 = ALIGN(y1 * 2, 4);
+ assert(dst_mt->msaa_layout == INTEL_MSAA_LAYOUT_IMS);
+ switch (dst_mt->num_samples) {
+ case 4:
+ x0 = ROUND_DOWN_TO(x0 * 2, 4);
+ y0 = ROUND_DOWN_TO(y0 * 2, 4);
+ x1 = ALIGN(x1 * 2, 4);
+ y1 = ALIGN(y1 * 2, 4);
+ break;
+ case 8:
+ x0 = ROUND_DOWN_TO(x0 * 4, 8);
+ y0 = ROUND_DOWN_TO(y0 * 2, 4);
+ x1 = ALIGN(x1 * 4, 8);
+ y1 = ALIGN(y1 * 2, 4);
+ break;
+ default:
+ assert(!"Unrecognized sample count in brw_blorp_blit_params ctor");
+ break;
+ }
wm_prog_key.use_kill = true;
}
if (dst.map_stencil_as_y_tiled) {
- /* We must modify the rectangle we send through the rendering pipeline,
- * to account for the fact that we are mapping it as Y-tiled when it is
- * in fact W-tiled. Y tiles have dimensions 128x32 whereas W tiles have
- * dimensions 64x64. We must also align it to a multiple of the tile
- * size, because the differences between W and Y tiling formats will
- * mean that pixels are scrambled within the tile.
+ /* We must modify the rectangle we send through the rendering pipeline
+ * (and the size and x/y offset of the destination surface), to account
+ * for the fact that we are mapping it as Y-tiled when it is in fact
+ * W-tiled.
*
- * Note: if the destination surface configured as an interleaved MSAA
- * surface, then the effective tile size we need to align it to is
- * smaller, because each pixel covers a 2x2 or a 4x2 block of samples.
+ * Both Y tiling and W tiling can be understood as organizations of
+ * 32-byte sub-tiles; within each 32-byte sub-tile, the layout of pixels
+ * is different, but the layout of the 32-byte sub-tiles within the 4k
+ * tile is the same (8 sub-tiles across by 16 sub-tiles down, in
+ * column-major order). In Y tiling, the sub-tiles are 16 bytes wide
+ * and 2 rows high; in W tiling, they are 8 bytes wide and 4 rows high.
*
- * TODO: what if this makes the coordinates too large?
+ * Therefore, to account for the layout differences within the 32-byte
+ * sub-tiles, we must expand the rectangle so the X coordinates of its
+ * edges are multiples of 8 (the W sub-tile width), and its Y
+ * coordinates of its edges are multiples of 4 (the W sub-tile height).
+ * Then we need to scale the X and Y coordinates of the rectangle to
+ * account for the differences in aspect ratio between the Y and W
+ * sub-tiles. We need to modify the layer width and height similarly.
+ *
+ * A correction needs to be applied when MSAA is in use: since
+ * INTEL_MSAA_LAYOUT_IMS uses an interleaving pattern whose height is 4,
+ * we need to align the Y coordinates to multiples of 8, so that when
+ * they are divided by two they are still multiples of 4.
+ *
+ * Note: Since the x/y offset of the surface will be applied using the
+ * SURFACE_STATE command packet, it will be invisible to the swizzling
+ * code in the shader; therefore it needs to be in a multiple of the
+ * 32-byte sub-tile size. Fortunately it is, since the sub-tile is 8
+ * pixels wide and 4 pixels high (when viewed as a W-tiled stencil
+ * buffer), and the miplevel alignment used for stencil buffers is 8
+ * pixels horizontally and either 4 or 8 pixels vertically (see
+ * intel_horizontal_texture_alignment_unit() and
+ * intel_vertical_texture_alignment_unit()).
+ *
+ * Note: Also, since the SURFACE_STATE command packet can only apply
+ * offsets that are multiples of 4 pixels horizontally and 2 pixels
+ * vertically, it is important that the offsets will be multiples of
+ * these sizes after they are converted into Y-tiled coordinates.
+ * Fortunately they will be, since we know from above that the offsets
+ * are a multiple of the 32-byte sub-tile size, and in Y-tiled
+ * coordinates the sub-tile is 16 pixels wide and 2 pixels high.
+ *
+ * TODO: what if this makes the coordinates (or the texture size) too
+ * large?
*/
- unsigned x_align = 64, y_align = 64;
- if (dst_mt->num_samples > 0 && dst_mt->msaa_is_interleaved) {
- x_align /= (dst_mt->num_samples == 4 ? 2 : 4);
- y_align /= 2;
- }
- x0 = (x0 & ~(x_align - 1)) * 2;
- y0 = (y0 & ~(y_align - 1)) / 2;
+ const unsigned x_align = 8, y_align = dst.num_samples != 0 ? 8 : 4;
+ x0 = ROUND_DOWN_TO(x0, x_align) * 2;
+ y0 = ROUND_DOWN_TO(y0, y_align) / 2;
x1 = ALIGN(x1, x_align) * 2;
y1 = ALIGN(y1, y_align) / 2;
+ dst.width = ALIGN(dst.width, x_align) * 2;
+ dst.height = ALIGN(dst.height, y_align) / 2;
+ dst.x_offset *= 2;
+ dst.y_offset /= 2;
wm_prog_key.use_kill = true;
}
+
+ if (src.map_stencil_as_y_tiled) {
+ /* We must modify the size and x/y offset of the source surface to
+ * account for the fact that we are mapping it as Y-tiled when it is in
+ * fact W tiled.
+ *
+ * See the comments above concerning x/y offset alignment for the
+ * destination surface.
+ *
+ * TODO: what if this makes the texture size too large?
+ */
+ const unsigned x_align = 8, y_align = src.num_samples != 0 ? 8 : 4;
+ src.width = ALIGN(src.width, x_align) * 2;
+ src.height = ALIGN(src.height, y_align) / 2;
+ src.x_offset *= 2;
+ src.y_offset /= 2;
+ }
}
uint32_t
brw_blorp_blit_params::get_wm_prog(struct brw_context *brw,
brw_blorp_prog_data **prog_data) const
{
- uint32_t prog_offset;
+ uint32_t prog_offset = 0;
if (!brw_search_cache(&brw->cache, BRW_BLORP_BLIT_PROG,
&this->wm_prog_key, sizeof(this->wm_prog_key),
&prog_offset, prog_data)) {
projects / mesa.git / blobdiff