*/
static void calculate_curbe_offsets( struct brw_context *brw )
{
- GLcontext *ctx = &brw->intel.ctx;
+ struct gl_context *ctx = &brw->intel.ctx;
/* CACHE_NEW_WM_PROG */
const GLuint nr_fp_regs = (brw->wm.prog_data->nr_params + 15) / 16;
/* _NEW_TRANSFORM */
if (ctx->Transform.ClipPlanesEnabled) {
- GLuint nr_planes = 6 + brw_count_bits(ctx->Transform.ClipPlanesEnabled);
+ GLuint nr_planes = 6 + _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
nr_clip_regs = (nr_planes * 4 + 15) / 16;
}
.brw = BRW_NEW_VERTEX_PROGRAM | BRW_NEW_CONTEXT,
.cache = CACHE_NEW_WM_PROG
},
- .prepare = calculate_curbe_offsets
+ .emit = calculate_curbe_offsets
};
*/
void brw_upload_cs_urb_state(struct brw_context *brw)
{
- struct brw_cs_urb_state cs_urb;
- memset(&cs_urb, 0, sizeof(cs_urb));
+ struct intel_context *intel = &brw->intel;
+ BEGIN_BATCH(2);
/* It appears that this is the state packet for the CS unit, ie. the
* urb entries detailed here are housed in the CS range from the
* URB_FENCE command.
*/
- cs_urb.header.opcode = CMD_CS_URB_STATE;
- cs_urb.header.length = sizeof(cs_urb)/4 - 2;
+ OUT_BATCH(CMD_CS_URB_STATE << 16 | (2-2));
/* BRW_NEW_URB_FENCE */
- cs_urb.bits0.nr_urb_entries = brw->urb.nr_cs_entries;
- cs_urb.bits0.urb_entry_size = brw->urb.csize - 1;
-
- assert(brw->urb.nr_cs_entries);
- BRW_CACHED_BATCH_STRUCT(brw, &cs_urb);
+ if (brw->urb.csize == 0) {
+ OUT_BATCH(0);
+ } else {
+ /* BRW_NEW_URB_FENCE */
+ assert(brw->urb.nr_cs_entries);
+ OUT_BATCH((brw->urb.csize - 1) << 4 | brw->urb.nr_cs_entries);
+ }
+ CACHED_BATCH();
}
static GLfloat fixed_plane[6][4] = {
* cache mechanism, but maybe would benefit from a comparison against
* the current uploaded set of constants.
*/
-static void prepare_constant_buffer(struct brw_context *brw)
+static void
+brw_upload_constant_buffer(struct brw_context *brw)
{
- GLcontext *ctx = &brw->intel.ctx;
+ struct intel_context *intel = &brw->intel;
+ struct gl_context *ctx = &intel->ctx;
const struct brw_vertex_program *vp =
brw_vertex_program_const(brw->vertex_program);
const GLuint sz = brw->curbe.total_size;
const GLuint bufsz = sz * 16 * sizeof(GLfloat);
GLfloat *buf;
GLuint i;
+ gl_clip_plane *clip_planes;
if (sz == 0) {
brw->curbe.last_bufsz = 0;
- return;
+ goto emit;
}
buf = brw->curbe.next_buf;
GLuint offset = brw->curbe.wm_start * 16;
/* copy float constants */
- for (i = 0; i < brw->wm.prog_data->nr_params; i++)
- buf[offset + i] = *brw->wm.prog_data->param[i];
+ for (i = 0; i < brw->wm.prog_data->nr_params; i++) {
+ buf[offset + i] = convert_param(brw->wm.prog_data->param_convert[i],
+ brw->wm.prog_data->param[i]);
+ }
}
- /* The clipplanes are actually delivered to both CLIP and VS units.
- * VS uses them to calculate the outcode bitmasks.
+ /* When using the old VS backend, the clipplanes are actually delivered to
+ * both CLIP and VS units. VS uses them to calculate the outcode bitmasks.
+ *
+ * When using the new VS backend, it is responsible for setting up its own
+ * clipplane constants if it needs them. This results in a slight waste of
+ * of curbe space, but the advantage is that the new VS backend can use its
+ * general-purpose uniform layout code to store the clipplanes.
*/
if (brw->curbe.clip_size) {
GLuint offset = brw->curbe.clip_start * 16;
/* Clip planes: _NEW_TRANSFORM plus _NEW_PROJECTION to get to
* clip-space:
*/
- assert(MAX_CLIP_PLANES == 6);
+ clip_planes = brw_select_clip_planes(ctx);
for (j = 0; j < MAX_CLIP_PLANES; j++) {
if (ctx->Transform.ClipPlanesEnabled & (1<<j)) {
- buf[offset + i * 4 + 0] = ctx->Transform._ClipUserPlane[j][0];
- buf[offset + i * 4 + 1] = ctx->Transform._ClipUserPlane[j][1];
- buf[offset + i * 4 + 2] = ctx->Transform._ClipUserPlane[j][2];
- buf[offset + i * 4 + 3] = ctx->Transform._ClipUserPlane[j][3];
+ buf[offset + i * 4 + 0] = clip_planes[j][0];
+ buf[offset + i * 4 + 1] = clip_planes[j][1];
+ buf[offset + i * 4 + 2] = clip_planes[j][2];
+ buf[offset + i * 4 + 3] = clip_planes[j][3];
i++;
}
}
GLuint offset = brw->curbe.vs_start * 16;
GLuint nr = brw->vs.prog_data->nr_params / 4;
- if (vp->use_const_buffer) {
+ if (brw->vs.prog_data->uses_new_param_layout) {
+ for (i = 0; i < brw->vs.prog_data->nr_params; i++) {
+ buf[offset + i] = *brw->vs.prog_data->param[i];
+ }
+ } else {
/* Load the subset of push constants that will get used when
* we also have a pull constant buffer.
*/
4 * sizeof(float));
}
}
- } else {
- for (i = 0; i < nr; i++) {
- memcpy(buf + offset + i * 4,
- vp->program.Base.Parameters->ParameterValues[i],
- 4 * sizeof(float));
- }
}
}
bufsz);
}
- brw_add_validated_bo(brw, brw->curbe.curbe_bo);
-
/* Because this provokes an action (ie copy the constants into the
* URB), it shouldn't be shortcircuited if identical to the
* previous time - because eg. the urb destination may have
* flushes as necessary when doublebuffering of CURBEs isn't
* possible.
*/
-}
-
-static void emit_constant_buffer(struct brw_context *brw)
-{
- struct intel_context *intel = &brw->intel;
- GLuint sz = brw->curbe.total_size;
+emit:
BEGIN_BATCH(2);
- if (sz == 0) {
+ if (brw->curbe.total_size == 0) {
OUT_BATCH((CMD_CONST_BUFFER << 16) | (2 - 2));
OUT_BATCH(0);
} else {
OUT_BATCH((CMD_CONST_BUFFER << 16) | (1 << 8) | (2 - 2));
OUT_RELOC(brw->curbe.curbe_bo,
I915_GEM_DOMAIN_INSTRUCTION, 0,
- (sz - 1) + brw->curbe.curbe_offset);
+ (brw->curbe.total_size - 1) + brw->curbe.curbe_offset);
}
ADVANCE_BATCH();
}
BRW_NEW_BATCH),
.cache = (CACHE_NEW_WM_PROG)
},
- .prepare = prepare_constant_buffer,
- .emit = emit_constant_buffer,
+ .emit = brw_upload_constant_buffer,
};