void brw_copy_indirect_to_indirect(struct brw_codegen *p, struct brw_indirect dst_ptr, struct brw_indirect src_ptr, unsigned count) { unsigned i; for (i = 0; i < count; i++) { unsigned delta = i*32; brw_MOV(p, deref_4f(dst_ptr, delta), deref_4f(src_ptr, delta)); brw_MOV(p, deref_4f(dst_ptr, delta+16), deref_4f(src_ptr, delta+16)); } }
/* Interpolate between two vertices and put the result into a0.0. * Increment a0.0 accordingly. */ void brw_clip_interp_vertex( struct brw_clip_compile *c, struct brw_indirect dest_ptr, struct brw_indirect v0_ptr, /* from */ struct brw_indirect v1_ptr, /* to */ struct brw_reg t0, GLboolean force_edgeflag) { struct brw_compile *p = &c->func; struct brw_reg tmp = get_tmp(c); GLuint i; /* Just copy the vertex header: */ brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1); /* Iterate over each attribute (could be done in pairs?) */ for (i = 0; i < c->nr_attrs; i++) { GLuint delta = i*16 + 32; if (delta == c->offset[VERT_RESULT_EDGE]) { if (force_edgeflag) brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1)); else brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta)); } else { /* Interpolate: * * New = attr0 + t*attr1 - t*attr0 */ brw_MUL(p, vec4(brw_null_reg()), deref_4f(v1_ptr, delta), t0); brw_MAC(p, tmp, negate(deref_4f(v0_ptr, delta)), t0); brw_ADD(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta), tmp); } } if (i & 1) { GLuint delta = i*16 + 32; brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0)); } release_tmp(c, tmp); /* Recreate the projected (NDC) coordinate in the new vertex * header: */ brw_clip_project_vertex(c, dest_ptr ); }
void brw_copy_indirect_to_indirect(struct brw_compile *p, struct brw_indirect dst_ptr, struct brw_indirect src_ptr, GLuint count) { GLuint i; for (i = 0; i < count; i++) { GLuint delta = i*32; brw_MOV(p, deref_4f(dst_ptr, delta), deref_4f(src_ptr, delta)); brw_MOV(p, deref_4f(dst_ptr, delta+16), deref_4f(src_ptr, delta+16)); } }
static void brw_clip_project_vertex( struct brw_clip_compile *c, struct brw_indirect vert_addr ) { struct brw_compile *p = &c->func; struct brw_reg tmp = get_tmp(c); /* Fixup position. Extract from the original vertex and re-project * to screen space: */ brw_MOV(p, tmp, deref_4f(vert_addr, c->offset[VERT_RESULT_HPOS])); brw_clip_project_position(c, tmp); brw_MOV(p, deref_4f(vert_addr, c->header_position_offset), tmp); release_tmp(c, tmp); }
void brw_copy_from_indirect(struct brw_compile *p, struct brw_reg dst, struct brw_indirect ptr, GLuint count) { GLuint i; dst = vec4(dst); for (i = 0; i < count; i++) { GLuint delta = i*32; brw_MOV(p, byte_offset(dst, delta), deref_4f(ptr, delta)); brw_MOV(p, byte_offset(dst, delta+16), deref_4f(ptr, delta+16)); } }
void brw_copy_from_indirect(struct brw_codegen *p, struct brw_reg dst, struct brw_indirect ptr, unsigned count) { unsigned i; dst = vec4(dst); for (i = 0; i < count; i++) { unsigned delta = i*32; brw_MOV(p, byte_offset(dst, delta), deref_4f(ptr, delta)); brw_MOV(p, byte_offset(dst, delta+16), deref_4f(ptr, delta+16)); } }
static void brw_clip_project_vertex( struct brw_clip_compile *c, struct brw_indirect vert_addr ) { struct brw_codegen *p = &c->func; struct brw_reg tmp = get_tmp(c); GLuint hpos_offset = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS); GLuint ndc_offset = brw_varying_to_offset(&c->vue_map, BRW_VARYING_SLOT_NDC); /* Fixup position. Extract from the original vertex and re-project * to screen space: */ brw_MOV(p, tmp, deref_4f(vert_addr, hpos_offset)); brw_clip_project_position(c, tmp); brw_MOV(p, deref_4f(vert_addr, ndc_offset), tmp); release_tmp(c, tmp); }
static void apply_one_offset( struct brw_clip_compile *c, struct brw_indirect vert ) { struct brw_compile *p = &c->func; struct brw_reg pos = deref_4f(vert, c->offset[VERT_RESULT_HPOS]); struct brw_reg z = get_element(pos, 2); brw_ADD(p, z, z, vec1(c->reg.offset)); }
/** * Loads the clip distance for a vertex into `dst`, and ends with * a comparison of it to zero with the condition `cond`. * * - If using a fixed plane, the distance is dot(hpos, plane). * - If using a user clip plane, the distance is directly available in the vertex. */ static inline void load_clip_distance(struct brw_clip_compile *c, struct brw_indirect vtx, struct brw_reg dst, GLuint hpos_offset, int cond) { struct brw_codegen *p = &c->func; dst = vec4(dst); brw_AND(p, vec1(brw_null_reg()), c->reg.vertex_src_mask, brw_imm_ud(1)); brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ); brw_IF(p, BRW_EXECUTE_1); { struct brw_indirect temp_ptr = brw_indirect(7, 0); brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx), c->reg.clipdistance_offset); brw_MOV(p, vec1(dst), deref_1f(temp_ptr, 0)); } brw_ELSE(p); { brw_MOV(p, dst, deref_4f(vtx, hpos_offset)); brw_DP4(p, dst, dst, c->reg.plane_equation); } brw_ENDIF(p); brw_CMP(p, brw_null_reg(), cond, vec1(dst), brw_imm_f(0.0f)); }
static void brw_clip_test( struct brw_clip_compile *c ) { struct brw_reg t = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); struct brw_reg t1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); struct brw_reg t2 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); struct brw_reg t3 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); struct brw_reg v0 = get_tmp(c); struct brw_reg v1 = get_tmp(c); struct brw_reg v2 = get_tmp(c); struct brw_indirect vt0 = brw_indirect(0, 0); struct brw_indirect vt1 = brw_indirect(1, 0); struct brw_indirect vt2 = brw_indirect(2, 0); struct brw_compile *p = &c->func; struct brw_instruction *is_outside; struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */ brw_MOV(p, get_addr_reg(vt0), brw_address(c->reg.vertex[0])); brw_MOV(p, get_addr_reg(vt1), brw_address(c->reg.vertex[1])); brw_MOV(p, get_addr_reg(vt2), brw_address(c->reg.vertex[2])); brw_MOV(p, v0, deref_4f(vt0, c->offset_hpos)); brw_MOV(p, v1, deref_4f(vt1, c->offset_hpos)); brw_MOV(p, v2, deref_4f(vt2, c->offset_hpos)); brw_AND(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(~0x3f)); /* test nearz, xmin, ymin plane */ /* clip.xyz < -clip.w */ brw_CMP(p, t1, BRW_CONDITIONAL_L, v0, negate(get_element(v0, 3))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_CMP(p, t2, BRW_CONDITIONAL_L, v1, negate(get_element(v1, 3))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_CMP(p, t3, BRW_CONDITIONAL_L, v2, negate(get_element(v2, 3))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); /* All vertices are outside of a plane, rejected */ brw_AND(p, t, t1, t2); brw_AND(p, t, t, t3); brw_OR(p, tmp0, get_element(t, 0), get_element(t, 1)); brw_OR(p, tmp0, tmp0, get_element(t, 2)); brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_AND(p, brw_null_reg(), tmp0, brw_imm_ud(0x1)); is_outside = brw_IF(p, BRW_EXECUTE_1); { brw_clip_kill_thread(c); } brw_ENDIF(p, is_outside); brw_set_predicate_control(p, BRW_PREDICATE_NONE); /* some vertices are inside a plane, some are outside,need to clip */ brw_XOR(p, t, t1, t2); brw_XOR(p, t1, t2, t3); brw_OR(p, t, t, t1); brw_AND(p, t, t, brw_imm_ud(0x1)); brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, get_element(t, 0), brw_imm_ud(0)); brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<5))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, get_element(t, 1), brw_imm_ud(0)); brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<3))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, get_element(t, 2), brw_imm_ud(0)); brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<1))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); /* test farz, xmax, ymax plane */ /* clip.xyz > clip.w */ brw_CMP(p, t1, BRW_CONDITIONAL_G, v0, get_element(v0, 3)); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_CMP(p, t2, BRW_CONDITIONAL_G, v1, get_element(v1, 3)); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_CMP(p, t3, BRW_CONDITIONAL_G, v2, get_element(v2, 3)); brw_set_predicate_control(p, BRW_PREDICATE_NONE); /* All vertices are outside of a plane, rejected */ brw_AND(p, t, t1, t2); brw_AND(p, t, t, t3); brw_OR(p, tmp0, get_element(t, 0), get_element(t, 1)); brw_OR(p, tmp0, tmp0, get_element(t, 2)); brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_AND(p, brw_null_reg(), tmp0, brw_imm_ud(0x1)); is_outside = brw_IF(p, BRW_EXECUTE_1); { brw_clip_kill_thread(c); } brw_ENDIF(p, is_outside); brw_set_predicate_control(p, BRW_PREDICATE_NONE); /* some vertices are inside a plane, some are outside,need to clip */ brw_XOR(p, t, t1, t2); brw_XOR(p, t1, t2, t3); brw_OR(p, t, t, t1); brw_AND(p, t, t, brw_imm_ud(0x1)); brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, get_element(t, 0), brw_imm_ud(0)); brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<4))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, get_element(t, 1), brw_imm_ud(0)); brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<2))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, get_element(t, 2), brw_imm_ud(0)); brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<0))); brw_set_predicate_control(p, BRW_PREDICATE_NONE); release_tmps(c); }
/* Use mesa's clipping algorithms, translated to GEN4 assembly. */ void brw_clip_tri( struct brw_clip_compile *c ) { struct brw_compile *p = &c->func; struct brw_indirect vtx = brw_indirect(0, 0); struct brw_indirect vtxPrev = brw_indirect(1, 0); struct brw_indirect vtxOut = brw_indirect(2, 0); struct brw_indirect plane_ptr = brw_indirect(3, 0); struct brw_indirect inlist_ptr = brw_indirect(4, 0); struct brw_indirect outlist_ptr = brw_indirect(5, 0); struct brw_indirect freelist_ptr = brw_indirect(6, 0); struct brw_instruction *plane_loop; struct brw_instruction *plane_active; struct brw_instruction *vertex_loop; struct brw_instruction *next_test; struct brw_instruction *prev_test; brw_MOV(p, get_addr_reg(vtxPrev), brw_address(c->reg.vertex[2]) ); brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c)); brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist)); brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist)); brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3]) ); plane_loop = brw_DO(p, BRW_EXECUTE_1); { /* if (planemask & 1) */ brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_AND(p, vec1(brw_null_reg()), c->reg.planemask, brw_imm_ud(1)); plane_active = brw_IF(p, BRW_EXECUTE_1); { /* vtxOut = freelist_ptr++ */ brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(freelist_ptr) ); brw_ADD(p, get_addr_reg(freelist_ptr), get_addr_reg(freelist_ptr), brw_imm_uw(c->nr_regs * REG_SIZE)); if (c->key.nr_userclip) brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0)); else brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0)); brw_MOV(p, c->reg.loopcount, c->reg.nr_verts); brw_MOV(p, c->reg.nr_verts, brw_imm_ud(0)); vertex_loop = brw_DO(p, BRW_EXECUTE_1); { /* vtx = *input_ptr; */ brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0)); /* IS_NEGATIVE(prev) */ brw_set_conditionalmod(p, BRW_CONDITIONAL_L); brw_DP4(p, vec4(c->reg.dpPrev), deref_4f(vtxPrev, c->offset_hpos), c->reg.plane_equation); prev_test = brw_IF(p, BRW_EXECUTE_1); { /* IS_POSITIVE(next) */ brw_set_conditionalmod(p, BRW_CONDITIONAL_GE); brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset_hpos), c->reg.plane_equation); next_test = brw_IF(p, BRW_EXECUTE_1); { /* Coming back in. */ brw_ADD(p, c->reg.t, c->reg.dpPrev, negate(c->reg.dp)); brw_math_invert(p, c->reg.t, c->reg.t); brw_MUL(p, c->reg.t, c->reg.t, c->reg.dpPrev); /* If (vtxOut == 0) vtxOut = vtxPrev */ brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) ); brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtxPrev) ); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_clip_interp_vertex(c, vtxOut, vtxPrev, vtx, c->reg.t, GL_FALSE); /* *outlist_ptr++ = vtxOut; * nr_verts++; * vtxOut = 0; */ brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut)); brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) ); } brw_ENDIF(p, next_test); } prev_test = brw_ELSE(p, prev_test); { /* *outlist_ptr++ = vtxPrev; * nr_verts++; */ brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxPrev)); brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); /* IS_NEGATIVE(next) */ brw_set_conditionalmod(p, BRW_CONDITIONAL_L); brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset_hpos), c->reg.plane_equation); next_test = brw_IF(p, BRW_EXECUTE_1); { /* Going out of bounds. Avoid division by zero as we * know dp != dpPrev from DIFFERENT_SIGNS, above. */ brw_ADD(p, c->reg.t, c->reg.dp, negate(c->reg.dpPrev)); brw_math_invert(p, c->reg.t, c->reg.t); brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp); /* If (vtxOut == 0) vtxOut = vtx */ brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) ); brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtx) ); brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_clip_interp_vertex(c, vtxOut, vtx, vtxPrev, c->reg.t, GL_TRUE); /* *outlist_ptr++ = vtxOut; * nr_verts++; * vtxOut = 0; */ brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut)); brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) ); } brw_ENDIF(p, next_test); } brw_ENDIF(p, prev_test); /* vtxPrev = vtx; * inlist_ptr++; */ brw_MOV(p, get_addr_reg(vtxPrev), get_addr_reg(vtx)); brw_ADD(p, get_addr_reg(inlist_ptr), get_addr_reg(inlist_ptr), brw_imm_uw(sizeof(short))); /* while (--loopcount != 0) */ brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1)); } brw_WHILE(p, vertex_loop); /* vtxPrev = *(outlist_ptr-1) OR: outlist[nr_verts-1] * inlist = outlist * inlist_ptr = &inlist[0] * outlist_ptr = &outlist[0] */ brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_w(-2)); brw_MOV(p, get_addr_reg(vtxPrev), deref_1uw(outlist_ptr, 0)); brw_MOV(p, brw_vec8_grf(c->reg.inlist.nr, 0), brw_vec8_grf(c->reg.outlist.nr, 0)); brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist)); brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist)); } brw_ENDIF(p, plane_active); /* plane_ptr++; */ brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c)); /* nr_verts >= 3 */ brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_GE, c->reg.nr_verts, brw_imm_ud(3)); /* && (planemask>>=1) != 0 */ brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1)); } brw_WHILE(p, plane_loop); }
/* Interpolate between two vertices and put the result into a0.0. * Increment a0.0 accordingly. * * Beware that dest_ptr can be equal to v0_ptr! */ void brw_clip_interp_vertex( struct brw_clip_compile *c, struct brw_indirect dest_ptr, struct brw_indirect v0_ptr, /* from */ struct brw_indirect v1_ptr, /* to */ struct brw_reg t0, bool force_edgeflag) { struct brw_codegen *p = &c->func; struct brw_reg t_nopersp, v0_ndc_copy; GLuint slot; /* Just copy the vertex header: */ /* * After CLIP stage, only first 256 bits of the VUE are read * back on Ironlake, so needn't change it */ brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1); /* First handle the 3D and NDC interpolation, in case we * need noperspective interpolation. Doing it early has no * performance impact in any case. */ /* Take a copy of the v0 NDC coordinates, in case dest == v0. */ if (c->has_noperspective_shading) { GLuint offset = brw_varying_to_offset(&c->vue_map, BRW_VARYING_SLOT_NDC); v0_ndc_copy = get_tmp(c); brw_MOV(p, v0_ndc_copy, deref_4f(v0_ptr, offset)); } /* Compute the new 3D position * * dest_hpos = v0_hpos * (1 - t0) + v1_hpos * t0 */ { GLuint delta = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS); struct brw_reg tmp = get_tmp(c); brw_MUL(p, vec4(brw_null_reg()), deref_4f(v1_ptr, delta), t0); brw_MAC(p, tmp, negate(deref_4f(v0_ptr, delta)), t0); brw_ADD(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta), tmp); release_tmp(c, tmp); } /* Recreate the projected (NDC) coordinate in the new vertex header */ brw_clip_project_vertex(c, dest_ptr); /* If we have noperspective attributes, * we need to compute the screen-space t */ if (c->has_noperspective_shading) { GLuint delta = brw_varying_to_offset(&c->vue_map, BRW_VARYING_SLOT_NDC); struct brw_reg tmp = get_tmp(c); t_nopersp = get_tmp(c); /* t_nopersp = vec4(v1.xy, dest.xy) */ brw_MOV(p, t_nopersp, deref_4f(v1_ptr, delta)); brw_MOV(p, tmp, deref_4f(dest_ptr, delta)); brw_set_default_access_mode(p, BRW_ALIGN_16); brw_MOV(p, brw_writemask(t_nopersp, WRITEMASK_ZW), brw_swizzle(tmp, 0, 1, 0, 1)); /* t_nopersp = vec4(v1.xy, dest.xy) - v0.xyxy */ brw_ADD(p, t_nopersp, t_nopersp, negate(brw_swizzle(v0_ndc_copy, 0, 1, 0, 1))); /* Add the absolute values of the X and Y deltas so that if * the points aren't in the same place on the screen we get * nonzero values to divide. * * After that, we have vert1 - vert0 in t_nopersp.x and * vertnew - vert0 in t_nopersp.y * * t_nopersp = vec2(|v1.x -v0.x| + |v1.y -v0.y|, * |dest.x-v0.x| + |dest.y-v0.y|) */ brw_ADD(p, brw_writemask(t_nopersp, WRITEMASK_XY), brw_abs(brw_swizzle(t_nopersp, 0, 2, 0, 0)), brw_abs(brw_swizzle(t_nopersp, 1, 3, 0, 0))); brw_set_default_access_mode(p, BRW_ALIGN_1); /* If the points are in the same place, just substitute a * value to avoid divide-by-zero */ brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, vec1(t_nopersp), brw_imm_f(0)); brw_IF(p, BRW_EXECUTE_1); brw_MOV(p, t_nopersp, brw_imm_vf4(brw_float_to_vf(1.0), brw_float_to_vf(0.0), brw_float_to_vf(0.0), brw_float_to_vf(0.0))); brw_ENDIF(p); /* Now compute t_nopersp = t_nopersp.y/t_nopersp.x and broadcast it. */ brw_math_invert(p, get_element(t_nopersp, 0), get_element(t_nopersp, 0)); brw_MUL(p, vec1(t_nopersp), vec1(t_nopersp), vec1(suboffset(t_nopersp, 1))); brw_set_default_access_mode(p, BRW_ALIGN_16); brw_MOV(p, t_nopersp, brw_swizzle(t_nopersp, 0, 0, 0, 0)); brw_set_default_access_mode(p, BRW_ALIGN_1); release_tmp(c, tmp); release_tmp(c, v0_ndc_copy); } /* Now we can iterate over each attribute * (could be done in pairs?) */ for (slot = 0; slot < c->vue_map.num_slots; slot++) { int varying = c->vue_map.slot_to_varying[slot]; GLuint delta = brw_vue_slot_to_offset(slot); /* HPOS, NDC already handled above */ if (varying == VARYING_SLOT_POS || varying == BRW_VARYING_SLOT_NDC) continue; if (varying == VARYING_SLOT_EDGE) { if (force_edgeflag) brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1)); else brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta)); } else if (varying == VARYING_SLOT_PSIZ) { /* PSIZ doesn't need interpolation because it isn't used by the * fragment shader. */ } else if (varying < VARYING_SLOT_MAX) { /* This is a true vertex result (and not a special value for the VUE * header), so interpolate: * * New = attr0 + t*attr1 - t*attr0 * * Unless the attribute is flat shaded -- in which case just copy * from one of the sources (doesn't matter which; already copied from pv) */ GLuint interp = c->key.interpolation_mode.mode[slot]; if (interp != INTERP_QUALIFIER_FLAT) { struct brw_reg tmp = get_tmp(c); struct brw_reg t = interp == INTERP_QUALIFIER_NOPERSPECTIVE ? t_nopersp : t0; brw_MUL(p, vec4(brw_null_reg()), deref_4f(v1_ptr, delta), t); brw_MAC(p, tmp, negate(deref_4f(v0_ptr, delta)), t); brw_ADD(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta), tmp); release_tmp(c, tmp); } else { brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta)); } } } if (c->vue_map.num_slots % 2) { GLuint delta = brw_vue_slot_to_offset(c->vue_map.num_slots); brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0)); } if (c->has_noperspective_shading) release_tmp(c, t_nopersp); }
/* Line clipping, more or less following the following algorithm: * * for (p=0;p<MAX_PLANES;p++) { * if (clipmask & (1 << p)) { * GLfloat dp0 = DOTPROD( vtx0, plane[p] ); * GLfloat dp1 = DOTPROD( vtx1, plane[p] ); * * if (dp1 < 0.0f) { * GLfloat t = dp1 / (dp1 - dp0); * if (t > t1) t1 = t; * } else { * GLfloat t = dp0 / (dp0 - dp1); * if (t > t0) t0 = t; * } * * if (t0 + t1 >= 1.0) * return; * } * } * * interp( ctx, newvtx0, vtx0, vtx1, t0 ); * interp( ctx, newvtx1, vtx1, vtx0, t1 ); * */ static void clip_and_emit_line( struct brw_clip_compile *c ) { struct brw_codegen *p = &c->func; struct brw_indirect vtx0 = brw_indirect(0, 0); struct brw_indirect vtx1 = brw_indirect(1, 0); struct brw_indirect newvtx0 = brw_indirect(2, 0); struct brw_indirect newvtx1 = brw_indirect(3, 0); struct brw_indirect plane_ptr = brw_indirect(4, 0); struct brw_reg v1_null_ud = retype(vec1(brw_null_reg()), BRW_REGISTER_TYPE_UD); GLuint hpos_offset = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS); GLint clipdist0_offset = c->key.nr_userclip ? brw_varying_to_offset(&c->vue_map, VARYING_SLOT_CLIP_DIST0) : 0; brw_MOV(p, get_addr_reg(vtx0), brw_address(c->reg.vertex[0])); brw_MOV(p, get_addr_reg(vtx1), brw_address(c->reg.vertex[1])); brw_MOV(p, get_addr_reg(newvtx0), brw_address(c->reg.vertex[2])); brw_MOV(p, get_addr_reg(newvtx1), brw_address(c->reg.vertex[3])); brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c)); /* Note: init t0, t1 together: */ brw_MOV(p, vec2(c->reg.t0), brw_imm_f(0)); brw_clip_init_planes(c); brw_clip_init_clipmask(c); /* -ve rhw workaround */ if (p->devinfo->has_negative_rhw_bug) { brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<20)); brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ); brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(0x3f)); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); } /* Set the initial vertex source mask: The first 6 planes are the bounds * of the view volume; the next 8 planes are the user clipping planes. */ brw_MOV(p, c->reg.vertex_src_mask, brw_imm_ud(0x3fc0)); /* Set the initial clipdistance offset to be 6 floats before gl_ClipDistance[0]. * We'll increment 6 times before we start hitting actual user clipping. */ brw_MOV(p, c->reg.clipdistance_offset, brw_imm_d(clipdist0_offset - 6*sizeof(float))); brw_DO(p, BRW_EXECUTE_1); { /* if (planemask & 1) */ brw_AND(p, v1_null_ud, c->reg.planemask, brw_imm_ud(1)); brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ); brw_IF(p, BRW_EXECUTE_1); { brw_AND(p, v1_null_ud, c->reg.vertex_src_mask, brw_imm_ud(1)); brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ); brw_IF(p, BRW_EXECUTE_1); { /* user clip distance: just fetch the correct float from each vertex */ struct brw_indirect temp_ptr = brw_indirect(7, 0); brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx0), c->reg.clipdistance_offset); brw_MOV(p, c->reg.dp0, deref_1f(temp_ptr, 0)); brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx1), c->reg.clipdistance_offset); brw_MOV(p, c->reg.dp1, deref_1f(temp_ptr, 0)); } brw_ELSE(p); { /* fixed plane: fetch the hpos, dp4 against the plane. */ if (c->key.nr_userclip) brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0)); else brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0)); brw_DP4(p, vec4(c->reg.dp0), deref_4f(vtx0, hpos_offset), c->reg.plane_equation); brw_DP4(p, vec4(c->reg.dp1), deref_4f(vtx1, hpos_offset), c->reg.plane_equation); } brw_ENDIF(p); brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, vec1(c->reg.dp1), brw_imm_f(0.0f)); brw_IF(p, BRW_EXECUTE_1); { /* * Both can be negative on GM965/G965 due to RHW workaround * if so, this object should be rejected. */ if (p->devinfo->has_negative_rhw_bug) { brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE, c->reg.dp0, brw_imm_f(0.0)); brw_IF(p, BRW_EXECUTE_1); { brw_clip_kill_thread(c); } brw_ENDIF(p); } brw_ADD(p, c->reg.t, c->reg.dp1, negate(c->reg.dp0)); brw_math_invert(p, c->reg.t, c->reg.t); brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp1); brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t1 ); brw_MOV(p, c->reg.t1, c->reg.t); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); } brw_ELSE(p); { /* Coming back in. We know that both cannot be negative * because the line would have been culled in that case. */ /* If both are positive, do nothing */ /* Only on GM965/G965 */ if (p->devinfo->has_negative_rhw_bug) { brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0)); brw_IF(p, BRW_EXECUTE_1); } { brw_ADD(p, c->reg.t, c->reg.dp0, negate(c->reg.dp1)); brw_math_invert(p, c->reg.t, c->reg.t); brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp0); brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t0 ); brw_MOV(p, c->reg.t0, c->reg.t); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); } if (p->devinfo->has_negative_rhw_bug) { brw_ENDIF(p); } } brw_ENDIF(p); } brw_ENDIF(p); /* plane_ptr++; */ brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c)); /* while (planemask>>=1) != 0 */ brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1)); brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ); brw_SHR(p, c->reg.vertex_src_mask, c->reg.vertex_src_mask, brw_imm_ud(1)); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); brw_ADD(p, c->reg.clipdistance_offset, c->reg.clipdistance_offset, brw_imm_w(sizeof(float))); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); } brw_WHILE(p); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); brw_ADD(p, c->reg.t, c->reg.t0, c->reg.t1); brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.t, brw_imm_f(1.0)); brw_IF(p, BRW_EXECUTE_1); { brw_clip_interp_vertex(c, newvtx0, vtx0, vtx1, c->reg.t0, false); brw_clip_interp_vertex(c, newvtx1, vtx1, vtx0, c->reg.t1, false); brw_clip_emit_vue(c, newvtx0, BRW_URB_WRITE_ALLOCATE_COMPLETE, (_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT) | URB_WRITE_PRIM_START); brw_clip_emit_vue(c, newvtx1, BRW_URB_WRITE_EOT_COMPLETE, (_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT) | URB_WRITE_PRIM_END); } brw_ENDIF(p); brw_clip_kill_thread(c); }
/* Use mesa's clipping algorithms, translated to GEN4 assembly. */ void brw_clip_tri( struct brw_clip_compile *c ) { struct brw_codegen *p = &c->func; struct brw_indirect vtx = brw_indirect(0, 0); struct brw_indirect vtxPrev = brw_indirect(1, 0); struct brw_indirect vtxOut = brw_indirect(2, 0); struct brw_indirect plane_ptr = brw_indirect(3, 0); struct brw_indirect inlist_ptr = brw_indirect(4, 0); struct brw_indirect outlist_ptr = brw_indirect(5, 0); struct brw_indirect freelist_ptr = brw_indirect(6, 0); GLuint hpos_offset = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS); GLint clipdist0_offset = c->key.nr_userclip ? brw_varying_to_offset(&c->vue_map, VARYING_SLOT_CLIP_DIST0) : 0; brw_MOV(p, get_addr_reg(vtxPrev), brw_address(c->reg.vertex[2]) ); brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c)); brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist)); brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist)); brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3]) ); /* Set the initial vertex source mask: The first 6 planes are the bounds * of the view volume; the next 8 planes are the user clipping planes. */ brw_MOV(p, c->reg.vertex_src_mask, brw_imm_ud(0x3fc0)); /* Set the initial clipdistance offset to be 6 floats before gl_ClipDistance[0]. * We'll increment 6 times before we start hitting actual user clipping. */ brw_MOV(p, c->reg.clipdistance_offset, brw_imm_d(clipdist0_offset - 6*sizeof(float))); brw_DO(p, BRW_EXECUTE_1); { /* if (planemask & 1) */ brw_AND(p, vec1(brw_null_reg()), c->reg.planemask, brw_imm_ud(1)); brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ); brw_IF(p, BRW_EXECUTE_1); { /* vtxOut = freelist_ptr++ */ brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(freelist_ptr) ); brw_ADD(p, get_addr_reg(freelist_ptr), get_addr_reg(freelist_ptr), brw_imm_uw(c->nr_regs * REG_SIZE)); if (c->key.nr_userclip) brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0)); else brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0)); brw_MOV(p, c->reg.loopcount, c->reg.nr_verts); brw_MOV(p, c->reg.nr_verts, brw_imm_ud(0)); brw_DO(p, BRW_EXECUTE_1); { /* vtx = *input_ptr; */ brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0)); load_clip_distance(c, vtxPrev, c->reg.dpPrev, hpos_offset, BRW_CONDITIONAL_L); /* (prev < 0.0f) */ brw_IF(p, BRW_EXECUTE_1); { load_clip_distance(c, vtx, c->reg.dp, hpos_offset, BRW_CONDITIONAL_GE); /* IS_POSITIVE(next) */ brw_IF(p, BRW_EXECUTE_1); { /* Coming back in. */ brw_ADD(p, c->reg.t, c->reg.dpPrev, negate(c->reg.dp)); brw_math_invert(p, c->reg.t, c->reg.t); brw_MUL(p, c->reg.t, c->reg.t, c->reg.dpPrev); /* If (vtxOut == 0) vtxOut = vtxPrev */ brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) ); brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtxPrev)); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); brw_clip_interp_vertex(c, vtxOut, vtxPrev, vtx, c->reg.t, false); /* *outlist_ptr++ = vtxOut; * nr_verts++; * vtxOut = 0; */ brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut)); brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) ); } brw_ENDIF(p); } brw_ELSE(p); { /* *outlist_ptr++ = vtxPrev; * nr_verts++; */ brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxPrev)); brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); load_clip_distance(c, vtx, c->reg.dp, hpos_offset, BRW_CONDITIONAL_L); /* (next < 0.0f) */ brw_IF(p, BRW_EXECUTE_1); { /* Going out of bounds. Avoid division by zero as we * know dp != dpPrev from DIFFERENT_SIGNS, above. */ brw_ADD(p, c->reg.t, c->reg.dp, negate(c->reg.dpPrev)); brw_math_invert(p, c->reg.t, c->reg.t); brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp); /* If (vtxOut == 0) vtxOut = vtx */ brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) ); brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtx)); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); brw_clip_interp_vertex(c, vtxOut, vtx, vtxPrev, c->reg.t, true); /* *outlist_ptr++ = vtxOut; * nr_verts++; * vtxOut = 0; */ brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut)); brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) ); } brw_ENDIF(p); } brw_ENDIF(p); /* vtxPrev = vtx; * inlist_ptr++; */ brw_MOV(p, get_addr_reg(vtxPrev), get_addr_reg(vtx)); brw_ADD(p, get_addr_reg(inlist_ptr), get_addr_reg(inlist_ptr), brw_imm_uw(sizeof(short))); /* while (--loopcount != 0) */ brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1)); brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ); } brw_WHILE(p); brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL); /* vtxPrev = *(outlist_ptr-1) OR: outlist[nr_verts-1] * inlist = outlist * inlist_ptr = &inlist[0] * outlist_ptr = &outlist[0] */ brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_w(-2)); brw_MOV(p, get_addr_reg(vtxPrev), deref_1uw(outlist_ptr, 0)); brw_MOV(p, brw_vec8_grf(c->reg.inlist.nr, 0), brw_vec8_grf(c->reg.outlist.nr, 0)); brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist)); brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist)); } brw_ENDIF(p); /* plane_ptr++; */ brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c)); /* nr_verts >= 3 */ brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_GE, c->reg.nr_verts, brw_imm_ud(3)); brw_set_default_predicate_control(p, BRW_PREDICATE_NORMAL); /* && (planemask>>=1) != 0 */ brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1)); brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ); brw_SHR(p, c->reg.vertex_src_mask, c->reg.vertex_src_mask, brw_imm_ud(1)); brw_ADD(p, c->reg.clipdistance_offset, c->reg.clipdistance_offset, brw_imm_w(sizeof(float))); } brw_WHILE(p); brw_set_default_predicate_control(p, BRW_PREDICATE_NONE); }
/* Interpolate between two vertices and put the result into a0.0. * Increment a0.0 accordingly. */ void brw_clip_interp_vertex( struct brw_clip_compile *c, struct brw_indirect dest_ptr, struct brw_indirect v0_ptr, /* from */ struct brw_indirect v1_ptr, /* to */ struct brw_reg t0, bool force_edgeflag) { struct brw_compile *p = &c->func; struct brw_reg tmp = get_tmp(c); GLuint slot; /* Just copy the vertex header: */ /* * After CLIP stage, only first 256 bits of the VUE are read * back on Ironlake, so needn't change it */ brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1); /* Iterate over each attribute (could be done in pairs?) */ for (slot = 0; slot < c->vue_map.num_slots; slot++) { int varying = c->vue_map.slot_to_varying[slot]; GLuint delta = brw_vue_slot_to_offset(slot); if (varying == VARYING_SLOT_EDGE) { if (force_edgeflag) brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1)); else brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta)); } else if (varying == VARYING_SLOT_PSIZ || varying == VARYING_SLOT_CLIP_DIST0 || varying == VARYING_SLOT_CLIP_DIST1) { /* PSIZ doesn't need interpolation because it isn't used by the * fragment shader. CLIP_DIST0 and CLIP_DIST1 don't need * intepolation because on pre-GEN6, these are just placeholder VUE * slots that don't perform any action. */ } else if (varying < VARYING_SLOT_MAX) { /* This is a true vertex result (and not a special value for the VUE * header), so interpolate: * * New = attr0 + t*attr1 - t*attr0 */ brw_MUL(p, vec4(brw_null_reg()), deref_4f(v1_ptr, delta), t0); brw_MAC(p, tmp, negate(deref_4f(v0_ptr, delta)), t0); brw_ADD(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta), tmp); } } if (c->vue_map.num_slots % 2) { GLuint delta = brw_vue_slot_to_offset(c->vue_map.num_slots); brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0)); } release_tmp(c, tmp); /* Recreate the projected (NDC) coordinate in the new vertex * header: */ brw_clip_project_vertex(c, dest_ptr ); }