static void do_flatshade_line( struct brw_sf_compile *c )
{
struct brw_compile *p = &c->func;
struct intel_context *intel = &p->brw->intel;
struct brw_reg ip = brw_ip_reg();
GLuint nr = brw_count_bits(c->key.attrs & VERT_RESULT_COLOR_BITS);
GLuint jmpi = 1;
if (!nr)
return;
/* Already done in clip program:
*/
if (c->key.primitive == SF_UNFILLED_TRIS)
return;
if (intel->gen == 5)
jmpi = 2;
brw_push_insn_state(p);
brw_MUL(p, c->pv, c->pv, brw_imm_d(jmpi*(nr+1)));
brw_JMPI(p, ip, ip, c->pv);
copy_colors(c, c->vert[1], c->vert[0]);
brw_JMPI(p, ip, ip, brw_imm_ud(jmpi*nr));
copy_colors(c, c->vert[0], c->vert[1]);
brw_pop_insn_state(p);
}
/* Need to use a computed jump to copy flatshaded attributes as the
* vertices are ordered according to y-coordinate before reaching this
* point, so the PV could be anywhere.
*/
static void do_flatshade_triangle( struct brw_sf_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_reg ip = brw_ip_reg();
GLuint nr = brw_count_bits(c->key.attrs & VERT_RESULT_COLOR_BITS);
if (!nr)
return;
/* Already done in clip program:
*/
if (c->key.primitive == SF_UNFILLED_TRIS)
return;
brw_push_insn_state(p);
brw_MUL(p, c->pv, c->pv, brw_imm_ud(nr*2+1));
brw_JMPI(p, ip, ip, c->pv);
copy_colors(c, c->vert[1], c->vert[0]);
copy_colors(c, c->vert[2], c->vert[0]);
brw_JMPI(p, ip, ip, brw_imm_ud(nr*4+1));
copy_colors(c, c->vert[0], c->vert[1]);
copy_colors(c, c->vert[2], c->vert[1]);
brw_JMPI(p, ip, ip, brw_imm_ud(nr*2));
copy_colors(c, c->vert[0], c->vert[2]);
copy_colors(c, c->vert[1], c->vert[2]);
brw_pop_insn_state(p);
}
static void
upload_clip_state(struct brw_context *brw)
{
struct intel_context *intel = &brw->intel;
struct gl_context *ctx = &intel->ctx;
uint32_t depth_clamp = 0;
uint32_t provoking, userclip;
if (!ctx->Transform.DepthClamp)
depth_clamp = GEN6_CLIP_Z_TEST;
/* _NEW_LIGHT */
if (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION) {
provoking =
(0 << GEN6_CLIP_TRI_PROVOKE_SHIFT) |
(1 << GEN6_CLIP_TRIFAN_PROVOKE_SHIFT) |
(0 << GEN6_CLIP_LINE_PROVOKE_SHIFT);
} else {
provoking =
(2 << GEN6_CLIP_TRI_PROVOKE_SHIFT) |
(2 << GEN6_CLIP_TRIFAN_PROVOKE_SHIFT) |
(1 << GEN6_CLIP_LINE_PROVOKE_SHIFT);
}
/* _NEW_TRANSFORM */
userclip = (1 << brw_count_bits(ctx->Transform.ClipPlanesEnabled)) - 1;
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_CLIP << 16 | (4 - 2));
OUT_BATCH(GEN6_CLIP_STATISTICS_ENABLE);
OUT_BATCH(GEN6_CLIP_ENABLE |
GEN6_CLIP_API_OGL |
GEN6_CLIP_MODE_NORMAL |
GEN6_CLIP_XY_TEST |
userclip << GEN6_USER_CLIP_CLIP_DISTANCES_SHIFT |
depth_clamp |
provoking);
OUT_BATCH(U_FIXED(0.125, 3) << GEN6_CLIP_MIN_POINT_WIDTH_SHIFT |
U_FIXED(255.875, 3) << GEN6_CLIP_MAX_POINT_WIDTH_SHIFT |
GEN6_CLIP_FORCE_ZERO_RTAINDEX);
ADVANCE_BATCH();
}
static void brw_upload_vs_prog(struct brw_context *brw)
{
struct gl_context *ctx = &brw->intel.ctx;
struct brw_vs_prog_key key;
struct brw_vertex_program *vp =
(struct brw_vertex_program *)brw->vertex_program;
int i;
memset(&key, 0, sizeof(key));
/* Just upload the program verbatim for now. Always send it all
* the inputs it asks for, whether they are varying or not.
*/
key.program_string_id = vp->id;
key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
key.copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
ctx->Polygon.BackMode != GL_FILL);
key.two_side_color = (ctx->Light.Enabled && ctx->Light.Model.TwoSide);
/* _NEW_POINT */
if (ctx->Point.PointSprite) {
for (i = 0; i < 8; i++) {
if (ctx->Point.CoordReplace[i])
key.point_coord_replace |= (1 << i);
}
}
/* Make an early check for the key.
*/
drm_intel_bo_unreference(brw->vs.prog_bo);
brw->vs.prog_bo = brw_search_cache(&brw->cache, BRW_VS_PROG,
&key, sizeof(key),
NULL, 0,
&brw->vs.prog_data);
if (brw->vs.prog_bo == NULL)
do_vs_prog(brw, vp, &key);
brw->vs.constant_map = ((int8_t *)brw->vs.prog_data +
sizeof(*brw->vs.prog_data));
}
static void compile_gs_prog( struct brw_context *brw,
struct brw_gs_prog_key *key )
{
struct intel_context *intel = &brw->intel;
struct brw_gs_compile c;
const GLuint *program;
void *mem_ctx;
GLuint program_size;
/* Gen6: VF has already converted into polygon, and LINELOOP is
* converted to LINESTRIP at the beginning of the 3D pipeline.
*/
if (intel->gen >= 6)
return;
memset(&c, 0, sizeof(c));
c.key = *key;
/* Need to locate the two positions present in vertex + header.
* These are currently hardcoded:
*/
c.nr_attrs = brw_count_bits(c.key.attrs);
if (intel->gen >= 5)
c.nr_regs = (c.nr_attrs + 1) / 2 + 3; /* are vertices packed, or reg-aligned? */
else
c.nr_regs = (c.nr_attrs + 1) / 2 + 1; /* are vertices packed, or reg-aligned? */
c.nr_bytes = c.nr_regs * REG_SIZE;
mem_ctx = NULL;
/* Begin the compilation:
*/
brw_init_compile(brw, &c.func, mem_ctx);
c.func.single_program_flow = 1;
/* For some reason the thread is spawned with only 4 channels
* unmasked.
*/
brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
/* Note that primitives which don't require a GS program have
* already been weeded out by this stage:
*/
switch (key->primitive) {
case GL_QUADS:
brw_gs_quads( &c, key );
break;
case GL_QUAD_STRIP:
brw_gs_quad_strip( &c, key );
break;
case GL_LINE_LOOP:
brw_gs_lines( &c );
break;
default:
ralloc_free(mem_ctx);
return;
}
/* get the program
*/
program = brw_get_program(&c.func, &program_size);
if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
int i;
printf("gs:\n");
for (i = 0; i < program_size / sizeof(struct brw_instruction); i++)
brw_disasm(stdout, &((struct brw_instruction *)program)[i],
intel->gen);
printf("\n");
}
brw_upload_cache(&brw->cache, BRW_GS_PROG,
&c.key, sizeof(c.key),
program, program_size,
&c.prog_data, sizeof(c.prog_data),
&brw->gs.prog_offset, &brw->gs.prog_data);
ralloc_free(mem_ctx);
}
static void compile_sf_prog( struct brw_context *brw,
struct brw_sf_prog_key *key )
{
GLcontext *ctx = &brw->intel.ctx;
struct brw_sf_compile c;
const GLuint *program;
GLuint program_size;
GLuint i, idx;
memset(&c, 0, sizeof(c));
/* Begin the compilation:
*/
brw_init_compile(brw, &c.func);
c.key = *key;
c.nr_attrs = brw_count_bits(c.key.attrs);
c.nr_attr_regs = (c.nr_attrs+1)/2;
c.nr_setup_attrs = brw_count_bits(c.key.attrs & DO_SETUP_BITS);
c.nr_setup_regs = (c.nr_setup_attrs+1)/2;
c.prog_data.urb_read_length = c.nr_attr_regs;
c.prog_data.urb_entry_size = c.nr_setup_regs * 2;
/* Construct map from attribute number to position in the vertex.
*/
for (i = idx = 0; i < VERT_RESULT_MAX; i++)
if (c.key.attrs & (1<<i)) {
c.attr_to_idx[i] = idx;
c.idx_to_attr[idx] = i;
if (i >= VERT_RESULT_TEX0 && i <= VERT_RESULT_TEX7) {
c.point_attrs[i].CoordReplace =
ctx->Point.CoordReplace[i - VERT_RESULT_TEX0];
}
else {
c.point_attrs[i].CoordReplace = GL_FALSE;
}
idx++;
}
/* Which primitive? Or all three?
*/
switch (key->primitive) {
case SF_TRIANGLES:
c.nr_verts = 3;
brw_emit_tri_setup( &c, GL_TRUE );
break;
case SF_LINES:
c.nr_verts = 2;
brw_emit_line_setup( &c, GL_TRUE );
break;
case SF_POINTS:
c.nr_verts = 1;
if (key->do_point_sprite)
brw_emit_point_sprite_setup( &c, GL_TRUE );
else
brw_emit_point_setup( &c, GL_TRUE );
break;
case SF_UNFILLED_TRIS:
c.nr_verts = 3;
brw_emit_anyprim_setup( &c );
break;
default:
assert(0);
return;
}
/* get the program
*/
program = brw_get_program(&c.func, &program_size);
/* Upload
*/
dri_bo_unreference(brw->sf.prog_bo);
brw->sf.prog_bo = brw_upload_cache( &brw->cache, BRW_SF_PROG,
&c.key, sizeof(c.key),
NULL, 0,
program, program_size,
&c.prog_data,
&brw->sf.prog_data );
}
/* Partition the CURBE between the various users of constant values:
*/
static void calculate_curbe_offsets( struct brw_context *brw )
{
/* CACHE_NEW_WM_PROG */
GLuint nr_fp_regs = (brw->wm.prog_data->nr_params + 15) / 16;
/* BRW_NEW_VERTEX_PROGRAM */
struct brw_vertex_program *vp = (struct brw_vertex_program *)brw->vertex_program;
GLuint nr_vp_regs = (vp->program.Base.Parameters->NumParameters * 4 + 15) / 16;
GLuint nr_clip_regs = 0;
GLuint total_regs;
/* _NEW_TRANSFORM */
if (brw->attribs.Transform->ClipPlanesEnabled) {
GLuint nr_planes = 6 + brw_count_bits(brw->attribs.Transform->ClipPlanesEnabled);
nr_clip_regs = (nr_planes * 4 + 15) / 16;
}
total_regs = nr_fp_regs + nr_vp_regs + nr_clip_regs;
/* This can happen - what to do? Probably rather than falling
* back, the best thing to do is emit programs which code the
* constants as immediate values. Could do this either as a static
* cap on WM and VS, or adaptively.
*
* Unfortunately, this is currently dependent on the results of the
* program generation process (in the case of wm), so this would
* introduce the need to re-generate programs in the event of a
* curbe allocation failure.
*/
/* Max size is 32 - just large enough to
* hold the 128 parameters allowed by
* the fragment and vertex program
* api's. It's not clear what happens
* when both VP and FP want to use 128
* parameters, though.
*/
assert(total_regs <= 32);
/* Lazy resize:
*/
if (nr_fp_regs > brw->curbe.wm_size ||
nr_vp_regs > brw->curbe.vs_size ||
nr_clip_regs > brw->curbe.clip_size ||
(total_regs < brw->curbe.total_size / 4 &&
brw->curbe.total_size > 16)) {
GLuint reg = 0;
/* Calculate a new layout:
*/
reg = 0;
brw->curbe.wm_start = reg;
brw->curbe.wm_size = nr_fp_regs; reg += nr_fp_regs;
brw->curbe.clip_start = reg;
brw->curbe.clip_size = nr_clip_regs; reg += nr_clip_regs;
brw->curbe.vs_start = reg;
brw->curbe.vs_size = nr_vp_regs; reg += nr_vp_regs;
brw->curbe.total_size = reg;
if (0)
_mesa_printf("curbe wm %d+%d clip %d+%d vs %d+%d\n",
brw->curbe.wm_start,
brw->curbe.wm_size,
brw->curbe.clip_start,
brw->curbe.clip_size,
brw->curbe.vs_start,
brw->curbe.vs_size );
brw->state.dirty.brw |= BRW_NEW_CURBE_OFFSETS;
}
}
static void compile_gs_prog( struct brw_context *brw,
struct brw_gs_prog_key *key )
{
struct brw_gs_compile c;
const unsigned *program;
unsigned program_size;
memset(&c, 0, sizeof(c));
c.key = *key;
/* Need to locate the two positions present in vertex + header.
* These are currently hardcoded:
*/
c.nr_attrs = brw_count_bits(c.key.attrs);
c.nr_regs = (c.nr_attrs + 1) / 2 + 1; /* are vertices packed, or reg-aligned? */
c.nr_bytes = c.nr_regs * REG_SIZE;
/* Begin the compilation:
*/
brw_init_compile(&c.func);
c.func.single_program_flow = 1;
/* For some reason the thread is spawned with only 4 channels
* unmasked.
*/
brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
/* Note that primitives which don't require a GS program have
* already been weeded out by this stage:
*/
switch (key->primitive) {
case PIPE_PRIM_QUADS:
brw_gs_quads( &c );
break;
case PIPE_PRIM_QUAD_STRIP:
brw_gs_quad_strip( &c );
break;
case PIPE_PRIM_LINE_LOOP:
brw_gs_lines( &c );
break;
case PIPE_PRIM_LINES:
if (key->hint_gs_always)
brw_gs_lines( &c );
else {
return;
}
break;
case PIPE_PRIM_TRIANGLES:
if (key->hint_gs_always)
brw_gs_tris( &c );
else {
return;
}
break;
case PIPE_PRIM_POINTS:
if (key->hint_gs_always)
brw_gs_points( &c );
else {
return;
}
break;
default:
return;
}
/* get the program
*/
program = brw_get_program(&c.func, &program_size);
/* Upload
*/
brw->gs.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_GS_PROG],
&c.key,
sizeof(c.key),
program,
program_size,
&c.prog_data,
&brw->gs.prog_data );
}
/* Calculate interpolants for triangle and line rasterization.
*/
static void upload_clip_prog(struct brw_context *brw)
{
struct intel_context *intel = &brw->intel;
GLcontext *ctx = &intel->ctx;
struct brw_clip_prog_key key;
memset(&key, 0, sizeof(key));
/* Populate the key:
*/
/* BRW_NEW_REDUCED_PRIMITIVE */
key.primitive = brw->intel.reduced_primitive;
/* CACHE_NEW_VS_PROG */
key.attrs = brw->vs.prog_data->outputs_written;
/* _NEW_LIGHT */
key.do_flat_shading = (ctx->Light.ShadeModel == GL_FLAT);
key.pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
/* _NEW_TRANSFORM */
key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
if (intel->gen == 5)
key.clip_mode = BRW_CLIPMODE_KERNEL_CLIP;
else
key.clip_mode = BRW_CLIPMODE_NORMAL;
/* _NEW_POLYGON */
if (key.primitive == GL_TRIANGLES) {
if (ctx->Polygon.CullFlag &&
ctx->Polygon.CullFaceMode == GL_FRONT_AND_BACK)
key.clip_mode = BRW_CLIPMODE_REJECT_ALL;
else {
GLuint fill_front = CLIP_CULL;
GLuint fill_back = CLIP_CULL;
GLuint offset_front = 0;
GLuint offset_back = 0;
if (!ctx->Polygon.CullFlag ||
ctx->Polygon.CullFaceMode != GL_FRONT) {
switch (ctx->Polygon.FrontMode) {
case GL_FILL:
fill_front = CLIP_FILL;
offset_front = 0;
break;
case GL_LINE:
fill_front = CLIP_LINE;
offset_front = ctx->Polygon.OffsetLine;
break;
case GL_POINT:
fill_front = CLIP_POINT;
offset_front = ctx->Polygon.OffsetPoint;
break;
}
}
if (!ctx->Polygon.CullFlag ||
ctx->Polygon.CullFaceMode != GL_BACK) {
switch (ctx->Polygon.BackMode) {
case GL_FILL:
fill_back = CLIP_FILL;
offset_back = 0;
break;
case GL_LINE:
fill_back = CLIP_LINE;
offset_back = ctx->Polygon.OffsetLine;
break;
case GL_POINT:
fill_back = CLIP_POINT;
offset_back = ctx->Polygon.OffsetPoint;
break;
}
}
if (ctx->Polygon.BackMode != GL_FILL ||
ctx->Polygon.FrontMode != GL_FILL) {
key.do_unfilled = 1;
/* Most cases the fixed function units will handle. Cases where
* one or more polygon faces are unfilled will require help:
*/
key.clip_mode = BRW_CLIPMODE_CLIP_NON_REJECTED;
if (offset_back || offset_front) {
/* _NEW_POLYGON, _NEW_BUFFERS */
key.offset_units = ctx->Polygon.OffsetUnits * brw->intel.polygon_offset_scale;
key.offset_factor = ctx->Polygon.OffsetFactor * ctx->DrawBuffer->_MRD;
}
switch (ctx->Polygon.FrontFace) {
case GL_CCW:
key.fill_ccw = fill_front;
key.fill_cw = fill_back;
key.offset_ccw = offset_front;
key.offset_cw = offset_back;
if (ctx->Light.Model.TwoSide &&
key.fill_cw != CLIP_CULL)
key.copy_bfc_cw = 1;
break;
case GL_CW:
key.fill_cw = fill_front;
key.fill_ccw = fill_back;
key.offset_cw = offset_front;
key.offset_ccw = offset_back;
if (ctx->Light.Model.TwoSide &&
key.fill_ccw != CLIP_CULL)
key.copy_bfc_ccw = 1;
break;
}
}
}
}
drm_intel_bo_unreference(brw->clip.prog_bo);
brw->clip.prog_bo = brw_search_cache(&brw->cache, BRW_CLIP_PROG,
&key, sizeof(key),
NULL, 0,
&brw->clip.prog_data);
if (brw->clip.prog_bo == NULL)
compile_clip_prog( brw, &key );
}
static void
upload_sf_state(struct brw_context *brw)
{
struct intel_context *intel = &brw->intel;
GLcontext *ctx = &intel->ctx;
/* CACHE_NEW_VS_PROG */
uint32_t num_inputs = brw_count_bits(brw->vs.prog_data->outputs_written);
uint32_t num_outputs = brw_count_bits(brw->fragment_program->Base.InputsRead);
uint32_t dw1, dw2, dw3, dw4, dw16;
int i;
/* _NEW_BUFFER */
GLboolean render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
int attr = 0;
dw1 =
num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT |
(num_inputs + 1) / 2 << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
1 << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
dw2 = GEN6_SF_VIEWPORT_TRANSFORM_ENABLE |
GEN6_SF_STATISTICS_ENABLE;
dw3 = 0;
dw4 = 0;
dw16 = 0;
/* _NEW_POLYGON */
if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
dw2 |= GEN6_SF_WINDING_CCW;
if (ctx->Polygon.OffsetFill)
dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
/* _NEW_SCISSOR */
if (ctx->Scissor.Enabled)
dw3 |= GEN6_SF_SCISSOR_ENABLE;
/* _NEW_POLYGON */
if (ctx->Polygon.CullFlag) {
switch (ctx->Polygon.CullFaceMode) {
case GL_FRONT:
dw3 |= GEN6_SF_CULL_FRONT;
break;
case GL_BACK:
dw3 |= GEN6_SF_CULL_BACK;
break;
case GL_FRONT_AND_BACK:
dw3 |= GEN6_SF_CULL_BOTH;
break;
default:
assert(0);
break;
}
} else {
dw3 |= GEN6_SF_CULL_NONE;
}
/* _NEW_LINE */
dw3 |= U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7) <<
GEN6_SF_LINE_WIDTH_SHIFT;
if (ctx->Line.SmoothFlag) {
dw3 |= GEN6_SF_LINE_AA_ENABLE;
dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
}
/* _NEW_POINT */
if (ctx->Point._Attenuated)
dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
dw4 |= U_FIXED(CLAMP(ctx->Point.Size, 0.125, 225.875), 3) <<
GEN6_SF_POINT_WIDTH_SHIFT;
if (ctx->Point.SpriteOrigin == GL_LOWER_LEFT)
dw1 |= GEN6_SF_POINT_SPRITE_LOWERLEFT;
/* _NEW_LIGHT */
if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
dw4 |=
(2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
(2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
(1 << GEN6_SF_LINE_PROVOKE_SHIFT);
} else {
dw4 |=
(1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
}
if (ctx->Point.PointSprite) {
for (i = 0; i < 8; i++) {
if (ctx->Point.CoordReplace[i])
dw16 |= (1 << i);
}
}
BEGIN_BATCH(20);
OUT_BATCH(CMD_3D_SF_STATE << 16 | (20 - 2));
OUT_BATCH(dw1);
OUT_BATCH(dw2);
OUT_BATCH(dw3);
OUT_BATCH(dw4);
OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant. copied from gen4 */
OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
for (i = 0; i < 8; i++) {
uint32_t attr_overrides = 0;
for (; attr < 64; attr++) {
if (brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)) {
attr_overrides |= get_attr_override(brw, attr);
attr++;
break;
}
}
for (; attr < 64; attr++) {
if (brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)) {
attr_overrides |= get_attr_override(brw, attr) << 16;
attr++;
break;
}
}
OUT_BATCH(attr_overrides);
}
OUT_BATCH(dw16); /* point sprite texcoord bitmask */
OUT_BATCH(0); /* constant interp bitmask */
OUT_BATCH(0); /* wrapshortest enables 0-7 */
OUT_BATCH(0); /* wrapshortest enables 8-15 */
ADVANCE_BATCH();
intel_batchbuffer_emit_mi_flush(intel->batch);
}
