struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
GLfloat (*tc0)[4], (*tc1)[4], (*fog)[4];
GLfloat (*tc2)[4], (*tc3)[4];
- GLubyte (*col)[4], (*spec)[4];
- GLuint tc0_stride, tc1_stride, col_stride, spec_stride, fog_stride;
+ GLfloat (*spec)[4];
+ GLfloat (*col)[4];
+ GLuint col_stride;
+ GLuint tc0_stride, tc1_stride, spec_stride, fog_stride;
GLuint tc2_stride, tc3_stride;
GLuint tc0_size, tc1_size;
GLuint tc2_size, tc3_size;
}
if (DO_RGBA) {
- if (VB->ColorPtr[0]->Type != GL_UNSIGNED_BYTE)
- IMPORT_FLOAT_COLORS( ctx );
- col = (GLubyte (*)[4])VB->ColorPtr[0]->Ptr;
- col_stride = VB->ColorPtr[0]->StrideB;
+ col = VB->ColorPtr[0]->data;
+ col_stride = VB->ColorPtr[0]->stride;
}
if (DO_SPEC) {
- if (VB->SecondaryColorPtr[0]->Type != GL_UNSIGNED_BYTE)
- IMPORT_FLOAT_SPEC_COLORS( ctx );
- spec = (GLubyte (*)[4])VB->SecondaryColorPtr[0]->Ptr;
- spec_stride = VB->SecondaryColorPtr[0]->StrideB;
+ spec = VB->SecondaryColorPtr[0]->data;
+ spec_stride = VB->SecondaryColorPtr[0]->stride;
+ } else {
+ spec = (GLfloat (*)[4])ctx->Current.Attrib[VERT_ATTRIB_COLOR1];
+ spec_stride = 0;
}
if (DO_FOG) {
}
}
- if (VB->importable_data || (DO_SPEC && !spec_stride) || (DO_FOG && !fog_stride)) {
- /* May have nonstandard strides:
- */
- if (start) {
- coord = (GLfloat (*)[4])((GLubyte *)coord + start * coord_stride);
- if (DO_TEX0)
- tc0 = (GLfloat (*)[4])((GLubyte *)tc0 + start * tc0_stride);
- if (DO_TEX1)
- tc1 = (GLfloat (*)[4])((GLubyte *)tc1 + start * tc1_stride);
- if (DO_TEX2)
- tc2 = (GLfloat (*)[4])((GLubyte *)tc2 + start * tc2_stride);
- if (DO_TEX3)
- tc3 = (GLfloat (*)[4])((GLubyte *)tc3 + start * tc3_stride);
- if (DO_RGBA)
- STRIDE_4UB(col, start * col_stride);
- if (DO_SPEC)
- STRIDE_4UB(spec, start * spec_stride);
- if (DO_FOG)
- fog = (GLfloat (*)[4])((GLubyte *)fog + start * fog_stride);
- /* STRIDE_F(fog, start * fog_stride); */
- }
-
- for (i=start; i < end; i++, v = (VERTEX *)((GLubyte *)v + stride)) {
- if (DO_XYZW) {
- if (HAVE_HW_VIEWPORT || mask[i] == 0) {
- /* unclipped */
- VIEWPORT_X(v->v.x, coord[0][0]);
- VIEWPORT_Y(v->v.y, coord[0][1]);
- VIEWPORT_Z(v->v.z, coord[0][2]);
- v->v.w = coord[0][3];
- } else {
- /* clipped */
- v->v.w = 1.0;
- }
- if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
- fprintf(stderr, "%s: vert (importable) %d: %.2f %.2f %.2f %f\n",
- __FUNCTION__, i, v->v.x, v->v.y, v->v.z, v->v.w);
- }
- coord = (GLfloat (*)[4])((GLubyte *)coord + coord_stride);
+ /* May have nonstandard strides:
+ */
+ if (start) {
+ coord = (GLfloat (*)[4])((GLubyte *)coord + start * coord_stride);
+ if (DO_TEX0)
+ tc0 = (GLfloat (*)[4])((GLubyte *)tc0 + start * tc0_stride);
+ if (DO_TEX1)
+ tc1 = (GLfloat (*)[4])((GLubyte *)tc1 + start * tc1_stride);
+ if (DO_TEX2)
+ tc2 = (GLfloat (*)[4])((GLubyte *)tc2 + start * tc2_stride);
+ if (DO_TEX3)
+ tc3 = (GLfloat (*)[4])((GLubyte *)tc3 + start * tc3_stride);
+ if (DO_RGBA)
+ STRIDE_4F(col, start * col_stride);
+ if (DO_SPEC)
+ STRIDE_4F(spec, start * spec_stride);
+ if (DO_FOG)
+ STRIDE_4F(fog, start * fog_stride);
+ // fog = (GLfloat (*)[4])((GLubyte *)fog + start * fog_stride);
+ /* STRIDE_F(fog, start * fog_stride); */
+ }
+
+ for (i=start; i < end; i++, v = (VERTEX *)((GLubyte *)v + stride)) {
+ if (DO_XYZW) {
+ if (HAVE_HW_VIEWPORT || mask[i] == 0) {
+ /* unclipped */
+ VIEWPORT_X(v->v.x, coord[0][0]);
+ VIEWPORT_Y(v->v.y, coord[0][1]);
+ VIEWPORT_Z(v->v.z, coord[0][2]);
+ v->v.w = coord[0][3];
+ } else {
+ /* clipped */
+ v->v.w = 1.0;
}
- if (DO_RGBA) {
- if (HAVE_RGBA_COLOR) {
- *(GLuint *)&v->v.color = *(GLuint *)&col[0];
- STRIDE_4UB(col, col_stride);
- } else {
- v->v.color.blue = col[0][2];
- v->v.color.green = col[0][1];
- v->v.color.red = col[0][0];
- v->v.color.alpha = col[0][3];
- STRIDE_4UB(col, col_stride);
- }
+ if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
+ fprintf(stderr, "%s: vert (importable) %d: %.2f %.2f %.2f %f\n",
+ __FUNCTION__, i, v->v.x, v->v.y, v->v.z, v->v.w);
}
- if (DO_SPEC) {
- v->v.specular.red = spec[0][0];
- v->v.specular.green = spec[0][1];
- v->v.specular.blue = spec[0][2];
- STRIDE_4UB(spec, spec_stride);
+ coord = (GLfloat (*)[4])((GLubyte *)coord + coord_stride);
+ }
+ if (DO_RGBA) {
+ if (HAVE_RGBA_COLOR) {
+ *(GLuint *)&v->v.color = *(GLuint *)&col[0];
+ STRIDE_4F(col, col_stride);
+ } else {
+ v->v.color.blue = col[0][2];
+ v->v.color.green = col[0][1];
+ v->v.color.red = col[0][0];
+ v->v.color.alpha = col[0][3];
+ STRIDE_4F(col, col_stride);
}
- if (DO_FOG) {
- v->v.specular.alpha = fog[0][0] * 255.0;
- /* STRIDE_F(fog, fog_stride); */
- fog = (GLfloat (*)[4])((GLubyte *)fog + fog_stride);
+ }
+ if (DO_SPEC) {
+ v->v.specular.red = spec[0][0];
+ v->v.specular.green = spec[0][1];
+ v->v.specular.blue = spec[0][2];
+ STRIDE_4F(spec, spec_stride);
+ }
+ if (DO_FOG) {
+ v->v.specular.alpha = fog[0][0] * 255.0;
+ /* STRIDE_F(fog, fog_stride); */
+ fog = (GLfloat (*)[4])((GLubyte *)fog + fog_stride);
+ }
+ if (DO_TEX0) {
+ v->v.u0 = tc0[0][0];
+ v->v.v0 = tc0[0][1];
+ if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
+ fprintf(stderr, "%s: vert (importable) %d: u0: %.2f, v0: %.2f, w: %f\n",
+ __FUNCTION__, i, v->v.u0, v->v.v0, v->v.w);
}
- if (DO_TEX0) {
- v->v.u0 = tc0[0][0];
- v->v.v0 = tc0[0][1];
- if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
- fprintf(stderr, "%s: vert (importable) %d: u0: %.2f, v0: %.2f, w: %f\n",
- __FUNCTION__, i, v->v.u0, v->v.v0, v->v.w);
- }
#ifdef MACH64_PREMULT_TEXCOORDS
- v->v.u0 *= v->v.w;
- v->v.v0 *= v->v.w;
+ v->v.u0 *= v->v.w;
+ v->v.v0 *= v->v.w;
#endif
- if (DO_PTEX) {
- if (HAVE_PTEX_VERTICES) {
- if (tc0_size == 4)
- v->pv.q0 = tc0[0][3];
- else
- v->pv.q0 = 1.0;
- }
- else if (tc0_size == 4) {
-#ifdef MACH64_PREMULT_TEXCOORDS
- v->v.w *= tc0[0][3];
-#else
- float rhw = 1.0 / tc0[0][3];
- v->v.w *= tc0[0][3];
- v->v.u0 *= rhw;
- v->v.v0 *= rhw;
-#endif
- }
- }
- tc0 = (GLfloat (*)[4])((GLubyte *)tc0 + tc0_stride);
- }
- if (DO_TEX1) {
- if (DO_PTEX) {
- v->pv.u1 = tc1[0][0];
- v->pv.v1 = tc1[0][1];
- if (tc1_size == 4)
- v->pv.q1 = tc1[0][3];
+ if (DO_PTEX) {
+ if (HAVE_PTEX_VERTICES) {
+ if (tc0_size == 4)
+ v->pv.q0 = tc0[0][3];
else
- v->pv.q1 = 1.0;
+ v->pv.q0 = 1.0;
}
- else {
- v->v.u1 = tc1[0][0];
- v->v.v1 = tc1[0][1];
- }
+ else if (tc0_size == 4) {
#ifdef MACH64_PREMULT_TEXCOORDS
- v->v.u1 *= v->v.w;
- v->v.v1 *= v->v.w;
+ v->v.w *= tc0[0][3];
+#else
+ float rhw = 1.0 / tc0[0][3];
+ v->v.w *= tc0[0][3];
+ v->v.u0 *= rhw;
+ v->v.v0 *= rhw;
#endif
- tc1 = (GLfloat (*)[4])((GLubyte *)tc1 + tc1_stride);
- }
- else if (DO_PTEX) {
- *(GLuint *)&v->pv.q1 = 0; /* avoid culling on radeon */
- }
- if (DO_TEX2) {
- if (DO_PTEX) {
- v->pv.u2 = tc2[0][0];
- v->pv.v2 = tc2[0][1];
- if (tc2_size == 4)
- v->pv.q2 = tc2[0][3];
- else
- v->pv.q2 = 1.0;
- }
- else {
- v->v.u2 = tc2[0][0];
- v->v.v2 = tc2[0][1];
- }
- tc2 = (GLfloat (*)[4])((GLubyte *)tc2 + tc2_stride);
- }
- if (DO_TEX3) {
- if (DO_PTEX) {
- v->pv.u3 = tc3[0][0];
- v->pv.v3 = tc3[0][1];
- if (tc3_size == 4)
- v->pv.q3 = tc3[0][3];
- else
- v->pv.q3 = 1.0;
}
- else {
- v->v.u3 = tc3[0][0];
- v->v.v3 = tc3[0][1];
- }
- tc3 = (GLfloat (*)[4])((GLubyte *)tc3 + tc3_stride);
}
+ tc0 = (GLfloat (*)[4])((GLubyte *)tc0 + tc0_stride);
}
- }
- else {
- for (i=start; i < end; i++, v = (VERTEX *)((GLubyte *)v + stride)) {
- if (DO_XYZW) {
- if (HAVE_HW_VIEWPORT || mask[i] == 0) {
- /* unclipped */
- VIEWPORT_X(v->v.x, coord[i][0]);
- VIEWPORT_Y(v->v.y, coord[i][1]);
- VIEWPORT_Z(v->v.z, coord[i][2]);
- v->v.w = coord[i][3];
- } else {
- /* clipped */
- v->v.w = 1.0;
- }
- if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
- fprintf(stderr, "%s: vert %d: %.2f %.2f %.2f %f\n",
- __FUNCTION__, i, v->v.x, v->v.y, v->v.z, v->v.w);
- }
- }
- if (DO_RGBA) {
- if (HAVE_RGBA_COLOR) {
- *(GLuint *)&v->v.color = *(GLuint *)&col[i];
- }
- else {
- v->v.color.blue = col[i][2];
- v->v.color.green = col[i][1];
- v->v.color.red = col[i][0];
- v->v.color.alpha = col[i][3];
- }
- }
- if (DO_SPEC) {
- v->v.specular.red = spec[i][0];
- v->v.specular.green = spec[i][1];
- v->v.specular.blue = spec[i][2];
- }
- if (DO_FOG) {
- v->v.specular.alpha = fog[i][0] * 255.0;
+ if (DO_TEX1) {
+ if (DO_PTEX) {
+ v->pv.u1 = tc1[0][0];
+ v->pv.v1 = tc1[0][1];
+ if (tc1_size == 4)
+ v->pv.q1 = tc1[0][3];
+ else
+ v->pv.q1 = 1.0;
+ }
+ else {
+ v->v.u1 = tc1[0][0];
+ v->v.v1 = tc1[0][1];
}
- if (DO_TEX0) {
- v->v.u0 = tc0[i][0];
- v->v.v0 = tc0[i][1];
- if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
- fprintf(stderr, "%s: vert %d: u0: %.2f, v0: %.2f, w: %f\n",
- __FUNCTION__, i, v->v.u0, v->v.v0, v->v.w);
- }
-#ifdef MACH64_PREMULT_TEXCOORDS
- v->v.u0 *= v->v.w;
- v->v.v0 *= v->v.w;
-#endif
- if (DO_PTEX) {
- if (HAVE_PTEX_VERTICES) {
- if (tc0_size == 4)
- v->pv.q0 = tc0[i][3];
- else
- v->pv.q0 = 1.0;
-
- v->pv.q1 = 0; /* radeon */
- }
- else if (tc0_size == 4) {
#ifdef MACH64_PREMULT_TEXCOORDS
- v->v.w *= tc0[i][3];
-#else
- float rhw = 1.0 / tc0[i][3];
- v->v.w *= tc0[i][3];
- v->v.u0 *= rhw;
- v->v.v0 *= rhw;
+ v->v.u1 *= v->v.w;
+ v->v.v1 *= v->v.w;
#endif
- }
- }
+ tc1 = (GLfloat (*)[4])((GLubyte *)tc1 + tc1_stride);
+ }
+ else if (DO_PTEX) {
+ *(GLuint *)&v->pv.q1 = 0; /* avoid culling on radeon */
+ }
+ if (DO_TEX2) {
+ if (DO_PTEX) {
+ v->pv.u2 = tc2[0][0];
+ v->pv.v2 = tc2[0][1];
+ if (tc2_size == 4)
+ v->pv.q2 = tc2[0][3];
+ else
+ v->pv.q2 = 1.0;
+ }
+ else {
+ v->v.u2 = tc2[0][0];
+ v->v.v2 = tc2[0][1];
}
- if (DO_TEX1) {
- if (DO_PTEX) {
- v->pv.u1 = tc1[i][0];
- v->pv.v1 = tc1[i][1];
- if (tc1_size == 4)
- v->pv.q1 = tc1[i][3];
- else
- v->pv.q1 = 1.0;
+ tc2 = (GLfloat (*)[4])((GLubyte *)tc2 + tc2_stride);
+ }
+ if (DO_TEX3) {
+ if (DO_PTEX) {
+ v->pv.u3 = tc3[0][0];
+ v->pv.v3 = tc3[0][1];
+ if (tc3_size == 4)
+ v->pv.q3 = tc3[0][3];
+ else
+ v->pv.q3 = 1.0;
}
- else {
- v->v.u1 = tc1[i][0];
- v->v.v1 = tc1[i][1];
- }
-#ifdef MACH64_PREMULT_TEXCOORDS
- v->v.u1 *= v->v.w;
- v->v.v1 *= v->v.w;
-#endif
+ else {
+ v->v.u3 = tc3[0][0];
+ v->v.v3 = tc3[0][1];
}
- }
+ tc3 = (GLfloat (*)[4])((GLubyte *)tc3 + tc3_stride);
+ }
}
}
+
#else
#if DO_XYZW
{
LOCALVARS
struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
- GLubyte (*col)[4];
+ GLfloat (*col)[4];
GLuint col_stride;
GLfloat (*coord)[4] = VB->NdcPtr->data;
GLuint coord_stride = VB->NdcPtr->stride;
ASSERT(stride == 4);
- if (VB->ColorPtr[0]->Type != GL_UNSIGNED_BYTE)
- IMPORT_FLOAT_COLORS( ctx );
-
- col = (GLubyte (*)[4])VB->ColorPtr[0]->Ptr;
- col_stride = VB->ColorPtr[0]->StrideB;
- ASSERT(VB->ColorPtr[0]->Type == GL_UNSIGNED_BYTE);
-
-/* fprintf(stderr, "%s(small) importable %x\n", */
-/* __FUNCTION__, VB->importable_data); */
+ col = VB->ColorPtr[0]->data;
+ col_stride = VB->ColorPtr[0]->stride;
/* Pack what's left into a 4-dword vertex. Color is in a different
* place, and there is no 'w' coordinate.
*/
- if (VB->importable_data) {
- if (start) {
- coord = (GLfloat (*)[4])((GLubyte *)coord + start * coord_stride);
- STRIDE_4UB(col, start * col_stride);
+ if (start) {
+ coord = (GLfloat (*)[4])((GLubyte *)coord + start * coord_stride);
+ STRIDE_4F(col, start * col_stride);
+ }
+
+ for (i=start; i < end; i++, v+=4) {
+ if (HAVE_HW_VIEWPORT || mask[i] == 0) {
+ VIEWPORT_X(v[0], coord[0][0]);
+ VIEWPORT_Y(v[1], coord[0][1]);
+ VIEWPORT_Z(v[2], coord[0][2]);
}
-
- for (i=start; i < end; i++, v+=4) {
- if (HAVE_HW_VIEWPORT || mask[i] == 0) {
- VIEWPORT_X(v[0], coord[0][0]);
- VIEWPORT_Y(v[1], coord[0][1]);
- VIEWPORT_Z(v[2], coord[0][2]);
- }
- coord = (GLfloat (*)[4])((GLubyte *)coord + coord_stride);
- if (DO_RGBA) {
- if (HAVE_RGBA_COLOR) {
- *(GLuint *)&v[3] = *(GLuint *)col;
- }
- else {
- GLubyte *b = (GLubyte *)&v[3];
- b[0] = col[0][2];
- b[1] = col[0][1];
- b[2] = col[0][0];
- b[3] = col[0][3];
- }
- STRIDE_4UB( col, col_stride );
+ coord = (GLfloat (*)[4])((GLubyte *)coord + coord_stride);
+ if (DO_RGBA) {
+ if (HAVE_RGBA_COLOR) {
+ *(GLuint *)&v[3] = *(GLuint *)col;
}
- if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
- fprintf(stderr, "vert (importable) %d: %.2f %.2f %.2f %x\n",
- i, v[0], v[1], v[2], *(int *)&v[3]);
+ else {
+ GLubyte *b = (GLubyte *)&v[3];
+ UNCLAMPED_FLOAT_TO_UBYTE(b[0], col[0][2]);
+ UNCLAMPED_FLOAT_TO_UBYTE(b[1], col[0][1]);
+ UNCLAMPED_FLOAT_TO_UBYTE(b[2], col[0][0]);
+ UNCLAMPED_FLOAT_TO_UBYTE(b[3], col[0][3]);
}
+ STRIDE_4F( col, col_stride );
}
- }
- else {
- for (i=start; i < end; i++, v+=4) {
- if (HAVE_HW_VIEWPORT || mask[i] == 0) {
- VIEWPORT_X(v[0], coord[i][0]);
- VIEWPORT_Y(v[1], coord[i][1]);
- VIEWPORT_Z(v[2], coord[i][2]);
- }
- if (DO_RGBA) {
- if (HAVE_RGBA_COLOR) {
- *(GLuint *)&v[3] = *(GLuint *)&col[i];
- }
- else {
- GLubyte *b = (GLubyte *)&v[3];
- b[0] = col[i][2];
- b[1] = col[i][1];
- b[2] = col[i][0];
- b[3] = col[i][3];
- }
- }
- if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
- fprintf(stderr, "vert %d: %.2f %.2f %.2f %x\n",
- i, v[0], v[1], v[2], *(int *)&v[3]);
- }
+ if (MACH64_DEBUG & DEBUG_VERBOSE_PRIMS) {
+ fprintf(stderr, "vert (importable) %d: %.2f %.2f %.2f %x\n",
+ i, v[0], v[1], v[2], *(int *)&v[3]);
}
}
}
{
LOCALVARS
struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
- GLubyte (*col)[4];
+ GLfloat (*col)[4];
GLuint col_stride;
GLfloat *v = (GLfloat *)dest;
int i;
- if (VB->ColorPtr[0]->Type != GL_UNSIGNED_BYTE)
- IMPORT_FLOAT_COLORS( ctx );
-
- col = VB->ColorPtr[0]->Ptr;
- col_stride = VB->ColorPtr[0]->StrideB;
+ col = VB->ColorPtr[0]->data;
+ col_stride = VB->ColorPtr[0]->stride;
if (start)
- STRIDE_4UB(col, col_stride * start);
+ STRIDE_4F(col, col_stride * start);
/* Need to figure out where color is:
*/
}
else {
GLubyte *b = (GLubyte *)v;
- b[0] = col[0][2];
- b[1] = col[0][1];
- b[2] = col[0][0];
- b[3] = col[0][3];
+ UNCLAMPED_FLOAT_TO_UBYTE(b[0], col[0][2]);
+ UNCLAMPED_FLOAT_TO_UBYTE(b[1], col[0][1]);
+ UNCLAMPED_FLOAT_TO_UBYTE(b[2], col[0][0]);
+ UNCLAMPED_FLOAT_TO_UBYTE(b[3], col[0][3]);
}
- STRIDE_4UB( col, col_stride );
+ STRIDE_4F( col, col_stride );
}
}
#endif /* emit */