#include "brw_context.h"
#include "brw_state.h"
#include "brw_defines.h"
-#include "macros.h"
+#include "main/macros.h"
#include "intel_fbo.h"
static void upload_sf_vp(struct brw_context *brw)
GLcontext *ctx = &brw->intel.ctx;
const GLfloat depth_scale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
struct brw_sf_viewport sfv;
- struct intel_renderbuffer *irb =
- intel_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[0][0]);
GLfloat y_scale, y_bias;
+ const GLboolean render_to_fbo = (ctx->DrawBuffer->Name != 0);
memset(&sfv, 0, sizeof(sfv));
- if (ctx->DrawBuffer->Name) {
- /* User-created FBO */
- if (irb && !irb->RenderToTexture) {
- y_scale = -1.0;
- y_bias = ctx->DrawBuffer->Height;
- } else {
- y_scale = 1.0;
- y_bias = 0;
- }
- } else {
+ if (render_to_fbo) {
+ y_scale = 1.0;
+ y_bias = 0;
+ }
+ else {
y_scale = -1.0;
y_bias = ctx->DrawBuffer->Height;
}
- /* _NEW_VIEWPORT, BRW_NEW_METAOPS */
-
- if (!brw->metaops.active) {
- const GLfloat *v = ctx->Viewport._WindowMap.m;
-
- sfv.viewport.m00 = v[MAT_SX];
- sfv.viewport.m11 = v[MAT_SY] * y_scale;
- sfv.viewport.m22 = v[MAT_SZ] * depth_scale;
- sfv.viewport.m30 = v[MAT_TX];
- sfv.viewport.m31 = v[MAT_TY] * y_scale + y_bias;
- sfv.viewport.m32 = v[MAT_TZ] * depth_scale;
- } else {
- sfv.viewport.m00 = 1;
- sfv.viewport.m11 = - 1;
- sfv.viewport.m22 = 1;
- sfv.viewport.m30 = 0;
- sfv.viewport.m31 = ctx->DrawBuffer->Height;
- sfv.viewport.m32 = 0;
- }
+ /* _NEW_VIEWPORT */
+
+ const GLfloat *v = ctx->Viewport._WindowMap.m;
+
+ sfv.viewport.m00 = v[MAT_SX];
+ sfv.viewport.m11 = v[MAT_SY] * y_scale;
+ sfv.viewport.m22 = v[MAT_SZ] * depth_scale;
+ sfv.viewport.m30 = v[MAT_TX];
+ sfv.viewport.m31 = v[MAT_TY] * y_scale + y_bias;
+ sfv.viewport.m32 = v[MAT_TZ] * depth_scale;
/* _NEW_SCISSOR */
* Note that the hardware's coordinates are inclusive, while Mesa's min is
* inclusive but max is exclusive.
*/
- sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
- sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
- sfv.scissor.ymin = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymax;
- sfv.scissor.ymax = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymin - 1;
+ if (render_to_fbo) {
+ /* texmemory: Y=0=bottom */
+ sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
+ sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
+ sfv.scissor.ymin = ctx->DrawBuffer->_Ymin;
+ sfv.scissor.ymax = ctx->DrawBuffer->_Ymax - 1;
+ }
+ else {
+ /* memory: Y=0=top */
+ sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
+ sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
+ sfv.scissor.ymin = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymax;
+ sfv.scissor.ymax = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymin - 1;
+ }
dri_bo_unreference(brw->sf.vp_bo);
brw->sf.vp_bo = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0 );
.dirty = {
.mesa = (_NEW_VIEWPORT |
_NEW_SCISSOR),
- .brw = BRW_NEW_METAOPS,
+ .brw = 0,
.cache = 0
},
- .update = upload_sf_vp
+ .prepare = upload_sf_vp
};
+struct brw_sf_unit_key {
+ unsigned int total_grf;
+ unsigned int urb_entry_read_length;
+ unsigned int nr_urb_entries, urb_size, sfsize;
-static void upload_sf_unit( struct brw_context *brw )
+ GLenum front_face, cull_face;
+ unsigned scissor:1;
+ unsigned line_smooth:1;
+ unsigned point_sprite:1;
+ unsigned point_attenuated:1;
+ unsigned render_to_fbo:1;
+ float line_width;
+ float point_size;
+};
+
+static void
+sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key)
+{
+ GLcontext *ctx = &brw->intel.ctx;
+ memset(key, 0, sizeof(*key));
+
+ /* CACHE_NEW_SF_PROG */
+ key->total_grf = brw->sf.prog_data->total_grf;
+ key->urb_entry_read_length = brw->sf.prog_data->urb_read_length;
+
+ /* BRW_NEW_URB_FENCE */
+ key->nr_urb_entries = brw->urb.nr_sf_entries;
+ key->urb_size = brw->urb.vsize;
+ key->sfsize = brw->urb.sfsize;
+
+ key->scissor = ctx->Scissor.Enabled;
+ key->front_face = ctx->Polygon.FrontFace;
+
+ if (ctx->Polygon.CullFlag)
+ key->cull_face = ctx->Polygon.CullFaceMode;
+ else
+ key->cull_face = GL_NONE;
+
+ key->line_width = ctx->Line.Width;
+ key->line_smooth = ctx->Line.SmoothFlag;
+
+ key->point_sprite = ctx->Point.PointSprite;
+ key->point_size = ctx->Point.Size;
+ key->point_attenuated = ctx->Point._Attenuated;
+
+ key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
+}
+
+static dri_bo *
+sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
+ dri_bo **reloc_bufs)
{
struct brw_sf_unit_state sf;
+ dri_bo *bo;
+
memset(&sf, 0, sizeof(sf));
- dri_bo *reloc_bufs[2];
- /* CACHE_NEW_SF_PROG */
- sf.thread0.grf_reg_count = ALIGN(brw->sf.prog_data->total_grf, 16) / 16 - 1;
+ sf.thread0.grf_reg_count = ALIGN(key->total_grf, 16) / 16 - 1;
sf.thread0.kernel_start_pointer = brw->sf.prog_bo->offset >> 6; /* reloc */
- sf.thread3.urb_entry_read_length = brw->sf.prog_data->urb_read_length;
sf.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
+
sf.thread3.dispatch_grf_start_reg = 3;
sf.thread3.urb_entry_read_offset = 1;
+ sf.thread3.urb_entry_read_length = key->urb_entry_read_length;
- /* BRW_NEW_URB_FENCE */
- sf.thread4.nr_urb_entries = brw->urb.nr_sf_entries;
- sf.thread4.urb_entry_allocation_size = brw->urb.sfsize - 1;
- sf.thread4.max_threads = MIN2(12, brw->urb.nr_sf_entries / 2) - 1;
+ sf.thread4.nr_urb_entries = key->nr_urb_entries;
+ sf.thread4.urb_entry_allocation_size = key->sfsize - 1;
+ /* Each SF thread produces 1 PUE, and there can be up to 24 threads */
+ sf.thread4.max_threads = MIN2(24, key->nr_urb_entries) - 1;
if (INTEL_DEBUG & DEBUG_SINGLE_THREAD)
- sf.thread4.max_threads = 0;
+ sf.thread4.max_threads = 0;
if (INTEL_DEBUG & DEBUG_STATS)
- sf.thread4.stats_enable = 1;
+ sf.thread4.stats_enable = 1;
/* CACHE_NEW_SF_VP */
sf.sf5.sf_viewport_state_offset = brw->sf.vp_bo->offset >> 5; /* reloc */
-
+
sf.sf5.viewport_transform = 1;
-
+
/* _NEW_SCISSOR */
- if (brw->attribs.Scissor->Enabled)
- sf.sf6.scissor = 1;
+ if (key->scissor)
+ sf.sf6.scissor = 1;
/* _NEW_POLYGON */
- if (brw->attribs.Polygon->FrontFace == GL_CCW)
+ if (key->front_face == GL_CCW)
sf.sf5.front_winding = BRW_FRONTWINDING_CCW;
else
sf.sf5.front_winding = BRW_FRONTWINDING_CW;
- if (brw->attribs.Polygon->CullFlag) {
- switch (brw->attribs.Polygon->CullFaceMode) {
- case GL_FRONT:
- sf.sf6.cull_mode = BRW_CULLMODE_FRONT;
- break;
- case GL_BACK:
- sf.sf6.cull_mode = BRW_CULLMODE_BACK;
- break;
- case GL_FRONT_AND_BACK:
- sf.sf6.cull_mode = BRW_CULLMODE_BOTH;
- break;
- default:
- assert(0);
- break;
- }
- }
- else
+ /* The viewport is inverted for rendering to a FBO, and that inverts
+ * polygon front/back orientation.
+ */
+ sf.sf5.front_winding ^= key->render_to_fbo;
+
+ switch (key->cull_face) {
+ case GL_FRONT:
+ sf.sf6.cull_mode = BRW_CULLMODE_FRONT;
+ break;
+ case GL_BACK:
+ sf.sf6.cull_mode = BRW_CULLMODE_BACK;
+ break;
+ case GL_FRONT_AND_BACK:
+ sf.sf6.cull_mode = BRW_CULLMODE_BOTH;
+ break;
+ case GL_NONE:
sf.sf6.cull_mode = BRW_CULLMODE_NONE;
-
+ break;
+ default:
+ assert(0);
+ break;
+ }
/* _NEW_LINE */
/* XXX use ctx->Const.Min/MaxLineWidth here */
- sf.sf6.line_width = CLAMP(brw->attribs.Line->Width, 1.0, 5.0) * (1<<1);
+ sf.sf6.line_width = CLAMP(key->line_width, 1.0, 5.0) * (1<<1);
sf.sf6.line_endcap_aa_region_width = 1;
- if (brw->attribs.Line->SmoothFlag)
+ if (key->line_smooth)
sf.sf6.aa_enable = 1;
- else if (sf.sf6.line_width <= 0x2)
- sf.sf6.line_width = 0;
+ else if (sf.sf6.line_width <= 0x2)
+ sf.sf6.line_width = 0;
/* _NEW_POINT */
- sf.sf6.point_rast_rule = 1; /* opengl conventions */
+ sf.sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT; /* opengl conventions */
/* XXX clamp max depends on AA vs. non-AA */
- sf.sf7.sprite_point = brw->attribs.Point->PointSprite;
- sf.sf7.point_size = CLAMP(brw->attribs.Point->Size, 1.0, 255.0) * (1<<3);
- sf.sf7.use_point_size_state = !brw->attribs.Point->_Attenuated;
-
+ sf.sf7.sprite_point = key->point_sprite;
+ sf.sf7.point_size = CLAMP(rint(key->point_size), 1, 255) * (1<<3);
+ sf.sf7.use_point_size_state = !key->point_attenuated;
+ sf.sf7.aa_line_distance_mode = 0;
+
/* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
*/
sf.sf7.trifan_pv = 2;
sf.sf6.dest_org_vbias = 0x8;
sf.sf6.dest_org_hbias = 0x8;
- reloc_bufs[0] = brw->sf.prog_bo;
- reloc_bufs[1] = brw->sf.vp_bo;
+ bo = brw_upload_cache(&brw->cache, BRW_SF_UNIT,
+ key, sizeof(*key),
+ reloc_bufs, 2,
+ &sf, sizeof(sf),
+ NULL, NULL);
+
+ /* Emit SF program relocation */
+ dri_bo_emit_reloc(bo,
+ I915_GEM_DOMAIN_INSTRUCTION, 0,
+ sf.thread0.grf_reg_count << 1,
+ offsetof(struct brw_sf_unit_state, thread0),
+ brw->sf.prog_bo);
- brw->sf.thread0_delta = sf.thread0.grf_reg_count << 1;
- brw->sf.sf5_delta = sf.sf5.front_winding | (sf.sf5.viewport_transform << 1);
+ /* Emit SF viewport relocation */
+ dri_bo_emit_reloc(bo,
+ I915_GEM_DOMAIN_INSTRUCTION, 0,
+ sf.sf5.front_winding | (sf.sf5.viewport_transform << 1),
+ offsetof(struct brw_sf_unit_state, sf5),
+ brw->sf.vp_bo);
- dri_bo_unreference(brw->sf.state_bo);
- brw->sf.state_bo = brw_cache_data( &brw->cache, BRW_SF_UNIT, &sf,
- reloc_bufs, 2 );
+ return bo;
}
-static void emit_reloc_sf_unit(struct brw_context *brw)
+static void upload_sf_unit( struct brw_context *brw )
{
- /* Emit SF program relocation */
- dri_emit_reloc(brw->sf.state_bo,
- DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
- brw->sf.thread0_delta,
- offsetof(struct brw_sf_unit_state, thread0),
- brw->sf.prog_bo);
+ struct brw_sf_unit_key key;
+ dri_bo *reloc_bufs[2];
- /* Emit SF viewport relocation */
- dri_emit_reloc(brw->sf.state_bo,
- DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
- brw->sf.sf5_delta,
- offsetof(struct brw_sf_unit_state, sf5),
- brw->sf.vp_bo);
+ sf_unit_populate_key(brw, &key);
+
+ reloc_bufs[0] = brw->sf.prog_bo;
+ reloc_bufs[1] = brw->sf.vp_bo;
+
+ dri_bo_unreference(brw->sf.state_bo);
+ brw->sf.state_bo = brw_search_cache(&brw->cache, BRW_SF_UNIT,
+ &key, sizeof(key),
+ reloc_bufs, 2,
+ NULL);
+ if (brw->sf.state_bo == NULL) {
+ brw->sf.state_bo = sf_unit_create_from_key(brw, &key, reloc_bufs);
+ }
}
const struct brw_tracked_state brw_sf_unit = {
_NEW_LINE |
_NEW_POINT |
_NEW_SCISSOR),
- .brw = (BRW_NEW_URB_FENCE |
- BRW_NEW_METAOPS),
+ .brw = BRW_NEW_URB_FENCE,
.cache = (CACHE_NEW_SF_VP |
CACHE_NEW_SF_PROG)
},
- .update = upload_sf_unit,
- .emit_reloc = emit_reloc_sf_unit,
+ .prepare = upload_sf_unit,
};
projects / mesa.git / blobdiff