*/
#include "pipe/p_defines.h"
+#include "util/u_format.h"
#include "util/u_memory.h"
#include "tgsi/tgsi_scan.h"
#include "sp_context.h"
#include "sp_quad.h"
-#include "sp_surface.h"
#include "sp_quad_pipe.h"
#include "sp_tile_cache.h"
#include "sp_state.h" /* for sp_fragment_shader */
data->bzzzz[j] = tile->data.depth32[y][x];
}
break;
- case PIPE_FORMAT_X8Z24_UNORM:
- case PIPE_FORMAT_S8Z24_UNORM:
+ case PIPE_FORMAT_Z24X8_UNORM:
+ case PIPE_FORMAT_Z24S8_UNORM:
for (j = 0; j < QUAD_SIZE; j++) {
int x = quad->input.x0 % TILE_SIZE + (j & 1);
int y = quad->input.y0 % TILE_SIZE + (j >> 1);
data->stencilVals[j] = tile->data.depth32[y][x] >> 24;
}
break;
- case PIPE_FORMAT_Z24X8_UNORM:
- case PIPE_FORMAT_Z24S8_UNORM:
+ case PIPE_FORMAT_X8Z24_UNORM:
+ case PIPE_FORMAT_S8Z24_UNORM:
for (j = 0; j < QUAD_SIZE; j++) {
int x = quad->input.x0 % TILE_SIZE + (j & 1);
int y = quad->input.y0 % TILE_SIZE + (j >> 1);
}
}
break;
- case PIPE_FORMAT_X8Z24_UNORM:
- case PIPE_FORMAT_S8Z24_UNORM:
+ case PIPE_FORMAT_Z24X8_UNORM:
+ case PIPE_FORMAT_Z24S8_UNORM:
{
float scale = (float) ((1 << 24) - 1);
}
}
break;
- case PIPE_FORMAT_Z24X8_UNORM:
- case PIPE_FORMAT_Z24S8_UNORM:
+ case PIPE_FORMAT_X8Z24_UNORM:
+ case PIPE_FORMAT_S8Z24_UNORM:
{
float scale = (float) ((1 << 24) - 1);
tile->data.depth16[y][x] = (ushort) data->bzzzz[j];
}
break;
- case PIPE_FORMAT_X8Z24_UNORM:
+ case PIPE_FORMAT_Z24X8_UNORM:
case PIPE_FORMAT_Z32_UNORM:
for (j = 0; j < QUAD_SIZE; j++) {
int x = quad->input.x0 % TILE_SIZE + (j & 1);
tile->data.depth32[y][x] = data->bzzzz[j];
}
break;
- case PIPE_FORMAT_S8Z24_UNORM:
+ case PIPE_FORMAT_Z24S8_UNORM:
for (j = 0; j < QUAD_SIZE; j++) {
int x = quad->input.x0 % TILE_SIZE + (j & 1);
int y = quad->input.y0 % TILE_SIZE + (j >> 1);
tile->data.depth32[y][x] = (data->stencilVals[j] << 24) | data->bzzzz[j];
}
break;
- case PIPE_FORMAT_Z24S8_UNORM:
+ case PIPE_FORMAT_S8Z24_UNORM:
for (j = 0; j < QUAD_SIZE; j++) {
int x = quad->input.x0 % TILE_SIZE + (j & 1);
int y = quad->input.y0 % TILE_SIZE + (j >> 1);
tile->data.depth32[y][x] = (data->bzzzz[j] << 8) | data->stencilVals[j];
}
break;
- case PIPE_FORMAT_Z24X8_UNORM:
+ case PIPE_FORMAT_X8Z24_UNORM:
for (j = 0; j < QUAD_SIZE; j++) {
int x = quad->input.x0 % TILE_SIZE + (j & 1);
int y = quad->input.y0 % TILE_SIZE + (j >> 1);
failOp = softpipe->depth_stencil->stencil[face].fail_op;
zFailOp = softpipe->depth_stencil->stencil[face].zfail_op;
zPassOp = softpipe->depth_stencil->stencil[face].zpass_op;
- ref = softpipe->depth_stencil->stencil[face].ref_value;
+ ref = softpipe->stencil_ref.ref_value[face];
wrtMask = softpipe->depth_stencil->stencil[face].writemask;
valMask = softpipe->depth_stencil->stencil[face].valuemask;
+/** helper to get number of Z buffer bits */
+static unsigned
+get_depth_bits(struct quad_stage *qs)
+{
+ struct pipe_surface *zsurf = qs->softpipe->framebuffer.zsbuf;
+ if (zsurf)
+ return util_format_get_component_bits(zsurf->format,
+ UTIL_FORMAT_COLORSPACE_ZS, 0);
+ else
+ return 0;
+}
+
+
+
static void
depth_test_quads_fallback(struct quad_stage *qs,
struct quad_header *quads[],
nr = alpha_test_quads(qs, quads, nr);
}
- if (qs->softpipe->framebuffer.zsbuf &&
+ if (get_depth_bits(qs) > 0 &&
(qs->softpipe->depth_stencil->depth.enabled ||
qs->softpipe->depth_stencil->stencil[0].enabled)) {
qs->next->run(qs->next, quads, nr);
}
-/* XXX: this function assumes setup function actually emits linear
- * spans of quads. It seems a lot more natural to do (early)
- * depth-testing on spans rather than quads.
+
+/**
+ * Special-case Z testing for 16-bit Zbuffer, PIPE_FUNC_LESS and
+ * Z buffer writes enabled.
+ *
+ * NOTE: there's no guarantee that the quads are sequentially side by
+ * side. The fragment shader may have culled some quads, etc. Sliver
+ * triangles may generate non-sequential quads.
*/
static void
depth_interp_z16_less_write(struct quad_stage *qs,
const float z0 = quads[0]->posCoef->a0[2] + dzdx * fx + dzdy * fy;
struct softpipe_cached_tile *tile;
ushort (*depth16)[TILE_SIZE];
- ushort idepth[4], depth_step;
+ ushort init_idepth[4], idepth[4], depth_step;
const float scale = 65535.0;
- idepth[0] = (ushort)((z0) * scale);
- idepth[1] = (ushort)((z0 + dzdx) * scale);
- idepth[2] = (ushort)((z0 + dzdy) * scale);
- idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
+ /* compute scaled depth of the four pixels in first quad */
+ init_idepth[0] = (ushort)((z0) * scale);
+ init_idepth[1] = (ushort)((z0 + dzdx) * scale);
+ init_idepth[2] = (ushort)((z0 + dzdy) * scale);
+ init_idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
- depth_step = (ushort)(dzdx * 2 * scale);
+ depth_step = (ushort)(dzdx * scale);
tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, ix, iy);
- depth16 = (ushort (*)[TILE_SIZE])
- &tile->data.depth16[iy % TILE_SIZE][ix % TILE_SIZE];
-
for (i = 0; i < nr; i++) {
- unsigned outmask = quads[i]->inout.mask;
+ const unsigned outmask = quads[i]->inout.mask;
+ const int dx = quads[i]->input.x0 - ix;
unsigned mask = 0;
-
+
+ /* compute depth for this quad */
+ idepth[0] = init_idepth[0] + dx * depth_step;
+ idepth[1] = init_idepth[1] + dx * depth_step;
+ idepth[2] = init_idepth[2] + dx * depth_step;
+ idepth[3] = init_idepth[3] + dx * depth_step;
+
+ depth16 = (ushort (*)[TILE_SIZE])
+ &tile->data.depth16[iy % TILE_SIZE][(ix + dx)% TILE_SIZE];
+
if ((outmask & 1) && idepth[0] < depth16[0][0]) {
depth16[0][0] = idepth[0];
mask |= (1 << 0);
mask |= (1 << 3);
}
- idepth[0] += depth_step;
- idepth[1] += depth_step;
- idepth[2] += depth_step;
- idepth[3] += depth_step;
-
- depth16 = (ushort (*)[TILE_SIZE]) &depth16[0][2];
-
quads[i]->inout.mask = mask;
if (quads[i]->inout.mask)
quads[pass++] = quads[i];
}
+/**
+ * Special-case Z testing for 16-bit Zbuffer, PIPE_FUNC_LEQUAL and
+ * Z buffer writes enabled.
+ *
+ * NOTE: there's no guarantee that the quads are sequentially side by
+ * side. The fragment shader may have culled some quads, etc. Sliver
+ * triangles may generate non-sequential quads.
+ */
static void
depth_interp_z16_lequal_write(struct quad_stage *qs,
struct quad_header *quads[],
const float z0 = quads[0]->posCoef->a0[2] + dzdx * fx + dzdy * fy;
struct softpipe_cached_tile *tile;
ushort (*depth16)[TILE_SIZE];
- ushort idepth[4], depth_step;
+ ushort init_idepth[4], idepth[4], depth_step;
const float scale = 65535.0;
- idepth[0] = (ushort)((z0) * scale);
- idepth[1] = (ushort)((z0 + dzdx) * scale);
- idepth[2] = (ushort)((z0 + dzdy) * scale);
- idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
+ /* compute scaled depth of the four pixels in first quad */
+ init_idepth[0] = (ushort)((z0) * scale);
+ init_idepth[1] = (ushort)((z0 + dzdx) * scale);
+ init_idepth[2] = (ushort)((z0 + dzdy) * scale);
+ init_idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
- depth_step = (ushort)(dzdx * 2 * scale);
+ depth_step = (ushort)(dzdx * scale);
tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, ix, iy);
- depth16 = (ushort (*)[TILE_SIZE])
- &tile->data.depth16[iy % TILE_SIZE][ix % TILE_SIZE];
-
for (i = 0; i < nr; i++) {
- unsigned outmask = quads[i]->inout.mask;
+ const unsigned outmask = quads[i]->inout.mask;
+ const int dx = quads[i]->input.x0 - ix;
unsigned mask = 0;
+ /* compute depth for this quad */
+ idepth[0] = init_idepth[0] + dx * depth_step;
+ idepth[1] = init_idepth[1] + dx * depth_step;
+ idepth[2] = init_idepth[2] + dx * depth_step;
+ idepth[3] = init_idepth[3] + dx * depth_step;
+
+ depth16 = (ushort (*)[TILE_SIZE])
+ &tile->data.depth16[iy % TILE_SIZE][(ix + dx)% TILE_SIZE];
+
if ((outmask & 1) && idepth[0] <= depth16[0][0]) {
depth16[0][0] = idepth[0];
mask |= (1 << 0);
mask |= (1 << 3);
}
- idepth[0] += depth_step;
- idepth[1] += depth_step;
- idepth[2] += depth_step;
- idepth[3] += depth_step;
-
depth16 = (ushort (*)[TILE_SIZE]) &depth16[0][2];
quads[i]->inout.mask = mask;
boolean alpha = qs->softpipe->depth_stencil->alpha.enabled;
- boolean depth = (qs->softpipe->framebuffer.zsbuf &&
+ boolean depth = (get_depth_bits(qs) > 0 &&
qs->softpipe->depth_stencil->depth.enabled);
unsigned depthfunc = qs->softpipe->depth_stencil->depth.func;
boolean occlusion = qs->softpipe->active_query_count;
-
if (!alpha &&
!depth &&
!stencil) {
projects / mesa.git / blobdiff