if ((dsa->alpha.func != PIPE_FUNC_NEVER) &&
(dsa->alpha.func != PIPE_FUNC_ALWAYS)) {
/* load/splat the alpha reference float value */
- spe_load_float(f, ref_reg, dsa->alpha.ref);
+ spe_load_float(f, ref_reg, dsa->alpha.ref_value);
}
/* emit code to do the alpha comparison, updating 'mask' */
spe_comment(f, 0, "Unpack framebuffer colors, convert to floats");
switch (color_format) {
- case PIPE_FORMAT_A8R8G8B8_UNORM:
+ case PIPE_FORMAT_B8G8R8A8_UNORM:
/* fbB = fbRGBA & mask */
spe_and(f, fbB_reg, fbRGBA_reg, mask0_reg);
spe_roti(f, fbA_reg, fbA_reg, -24);
break;
- case PIPE_FORMAT_B8G8R8A8_UNORM:
+ case PIPE_FORMAT_A8R8G8B8_UNORM:
/* fbA = fbRGBA & mask */
spe_and(f, fbA_reg, fbRGBA_reg, mask0_reg);
int one_reg = -1;
int constR_reg = -1, constG_reg = -1, constB_reg = -1, constA_reg = -1;
- ASSERT(blend->blend_enable);
+ ASSERT(blend->rt[0].blend_enable);
/* packed RGBA -> float colors */
unpack_colors(f, color_format, fbRGBA_reg,
* because in some cases (like PIPE_BLENDFACTOR_ONE and
* PIPE_BLENDFACTOR_ZERO) we can avoid doing unnecessary math.
*/
- switch (blend->rgb_src_factor) {
+ switch (blend->rt[0].rgb_src_factor) {
case PIPE_BLENDFACTOR_ONE:
/* factors = (1,1,1), so term = (R,G,B) */
spe_move(f, term1R_reg, fragR_reg);
* the full term A*factor, not just the factor itself, because
* in many cases we can avoid doing unnecessary multiplies.
*/
- switch (blend->alpha_src_factor) {
+ switch (blend->rt[0].alpha_src_factor) {
case PIPE_BLENDFACTOR_ZERO:
/* factor = 0, so term = 0 */
spe_load_float(f, term1A_reg, 0.0f);
* the full term (Rfb,Gfb,Bfb)*(factor), not just the factor itself, because
* in many cases we can avoid doing unnecessary multiplies.
*/
- switch (blend->rgb_dst_factor) {
+ switch (blend->rt[0].rgb_dst_factor) {
case PIPE_BLENDFACTOR_ONE:
/* factors = (1,1,1), so term = (Rfb,Gfb,Bfb) */
spe_move(f, term2R_reg, fbR_reg);
* the full term Afb*factor, not just the factor itself, because
* in many cases we can avoid doing unnecessary multiplies.
*/
- switch (blend->alpha_dst_factor) {
+ switch (blend->rt[0].alpha_dst_factor) {
case PIPE_BLENDFACTOR_ONE:
/* factor = 1, so term = Afb */
spe_move(f, term2A_reg, fbA_reg);
/*
* Combine Src/Dest RGB terms as per the blend equation.
*/
- switch (blend->rgb_func) {
+ switch (blend->rt[0].rgb_func) {
case PIPE_BLEND_ADD:
spe_fa(f, fragR_reg, term1R_reg, term2R_reg);
spe_fa(f, fragG_reg, term1G_reg, term2G_reg);
/*
* Combine Src/Dest A term
*/
- switch (blend->alpha_func) {
+ switch (blend->rt[0].alpha_func) {
case PIPE_BLEND_ADD:
spe_fa(f, fragA_reg, term1A_reg, term2A_reg);
break;
spe_rotmi(f, a_reg, a_reg, -24);
/* Shift the color bytes according to the surface format */
- if (color_format == PIPE_FORMAT_A8R8G8B8_UNORM) {
+ if (color_format == PIPE_FORMAT_B8G8R8A8_UNORM) {
spe_roti(f, g_reg, g_reg, 8); /* green <<= 8 */
spe_roti(f, r_reg, r_reg, 16); /* red <<= 16 */
spe_roti(f, a_reg, a_reg, 24); /* alpha <<= 24 */
}
- else if (color_format == PIPE_FORMAT_B8G8R8A8_UNORM) {
+ else if (color_format == PIPE_FORMAT_A8R8G8B8_UNORM) {
spe_roti(f, r_reg, r_reg, 8); /* red <<= 8 */
spe_roti(f, g_reg, g_reg, 16); /* green <<= 16 */
spe_roti(f, b_reg, b_reg, 24); /* blue <<= 24 */
* end up, so we can mask them correctly.
*/
switch(color_format) {
- case PIPE_FORMAT_A8R8G8B8_UNORM:
+ case PIPE_FORMAT_B8G8R8A8_UNORM:
/* ARGB */
a_mask = 0xff000000;
r_mask = 0x00ff0000;
g_mask = 0x0000ff00;
b_mask = 0x000000ff;
break;
- case PIPE_FORMAT_B8G8R8A8_UNORM:
+ case PIPE_FORMAT_A8R8G8B8_UNORM:
/* BGRA */
b_mask = 0xff000000;
g_mask = 0x00ff0000;
*/
static void
gen_stencil_test(struct spe_function *f,
- const struct pipe_stencil_state *state,
+ const struct pipe_stencil_state *state,
+ const unsigned ref_value,
uint stencil_max_value,
int fragment_mask_reg,
int fbS_reg,
*/
switch (state->func) {
case PIPE_FUNC_EQUAL:
- if (state->value_mask == stencil_max_value) {
+ if (state->valuemask == stencil_max_value) {
/* stencil_pass = fragment_mask & (s == reference) */
- spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, state->ref_value);
+ spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, ref_value);
spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
}
else {
/* stencil_pass = fragment_mask & ((s&mask) == (reference&mask)) */
uint tmp_masked_stencil = spe_allocate_available_register(f);
- spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
+ spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
spe_compare_equal_uint(f, stencil_pass_reg, tmp_masked_stencil,
- state->value_mask & state->ref_value);
+ state->valuemask & ref_value);
spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
spe_release_register(f, tmp_masked_stencil);
}
break;
case PIPE_FUNC_NOTEQUAL:
- if (state->value_mask == stencil_max_value) {
+ if (state->valuemask == stencil_max_value) {
/* stencil_pass = fragment_mask & ~(s == reference) */
- spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, state->ref_value);
+ spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, ref_value);
spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
}
else {
/* stencil_pass = fragment_mask & ~((s&mask) == (reference&mask)) */
int tmp_masked_stencil = spe_allocate_available_register(f);
- spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
+ spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
spe_compare_equal_uint(f, stencil_pass_reg, tmp_masked_stencil,
- state->value_mask & state->ref_value);
+ state->valuemask & ref_value);
spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
spe_release_register(f, tmp_masked_stencil);
}
break;
case PIPE_FUNC_LESS:
- if (state->value_mask == stencil_max_value) {
+ if (state->valuemask == stencil_max_value) {
/* stencil_pass = fragment_mask & (reference < s) */
- spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg, state->ref_value);
+ spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg, ref_value);
spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
}
else {
/* stencil_pass = fragment_mask & ((reference&mask) < (s & mask)) */
int tmp_masked_stencil = spe_allocate_available_register(f);
- spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
+ spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
spe_compare_greater_uint(f, stencil_pass_reg, tmp_masked_stencil,
- state->value_mask & state->ref_value);
+ state->valuemask & ref_value);
spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
spe_release_register(f, tmp_masked_stencil);
}
break;
case PIPE_FUNC_GREATER:
- if (state->value_mask == stencil_max_value) {
+ if (state->valuemask == stencil_max_value) {
/* stencil_pass = fragment_mask & (reference > s) */
/* There's no convenient Compare Less Than Immediate instruction, so
* we'll have to do this one the harder way, by loading a register and
* treats its operands as unsigned - no sign extension.
*/
int tmp_reg = spe_allocate_available_register(f);
- spe_load_uint(f, tmp_reg, state->ref_value);
+ spe_load_uint(f, tmp_reg, ref_value);
spe_clgt(f, stencil_pass_reg, tmp_reg, fbS_reg);
spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
spe_release_register(f, tmp_reg);
/* stencil_pass = fragment_mask & ((reference&mask) > (s&mask)) */
int tmp_reg = spe_allocate_available_register(f);
int tmp_masked_stencil = spe_allocate_available_register(f);
- spe_load_uint(f, tmp_reg, state->value_mask & state->ref_value);
- spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
+ spe_load_uint(f, tmp_reg, state->valuemask & ref_value);
+ spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
spe_clgt(f, stencil_pass_reg, tmp_reg, tmp_masked_stencil);
spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
spe_release_register(f, tmp_reg);
break;
case PIPE_FUNC_GEQUAL:
- if (state->value_mask == stencil_max_value) {
+ if (state->valuemask == stencil_max_value) {
/* stencil_pass = fragment_mask & (reference >= s)
* = fragment_mask & ~(s > reference) */
spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg,
- state->ref_value);
+ ref_value);
spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
}
else {
/* stencil_pass = fragment_mask & ~((s&mask) > (reference&mask)) */
int tmp_masked_stencil = spe_allocate_available_register(f);
- spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
+ spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
spe_compare_greater_uint(f, stencil_pass_reg, tmp_masked_stencil,
- state->value_mask & state->ref_value);
+ state->valuemask & ref_value);
spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
spe_release_register(f, tmp_masked_stencil);
}
break;
case PIPE_FUNC_LEQUAL:
- if (state->value_mask == stencil_max_value) {
+ if (state->valuemask == stencil_max_value) {
/* stencil_pass = fragment_mask & (reference <= s) ]
* = fragment_mask & ~(reference > s) */
/* As above, we have to do this by loading a register */
int tmp_reg = spe_allocate_available_register(f);
- spe_load_uint(f, tmp_reg, state->ref_value);
+ spe_load_uint(f, tmp_reg, ref_value);
spe_clgt(f, stencil_pass_reg, tmp_reg, fbS_reg);
spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
spe_release_register(f, tmp_reg);
/* stencil_pass = fragment_mask & ~((reference&mask) > (s&mask)) */
int tmp_reg = spe_allocate_available_register(f);
int tmp_masked_stencil = spe_allocate_available_register(f);
- spe_load_uint(f, tmp_reg, state->ref_value & state->value_mask);
- spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->value_mask);
+ spe_load_uint(f, tmp_reg, ref_value & state->valuemask);
+ spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
spe_clgt(f, stencil_pass_reg, tmp_reg, tmp_masked_stencil);
spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
spe_release_register(f, tmp_reg);
static void
gen_get_stencil_values(struct spe_function *f,
const struct pipe_stencil_state *stencil,
+ const unsigned ref_value,
const uint depth_enabled,
int fbS_reg,
int *fail_reg,
}
else {
*fail_reg = spe_allocate_available_register(f);
- gen_stencil_values(f, stencil->fail_op, stencil->ref_value,
+ gen_stencil_values(f, stencil->fail_op, ref_value,
0xff, fbS_reg, *fail_reg);
}
}
else {
*zfail_reg = spe_allocate_available_register(f);
- gen_stencil_values(f, stencil->zfail_op, stencil->ref_value,
+ gen_stencil_values(f, stencil->zfail_op, ref_value,
0xff, fbS_reg, *zfail_reg);
}
}
else {
*zpass_reg = spe_allocate_available_register(f);
- gen_stencil_values(f, stencil->zpass_op, stencil->ref_value,
+ gen_stencil_values(f, stencil->zpass_op, ref_value,
0xff, fbS_reg, *zpass_reg);
}
}
*/
static boolean
gen_stencil_depth_test(struct spe_function *f,
- const struct pipe_depth_stencil_alpha_state *dsa,
+ const struct pipe_depth_stencil_alpha_state *dsa,
+ const struct pipe_stencil_ref *stencil_ref,
const uint facing,
const int mask_reg, const int fragZ_reg,
const int fbZ_reg, const int fbS_reg)
int stencil_writemask_reg;
int zmask_reg;
int newS_reg;
+ unsigned ref_value;
/* Stenciling is quite complex: up to six different configurable stencil
* operations/calculations can be required (three each for front-facing
*/
if (facing == CELL_FACING_BACK && dsa->stencil[1].enabled) {
stencil = &dsa->stencil[1];
+ ref_value = stencil_ref->ref_value[1];
}
else {
stencil = &dsa->stencil[0];
+ ref_value = stencil_ref->ref_value[0];
}
/* Calculate the writemask. If the writemask is trivial (either
need_to_calculate_stencil_values = FALSE;
need_to_writemask_stencil_values = FALSE;
}
- else if (stencil->write_mask == 0x0) {
+ else if (stencil->writemask == 0x0) {
/* All changes are writemasked out, so no need to calculate
* what those changes might be, and no need to write anything back.
*/
need_to_calculate_stencil_values = FALSE;
need_to_writemask_stencil_values = FALSE;
}
- else if (stencil->write_mask == 0xff) {
+ else if (stencil->writemask == 0xff) {
/* Still trivial, but a little less so. We need to write the stencil
* values, but we don't need to mask them.
*/
*/
spe_comment(f, 0, "Computing stencil writemask");
stencil_writemask_reg = spe_allocate_available_register(f);
- spe_load_uint(f, stencil_writemask_reg, dsa->stencil[facing].write_mask);
+ spe_load_uint(f, stencil_writemask_reg, dsa->stencil[facing].writemask);
}
/* At least one-sided stenciling must be on. Generate code that
*/
spe_comment(f, 0, "Running basic stencil test");
stencil_pass_reg = spe_allocate_available_register(f);
- gen_stencil_test(f, stencil, 0xff, mask_reg, fbS_reg, stencil_pass_reg);
+ gen_stencil_test(f, stencil, ref_value, 0xff, mask_reg, fbS_reg, stencil_pass_reg);
/* Generate code that, given the mask of valid fragments and the
* mask of valid fragments that passed the stencil test, computes
spe_comment(f, 0, facing == CELL_FACING_FRONT
? "Computing front-facing stencil values"
: "Computing back-facing stencil values");
- gen_get_stencil_values(f, stencil, dsa->depth.enabled, fbS_reg,
+ gen_get_stencil_values(f, stencil, ref_value, dsa->depth.enabled, fbS_reg,
&stencil_fail_values, &stencil_pass_depth_fail_values,
&stencil_pass_depth_pass_values);
}
static void
gen_depth_stencil(struct cell_context *cell,
const struct pipe_depth_stencil_alpha_state *dsa,
+ const struct pipe_stencil_ref *stencil_ref,
struct spe_function *f,
uint facing,
int mask_reg,
spe_comment(f, 0, "Fetch Z/stencil quad from tile");
switch(zs_format) {
- case PIPE_FORMAT_S8Z24_UNORM: /* fall through */
- case PIPE_FORMAT_X8Z24_UNORM:
+ case PIPE_FORMAT_Z24S8_UNORM: /* fall through */
+ case PIPE_FORMAT_Z24X8_UNORM:
/* prepare mask to extract Z vals from ZS vals */
spe_load_uint(f, zmask_reg, 0x00ffffff);
spe_rotmi(f, fbS_reg, fbZS_reg, -24);
break;
- case PIPE_FORMAT_Z24S8_UNORM: /* fall through */
- case PIPE_FORMAT_Z24X8_UNORM:
+ case PIPE_FORMAT_S8Z24_UNORM: /* fall through */
+ case PIPE_FORMAT_X8Z24_UNORM:
/* convert fragment Z from [0,1] to 32-bit ints */
spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
* gen_stencil_depth_test() function must ignore the
* fbZ_reg register if depth is not enabled.
*/
- write_depth_stencil = gen_stencil_depth_test(f, dsa, facing,
+ write_depth_stencil = gen_stencil_depth_test(f, dsa, stencil_ref, facing,
mask_reg, fragZ_reg,
fbZ_reg, fbS_reg);
}
* fbS_reg has four 8-bit Z values in bits [7..0].
*/
spe_comment(f, 0, "Store quad's depth/stencil values in tile");
- if (zs_format == PIPE_FORMAT_S8Z24_UNORM ||
- zs_format == PIPE_FORMAT_X8Z24_UNORM) {
+ if (zs_format == PIPE_FORMAT_Z24S8_UNORM ||
+ zs_format == PIPE_FORMAT_Z24X8_UNORM) {
spe_shli(f, fbS_reg, fbS_reg, 24); /* fbS = fbS << 24 */
spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */
}
- else if (zs_format == PIPE_FORMAT_Z24S8_UNORM ||
- zs_format == PIPE_FORMAT_Z24X8_UNORM) {
+ else if (zs_format == PIPE_FORMAT_S8Z24_UNORM ||
+ zs_format == PIPE_FORMAT_X8Z24_UNORM) {
spe_shli(f, fbZ_reg, fbZ_reg, 8); /* fbZ = fbZ << 8 */
spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */
}
* code before the fragment shader to cull fragments/quads that are
* totally occluded/discarded.
*
- * XXX we only support PIPE_FORMAT_Z24S8_UNORM z/stencil buffer right now.
+ * XXX we only support PIPE_FORMAT_S8Z24_UNORM z/stencil buffer right now.
*
* See the spu_default_fragment_ops() function to see how the per-fragment
* operations would be done with ordinary C code.
struct spe_function *f)
{
const struct pipe_depth_stencil_alpha_state *dsa = cell->depth_stencil;
+ const struct pipe_stencil_ref *stencil_ref = &cell->stencil_ref;
const struct pipe_blend_state *blend = cell->blend;
const struct pipe_blend_color *blend_color = &cell->blend_color;
const enum pipe_format color_format = cell->framebuffer.cbufs[0]->format;
/* generate depth and/or stencil test code */
if (dsa->depth.enabled || dsa->stencil[0].enabled) {
- gen_depth_stencil(cell, dsa, f,
+ gen_depth_stencil(cell, dsa, stencil_ref, f,
facing,
mask_reg,
depth_tile_reg,
spe_comment(f, 0, "Fetch quad colors from tile");
spe_lqx(f, fbRGBA_reg, color_tile_reg, quad_offset_reg);
- if (blend->blend_enable) {
+ if (blend->rt[0].blend_enable) {
spe_comment(f, 0, "Perform blending");
gen_blend(blend, blend_color, f, color_format,
fragR_reg, fragG_reg, fragB_reg, fragA_reg, fbRGBA_reg);
gen_logicop(blend, f, rgba_reg, fbRGBA_reg);
}
- if (blend->colormask != PIPE_MASK_RGBA) {
+ if (blend->rt[0].colormask != PIPE_MASK_RGBA) {
spe_comment(f, 0, "Compute color mask");
- gen_colormask(f, blend->colormask, color_format, rgba_reg, fbRGBA_reg);
+ gen_colormask(f, blend->rt[0].colormask, color_format, rgba_reg, fbRGBA_reg);
}
/* Mix fragment colors with framebuffer colors using the quad/pixel mask:
projects / mesa.git / blobdiff