void
brw_populate_sampler_prog_key_data(struct gl_context *ctx,
const struct gl_program *prog,
unsigned sampler_count,
struct brw_sampler_prog_key_data *key)
{
struct brw_context *brw = brw_context(ctx);
for (int s = 0; s < sampler_count; s++) {
key->swizzles[s] = SWIZZLE_NOOP;
if (!(prog->SamplersUsed & (1 << s)))
continue;
int unit_id = prog->SamplerUnits[s];
const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
const struct gl_texture_object *t = unit->_Current;
const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
const bool alpha_depth = t->DepthMode == GL_ALPHA &&
(img->_BaseFormat == GL_DEPTH_COMPONENT ||
img->_BaseFormat == GL_DEPTH_STENCIL);
/* Haswell handles texture swizzling as surface format overrides
* (except for GL_ALPHA); all other platforms need MOVs in the shader.
*/
if (!brw->is_haswell || alpha_depth)
key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
if (img->InternalFormat == GL_YCBCR_MESA) {
key->yuvtex_mask |= 1 << s;
if (img->TexFormat == MESA_FORMAT_YCBCR)
key->yuvtex_swap_mask |= 1 << s;
}
if (sampler->MinFilter != GL_NEAREST &&
sampler->MagFilter != GL_NEAREST) {
if (sampler->WrapS == GL_CLAMP)
key->gl_clamp_mask[0] |= 1 << s;
if (sampler->WrapT == GL_CLAMP)
key->gl_clamp_mask[1] |= 1 << s;
if (sampler->WrapR == GL_CLAMP)
key->gl_clamp_mask[2] |= 1 << s;
}
/* gather4's channel select for green from RG32F is broken;
* requires a shader w/a on IVB; fixable with just SCS on HSW. */
if (brw->gen >= 7 && !brw->is_haswell && prog->UsesGather) {
if (img->InternalFormat == GL_RG32F)
key->gather_channel_quirk_mask |= 1 << s;
}
}
}
}
void
brw_populate_sampler_prog_key_data(struct gl_context *ctx,
const struct gl_program *prog,
struct brw_sampler_prog_key_data *key)
{
struct intel_context *intel = intel_context(ctx);
for (int s = 0; s < MAX_SAMPLERS; s++) {
key->swizzles[s] = SWIZZLE_NOOP;
if (!(prog->SamplersUsed & (1 << s)))
continue;
int unit_id = prog->SamplerUnits[s];
const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
const struct gl_texture_object *t = unit->_Current;
const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
const bool alpha_depth = t->DepthMode == GL_ALPHA &&
(img->_BaseFormat == GL_DEPTH_COMPONENT ||
img->_BaseFormat == GL_DEPTH_STENCIL);
/* Haswell handles texture swizzling as surface format overrides
* (except for GL_ALPHA); all other platforms need MOVs in the shader.
*/
if (!intel->is_haswell || alpha_depth)
key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
if (img->InternalFormat == GL_YCBCR_MESA) {
key->yuvtex_mask |= 1 << s;
if (img->TexFormat == MESA_FORMAT_YCBCR)
key->yuvtex_swap_mask |= 1 << s;
}
if (sampler->MinFilter != GL_NEAREST &&
sampler->MagFilter != GL_NEAREST) {
if (sampler->WrapS == GL_CLAMP)
key->gl_clamp_mask[0] |= 1 << s;
if (sampler->WrapT == GL_CLAMP)
key->gl_clamp_mask[1] |= 1 << s;
if (sampler->WrapR == GL_CLAMP)
key->gl_clamp_mask[2] |= 1 << s;
}
}
}
}
void
brw_populate_sampler_prog_key_data(struct gl_context *ctx,
const struct gl_program *prog,
struct brw_sampler_prog_key_data *key)
{
struct brw_context *brw = brw_context(ctx);
const struct gen_device_info *devinfo = &brw->screen->devinfo;
GLbitfield mask = prog->SamplersUsed;
while (mask) {
const int s = u_bit_scan(&mask);
key->swizzles[s] = SWIZZLE_NOOP;
key->scale_factors[s] = 0.0f;
int unit_id = prog->SamplerUnits[s];
const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
const struct gl_texture_object *t = unit->_Current;
const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
const bool alpha_depth = t->DepthMode == GL_ALPHA &&
(img->_BaseFormat == GL_DEPTH_COMPONENT ||
img->_BaseFormat == GL_DEPTH_STENCIL);
/* Haswell handles texture swizzling as surface format overrides
* (except for GL_ALPHA); all other platforms need MOVs in the shader.
*/
if (alpha_depth || (devinfo->gen < 8 && !devinfo->is_haswell))
key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
if (devinfo->gen < 8 &&
sampler->MinFilter != GL_NEAREST &&
sampler->MagFilter != GL_NEAREST) {
if (sampler->WrapS == GL_CLAMP)
key->gl_clamp_mask[0] |= 1 << s;
if (sampler->WrapT == GL_CLAMP)
key->gl_clamp_mask[1] |= 1 << s;
if (sampler->WrapR == GL_CLAMP)
key->gl_clamp_mask[2] |= 1 << s;
}
/* gather4 for RG32* is broken in multiple ways on Gen7. */
if (devinfo->gen == 7 && prog->info.uses_texture_gather) {
switch (img->InternalFormat) {
case GL_RG32I:
case GL_RG32UI: {
/* We have to override the format to R32G32_FLOAT_LD.
* This means that SCS_ALPHA and SCS_ONE will return 0x3f8
* (1.0) rather than integer 1. This needs shader hacks.
*
* On Ivybridge, we whack W (alpha) to ONE in our key's
* swizzle. On Haswell, we look at the original texture
* swizzle, and use XYZW with channels overridden to ONE,
* leaving normal texture swizzling to SCS.
*/
unsigned src_swizzle =
devinfo->is_haswell ? t->_Swizzle : key->swizzles[s];
for (int i = 0; i < 4; i++) {
unsigned src_comp = GET_SWZ(src_swizzle, i);
if (src_comp == SWIZZLE_ONE || src_comp == SWIZZLE_W) {
key->swizzles[i] &= ~(0x7 << (3 * i));
key->swizzles[i] |= SWIZZLE_ONE << (3 * i);
}
}
/* fallthrough */
}
case GL_RG32F:
/* The channel select for green doesn't work - we have to
* request blue. Haswell can use SCS for this, but Ivybridge
* needs a shader workaround.
*/
if (!devinfo->is_haswell)
key->gather_channel_quirk_mask |= 1 << s;
break;
}
}
/* Gen6's gather4 is broken for UINT/SINT; we treat them as
* UNORM/FLOAT instead and fix it in the shader.
*/
if (devinfo->gen == 6 && prog->info.uses_texture_gather) {
key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
}
/* If this is a multisample sampler, and uses the CMS MSAA layout,
* then we need to emit slightly different code to first sample the
* MCS surface.
*/
struct intel_texture_object *intel_tex =
intel_texture_object((struct gl_texture_object *)t);
/* From gen9 onwards some single sampled buffers can also be
* compressed. These don't need ld2dms sampling along with mcs fetch.
*/
if (intel_tex->mt->aux_usage == ISL_AUX_USAGE_MCS) {
assert(devinfo->gen >= 7);
assert(intel_tex->mt->surf.samples > 1);
assert(intel_tex->mt->aux_buf);
assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
key->compressed_multisample_layout_mask |= 1 << s;
if (intel_tex->mt->surf.samples >= 16) {
assert(devinfo->gen >= 9);
key->msaa_16 |= 1 << s;
}
}
if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
/* Setup possible scaling factor. */
key->scale_factors[s] = intel_tex->planar_format->scaling_factor;
switch (intel_tex->planar_format->components) {
case __DRI_IMAGE_COMPONENTS_Y_UV:
key->y_uv_image_mask |= 1 << s;
break;
case __DRI_IMAGE_COMPONENTS_Y_U_V:
key->y_u_v_image_mask |= 1 << s;
break;
case __DRI_IMAGE_COMPONENTS_Y_XUXV:
key->yx_xuxv_image_mask |= 1 << s;
break;
case __DRI_IMAGE_COMPONENTS_Y_UXVX:
key->xy_uxvx_image_mask |= 1 << s;
break;
case __DRI_IMAGE_COMPONENTS_AYUV:
key->ayuv_image_mask |= 1 << s;
break;
case __DRI_IMAGE_COMPONENTS_XYUV:
key->xyuv_image_mask |= 1 << s;
break;
default:
break;
}
}
}
}
}
void
brw_populate_sampler_prog_key_data(struct gl_context *ctx,
const struct gl_program *prog,
unsigned sampler_count,
struct brw_sampler_prog_key_data *key)
{
struct brw_context *brw = brw_context(ctx);
for (int s = 0; s < sampler_count; s++) {
key->swizzles[s] = SWIZZLE_NOOP;
if (!(prog->SamplersUsed & (1 << s)))
continue;
int unit_id = prog->SamplerUnits[s];
const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
const struct gl_texture_object *t = unit->_Current;
const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
const bool alpha_depth = t->DepthMode == GL_ALPHA &&
(img->_BaseFormat == GL_DEPTH_COMPONENT ||
img->_BaseFormat == GL_DEPTH_STENCIL);
/* Haswell handles texture swizzling as surface format overrides
* (except for GL_ALPHA); all other platforms need MOVs in the shader.
*/
if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
if (brw->gen < 8 &&
sampler->MinFilter != GL_NEAREST &&
sampler->MagFilter != GL_NEAREST) {
if (sampler->WrapS == GL_CLAMP)
key->gl_clamp_mask[0] |= 1 << s;
if (sampler->WrapT == GL_CLAMP)
key->gl_clamp_mask[1] |= 1 << s;
if (sampler->WrapR == GL_CLAMP)
key->gl_clamp_mask[2] |= 1 << s;
}
/* gather4's channel select for green from RG32F is broken; requires
* a shader w/a on IVB; fixable with just SCS on HSW.
*/
if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
if (img->InternalFormat == GL_RG32F)
key->gather_channel_quirk_mask |= 1 << s;
}
/* Gen6's gather4 is broken for UINT/SINT; we treat them as
* UNORM/FLOAT instead and fix it in the shader.
*/
if (brw->gen == 6 && prog->UsesGather) {
key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
}
/* If this is a multisample sampler, and uses the CMS MSAA layout,
* then we need to emit slightly different code to first sample the
* MCS surface.
*/
struct intel_texture_object *intel_tex =
intel_texture_object((struct gl_texture_object *)t);
if (brw->gen >= 7 &&
intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
key->compressed_multisample_layout_mask |= 1 << s;
if (intel_tex->mt->num_samples >= 16) {
assert(brw->gen >= 9);
key->msaa_16 |= 1 << s;
}
}
}
}
}
