X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fvbo%2Fvbo_exec_api.c;h=cbf78b49631700c9385487e9faaf5d7ba672805d;hb=e81aaeba37f5419323d8f88bc10943c77e25ed14;hp=7534599c3134b1dce00aaf581c1f41ac462b59b9;hpb=1469ccb7464836c752fa2664c36d8fae7e80606c;p=mesa.git diff --git a/src/mesa/vbo/vbo_exec_api.c b/src/mesa/vbo/vbo_exec_api.c index 7534599c313..cbf78b49631 100644 --- a/src/mesa/vbo/vbo_exec_api.c +++ b/src/mesa/vbo/vbo_exec_api.c @@ -42,21 +42,18 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/api_arrayelt.h" #include "main/api_validate.h" #include "main/dispatch.h" +#include "util/bitscan.h" #include "vbo_context.h" #include "vbo_noop.h" -#ifdef ERROR -#undef ERROR -#endif - - /** ID/name for immediate-mode VBO */ #define IMM_BUFFER_NAME 0xaabbccdd -static void reset_attrfv( struct vbo_exec_context *exec ); +static void +vbo_reset_all_attr(struct vbo_exec_context *exec); /** @@ -64,7 +61,8 @@ static void reset_attrfv( struct vbo_exec_context *exec ); * primitive. This is called when we fill a vertex buffer before * hitting glEnd. */ -static void vbo_exec_wrap_buffers( struct vbo_exec_context *exec ) +static void +vbo_exec_wrap_buffers(struct vbo_exec_context *exec) { if (exec->vtx.prim_count == 0) { exec->vtx.copied.nr = 0; @@ -118,7 +116,7 @@ static void vbo_exec_wrap_buffers( struct vbo_exec_context *exec ) exec->vtx.prim[0].start = 0; exec->vtx.prim[0].count = 0; exec->vtx.prim_count++; - + if (exec->vtx.copied.nr == last_count) exec->vtx.prim[0].begin = last_begin; } @@ -139,13 +137,13 @@ vbo_exec_vtx_wrap(struct vbo_exec_context *exec) * to exec->vtx.copied. */ vbo_exec_wrap_buffers( exec ); - + if (!exec->vtx.buffer_ptr) { /* probably ran out of memory earlier when allocating the VBO */ return; } - /* Copy stored stored vertices to start of new list. + /* Copy stored stored vertices to start of new list. */ assert(exec->vtx.max_vert - exec->vtx.vert_count > exec->vtx.copied.nr); @@ -163,58 +161,62 @@ vbo_exec_vtx_wrap(struct vbo_exec_context *exec) /** * Copy the active vertex's values to the ctx->Current fields. */ -static void vbo_exec_copy_to_current( struct vbo_exec_context *exec ) +static void +vbo_exec_copy_to_current(struct vbo_exec_context *exec) { struct gl_context *ctx = exec->ctx; struct vbo_context *vbo = vbo_context(ctx); - GLuint i; + GLbitfield64 enabled = exec->vtx.enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS)); + + while (enabled) { + const int i = u_bit_scan64(&enabled); + + /* Note: the exec->vtx.current[i] pointers point into the + * ctx->Current.Attrib and ctx->Light.Material.Attrib arrays. + */ + GLfloat *current = (GLfloat *)vbo->currval[i].Ptr; + fi_type tmp[8]; /* space for doubles */ + int dmul = exec->vtx.attrtype[i] == GL_DOUBLE ? 2 : 1; + + assert(exec->vtx.attrsz[i]); + + if (exec->vtx.attrtype[i] == GL_DOUBLE) { + memset(tmp, 0, sizeof(tmp)); + memcpy(tmp, exec->vtx.attrptr[i], exec->vtx.attrsz[i] * sizeof(GLfloat)); + } else { + COPY_CLEAN_4V_TYPE_AS_UNION(tmp, + exec->vtx.attrsz[i], + exec->vtx.attrptr[i], + exec->vtx.attrtype[i]); + } + + if (exec->vtx.attrtype[i] != vbo->currval[i].Type || + memcmp(current, tmp, 4 * sizeof(GLfloat) * dmul) != 0) { + memcpy(current, tmp, 4 * sizeof(GLfloat) * dmul); - for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) { - if (exec->vtx.attrsz[i]) { - /* Note: the exec->vtx.current[i] pointers point into the - * ctx->Current.Attrib and ctx->Light.Material.Attrib arrays. + /* Given that we explicitly state size here, there is no need + * for the COPY_CLEAN above, could just copy 16 bytes and be + * done. The only problem is when Mesa accesses ctx->Current + * directly. */ - GLfloat *current = (GLfloat *)vbo->currval[i].Ptr; - fi_type tmp[8]; /* space for doubles */ - int dmul = exec->vtx.attrtype[i] == GL_DOUBLE ? 2 : 1; - - if (exec->vtx.attrtype[i] == GL_DOUBLE) { - memset(tmp, 0, sizeof(tmp)); - memcpy(tmp, exec->vtx.attrptr[i], exec->vtx.attrsz[i] * sizeof(GLfloat)); - } else { - COPY_CLEAN_4V_TYPE_AS_UNION(tmp, - exec->vtx.attrsz[i], - exec->vtx.attrptr[i], - exec->vtx.attrtype[i]); - } + /* Size here is in components - not bytes */ + vbo->currval[i].Size = exec->vtx.attrsz[i] / dmul; + vbo->currval[i]._ElementSize = + vbo->currval[i].Size * sizeof(GLfloat) * dmul; + vbo->currval[i].Type = exec->vtx.attrtype[i]; + vbo->currval[i].Integer = + vbo_attrtype_to_integer_flag(exec->vtx.attrtype[i]); + vbo->currval[i].Doubles = + vbo_attrtype_to_double_flag(exec->vtx.attrtype[i]); + + /* This triggers rather too much recalculation of Mesa state + * that doesn't get used (eg light positions). + */ + if (i >= VBO_ATTRIB_MAT_FRONT_AMBIENT && + i <= VBO_ATTRIB_MAT_BACK_INDEXES) + ctx->NewState |= _NEW_LIGHT; - if (exec->vtx.attrtype[i] != vbo->currval[i].Type || - memcmp(current, tmp, 4 * sizeof(GLfloat) * dmul) != 0) { - memcpy(current, tmp, 4 * sizeof(GLfloat) * dmul); - - /* Given that we explicitly state size here, there is no need - * for the COPY_CLEAN above, could just copy 16 bytes and be - * done. The only problem is when Mesa accesses ctx->Current - * directly. - */ - /* Size here is in components - not bytes */ - vbo->currval[i].Size = exec->vtx.attrsz[i] / dmul; - vbo->currval[i]._ElementSize = vbo->currval[i].Size * sizeof(GLfloat) * dmul; - vbo->currval[i].Type = exec->vtx.attrtype[i]; - vbo->currval[i].Integer = - vbo_attrtype_to_integer_flag(exec->vtx.attrtype[i]); - vbo->currval[i].Doubles = - vbo_attrtype_to_double_flag(exec->vtx.attrtype[i]); - - /* This triggers rather too much recalculation of Mesa state - * that doesn't get used (eg light positions). - */ - if (i >= VBO_ATTRIB_MAT_FRONT_AMBIENT && - i <= VBO_ATTRIB_MAT_BACK_INDEXES) - ctx->NewState |= _NEW_LIGHT; - - ctx->NewState |= _NEW_CURRENT_ATTRIB; - } + ctx->NewState |= _NEW_CURRENT_ATTRIB; } } @@ -222,7 +224,7 @@ static void vbo_exec_copy_to_current( struct vbo_exec_context *exec ) */ if (ctx->Light.ColorMaterialEnabled && exec->vtx.attrsz[VBO_ATTRIB_COLOR0]) { - _mesa_update_color_material(ctx, + _mesa_update_color_material(ctx, ctx->Current.Attrib[VBO_ATTRIB_COLOR0]); } } @@ -240,7 +242,8 @@ vbo_exec_copy_from_current(struct vbo_exec_context *exec) for (i = VBO_ATTRIB_POS + 1; i < VBO_ATTRIB_MAX; i++) { if (exec->vtx.attrtype[i] == GL_DOUBLE) { - memcpy(exec->vtx.attrptr[i], vbo->currval[i].Ptr, exec->vtx.attrsz[i] * sizeof(GLfloat)); + memcpy(exec->vtx.attrptr[i], vbo->currval[i].Ptr, + exec->vtx.attrsz[i] * sizeof(GLfloat)); } else { const fi_type *current = (fi_type *) vbo->currval[i].Ptr; switch (exec->vtx.attrsz[i]) { @@ -261,7 +264,7 @@ vbo_exec_copy_from_current(struct vbo_exec_context *exec) * to a larger one. Ex: glTexCoord2f -> glTexCoord4f. * We need to go back over the previous 2-component texcoords and insert * zero and one values. - */ + */ static void vbo_exec_wrap_upgrade_vertex(struct vbo_exec_context *exec, GLuint attr, GLuint newSize ) @@ -301,7 +304,7 @@ vbo_exec_wrap_upgrade_vertex(struct vbo_exec_context *exec, if (!_mesa_inside_begin_end(ctx) && !oldSize && lastcount > 8 && exec->vtx.vertex_size) { vbo_exec_copy_to_current( exec ); - reset_attrfv( exec ); + vbo_reset_all_attr(exec); } /* Fix up sizes: @@ -311,6 +314,7 @@ vbo_exec_wrap_upgrade_vertex(struct vbo_exec_context *exec, exec->vtx.max_vert = vbo_compute_max_verts(exec); exec->vtx.vert_count = 0; exec->vtx.buffer_ptr = exec->vtx.buffer_map; + exec->vtx.enabled |= BITFIELD64_BIT(attr); if (unlikely(oldSize)) { /* Size changed, recalculate all the attrptr[] values @@ -345,34 +349,34 @@ vbo_exec_wrap_upgrade_vertex(struct vbo_exec_context *exec, if (unlikely(exec->vtx.copied.nr)) { fi_type *data = exec->vtx.copied.buffer; fi_type *dest = exec->vtx.buffer_ptr; - GLuint j; assert(exec->vtx.buffer_ptr == exec->vtx.buffer_map); for (i = 0 ; i < exec->vtx.copied.nr ; i++) { - for (j = 0 ; j < VBO_ATTRIB_MAX ; j++) { + GLbitfield64 enabled = exec->vtx.enabled; + while (enabled) { + const int j = u_bit_scan64(&enabled); GLuint sz = exec->vtx.attrsz[j]; - - if (sz) { - GLint old_offset = old_attrptr[j] - exec->vtx.vertex; - GLint new_offset = exec->vtx.attrptr[j] - exec->vtx.vertex; - - if (j == attr) { - if (oldSize) { - fi_type tmp[4]; - COPY_CLEAN_4V_TYPE_AS_UNION(tmp, oldSize, - data + old_offset, - exec->vtx.attrtype[j]); - COPY_SZ_4V(dest + new_offset, newSize, tmp); - } else { - fi_type *current = (fi_type *)vbo->currval[j].Ptr; - COPY_SZ_4V(dest + new_offset, sz, current); - } - } - else { - COPY_SZ_4V(dest + new_offset, sz, data + old_offset); - } - } + GLint old_offset = old_attrptr[j] - exec->vtx.vertex; + GLint new_offset = exec->vtx.attrptr[j] - exec->vtx.vertex; + + assert(sz); + + if (j == attr) { + if (oldSize) { + fi_type tmp[4]; + COPY_CLEAN_4V_TYPE_AS_UNION(tmp, oldSize, + data + old_offset, + exec->vtx.attrtype[j]); + COPY_SZ_4V(dest + new_offset, newSize, tmp); + } else { + fi_type *current = (fi_type *)vbo->currval[j].Ptr; + COPY_SZ_4V(dest + new_offset, sz, current); + } + } + else { + COPY_SZ_4V(dest + new_offset, sz, data + old_offset); + } } data += old_vtx_size; @@ -414,16 +418,17 @@ vbo_exec_fixup_vertex(struct gl_context *ctx, GLuint attr, * zeros. Don't need to flush or wrap. */ for (i = newSize; i <= exec->vtx.attrsz[attr]; i++) - exec->vtx.attrptr[attr][i-1] = id[i-1]; + exec->vtx.attrptr[attr][i-1] = id[i-1]; } exec->vtx.active_sz[attr] = newSize; + exec->vtx.attrtype[attr] = newType; /* Does setting NeedFlush belong here? Necessitates resetting * vtxfmt on each flush (otherwise flags won't get reset * afterwards). */ - if (attr == 0) + if (attr == 0) ctx->Driver.NeedFlush |= FLUSH_STORED_VERTICES; } @@ -474,7 +479,7 @@ do { \ if (N>1) dest[1] = V1; \ if (N>2) dest[2] = V2; \ if (N>3) dest[3] = V3; \ - exec->vtx.attrtype[A] = T; \ + assert(exec->vtx.attrtype[A] == T); \ } \ \ if ((A) == 0) { \ @@ -508,6 +513,8 @@ do { \ } \ } while (0) + +#undef ERROR #define ERROR(err) _mesa_error( ctx, err, __func__ ) #define TAG(x) vbo_##x @@ -626,30 +633,31 @@ vbo_exec_FlushVertices_internal(struct vbo_exec_context *exec, GLboolean unmap) if (exec->vtx.vertex_size) { vbo_exec_copy_to_current( exec ); - reset_attrfv( exec ); + vbo_reset_all_attr(exec); } } -static void GLAPIENTRY vbo_exec_EvalCoord1f( GLfloat u ) +static void GLAPIENTRY +vbo_exec_EvalCoord1f(GLfloat u) { GET_CURRENT_CONTEXT( ctx ); struct vbo_exec_context *exec = &vbo_context(ctx)->exec; { GLint i; - if (exec->eval.recalculate_maps) + if (exec->eval.recalculate_maps) vbo_exec_eval_update( exec ); for (i = 0; i <= VBO_ATTRIB_TEX7; i++) { - if (exec->eval.map1[i].map) + if (exec->eval.map1[i].map) if (exec->vtx.active_sz[i] != exec->eval.map1[i].sz) vbo_exec_fixup_vertex( ctx, i, exec->eval.map1[i].sz, GL_FLOAT ); } } - memcpy( exec->vtx.copied.buffer, exec->vtx.vertex, + memcpy( exec->vtx.copied.buffer, exec->vtx.vertex, exec->vtx.vertex_size * sizeof(GLfloat)); vbo_exec_do_EvalCoord1f( exec, u ); @@ -658,47 +666,55 @@ static void GLAPIENTRY vbo_exec_EvalCoord1f( GLfloat u ) exec->vtx.vertex_size * sizeof(GLfloat)); } -static void GLAPIENTRY vbo_exec_EvalCoord2f( GLfloat u, GLfloat v ) + +static void GLAPIENTRY +vbo_exec_EvalCoord2f(GLfloat u, GLfloat v) { GET_CURRENT_CONTEXT( ctx ); struct vbo_exec_context *exec = &vbo_context(ctx)->exec; { GLint i; - if (exec->eval.recalculate_maps) + if (exec->eval.recalculate_maps) vbo_exec_eval_update( exec ); for (i = 0; i <= VBO_ATTRIB_TEX7; i++) { - if (exec->eval.map2[i].map) + if (exec->eval.map2[i].map) if (exec->vtx.active_sz[i] != exec->eval.map2[i].sz) vbo_exec_fixup_vertex( ctx, i, exec->eval.map2[i].sz, GL_FLOAT ); } - if (ctx->Eval.AutoNormal) + if (ctx->Eval.AutoNormal) if (exec->vtx.active_sz[VBO_ATTRIB_NORMAL] != 3) vbo_exec_fixup_vertex( ctx, VBO_ATTRIB_NORMAL, 3, GL_FLOAT ); } - memcpy( exec->vtx.copied.buffer, exec->vtx.vertex, + memcpy( exec->vtx.copied.buffer, exec->vtx.vertex, exec->vtx.vertex_size * sizeof(GLfloat)); vbo_exec_do_EvalCoord2f( exec, u, v ); - memcpy( exec->vtx.vertex, exec->vtx.copied.buffer, + memcpy( exec->vtx.vertex, exec->vtx.copied.buffer, exec->vtx.vertex_size * sizeof(GLfloat)); } -static void GLAPIENTRY vbo_exec_EvalCoord1fv( const GLfloat *u ) + +static void GLAPIENTRY +vbo_exec_EvalCoord1fv(const GLfloat *u) { vbo_exec_EvalCoord1f( u[0] ); } -static void GLAPIENTRY vbo_exec_EvalCoord2fv( const GLfloat *u ) + +static void GLAPIENTRY +vbo_exec_EvalCoord2fv(const GLfloat *u) { vbo_exec_EvalCoord2f( u[0], u[1] ); } -static void GLAPIENTRY vbo_exec_EvalPoint1( GLint i ) + +static void GLAPIENTRY +vbo_exec_EvalPoint1(GLint i) { GET_CURRENT_CONTEXT( ctx ); GLfloat du = ((ctx->Eval.MapGrid1u2 - ctx->Eval.MapGrid1u1) / @@ -709,12 +725,13 @@ static void GLAPIENTRY vbo_exec_EvalPoint1( GLint i ) } -static void GLAPIENTRY vbo_exec_EvalPoint2( GLint i, GLint j ) +static void GLAPIENTRY +vbo_exec_EvalPoint2(GLint i, GLint j) { GET_CURRENT_CONTEXT( ctx ); - GLfloat du = ((ctx->Eval.MapGrid2u2 - ctx->Eval.MapGrid2u1) / + GLfloat du = ((ctx->Eval.MapGrid2u2 - ctx->Eval.MapGrid2u1) / (GLfloat) ctx->Eval.MapGrid2un); - GLfloat dv = ((ctx->Eval.MapGrid2v2 - ctx->Eval.MapGrid2v1) / + GLfloat dv = ((ctx->Eval.MapGrid2v2 - ctx->Eval.MapGrid2v1) / (GLfloat) ctx->Eval.MapGrid2vn); GLfloat u = i * du + ctx->Eval.MapGrid2u1; GLfloat v = j * dv + ctx->Eval.MapGrid2v1; @@ -726,9 +743,10 @@ static void GLAPIENTRY vbo_exec_EvalPoint2( GLint i, GLint j ) /** * Called via glBegin. */ -static void GLAPIENTRY vbo_exec_Begin( GLenum mode ) +static void GLAPIENTRY +vbo_exec_Begin(GLenum mode) { - GET_CURRENT_CONTEXT( ctx ); + GET_CURRENT_CONTEXT( ctx ); struct vbo_exec_context *exec = &vbo_context(ctx)->exec; int i; @@ -819,9 +837,10 @@ try_vbo_merge(struct vbo_exec_context *exec) /** * Called via glEnd. */ -static void GLAPIENTRY vbo_exec_End( void ) +static void GLAPIENTRY +vbo_exec_End(void) { - GET_CURRENT_CONTEXT( ctx ); + GET_CURRENT_CONTEXT( ctx ); struct vbo_exec_context *exec = &vbo_context(ctx)->exec; if (!_mesa_inside_begin_end(ctx)) { @@ -887,7 +906,7 @@ static void GLAPIENTRY vbo_exec_PrimitiveRestartNV(void) { GLenum curPrim; - GET_CURRENT_CONTEXT( ctx ); + GET_CURRENT_CONTEXT(ctx); curPrim = ctx->Driver.CurrentExecPrimitive; @@ -901,8 +920,8 @@ vbo_exec_PrimitiveRestartNV(void) } - -static void vbo_exec_vtxfmt_init( struct vbo_exec_context *exec ) +static void +vbo_exec_vtxfmt_init(struct vbo_exec_context *exec) { struct gl_context *ctx = exec->ctx; GLvertexformat *vfmt = &exec->vtxfmt; @@ -957,7 +976,7 @@ static void vbo_exec_vtxfmt_init( struct vbo_exec_context *exec ) vfmt->Vertex3fv = vbo_Vertex3fv; vfmt->Vertex4f = vbo_Vertex4f; vfmt->Vertex4fv = vbo_Vertex4fv; - + if (ctx->API == API_OPENGLES2) { vfmt->VertexAttrib1fARB = _es_VertexAttrib1f; vfmt->VertexAttrib1fvARB = _es_VertexAttrib1fv; @@ -1040,7 +1059,7 @@ static void vbo_exec_vtxfmt_init( struct vbo_exec_context *exec ) vfmt->MultiTexCoordP3uiv = vbo_MultiTexCoordP3uiv; vfmt->MultiTexCoordP4ui = vbo_MultiTexCoordP4ui; vfmt->MultiTexCoordP4uiv = vbo_MultiTexCoordP4uiv; - + vfmt->NormalP3ui = vbo_NormalP3ui; vfmt->NormalP3uiv = vbo_NormalP3uiv; @@ -1079,7 +1098,8 @@ static void vbo_exec_vtxfmt_init( struct vbo_exec_context *exec ) * This replaces the malloced buffer which was created in * vb_exec_vtx_init() below. */ -void vbo_use_buffer_objects(struct gl_context *ctx) +void +vbo_use_buffer_objects(struct gl_context *ctx) { struct vbo_exec_context *exec = &vbo_context(ctx)->exec; /* Any buffer name but 0 can be used here since this bufferobj won't @@ -1124,7 +1144,8 @@ vbo_always_unmap_buffers(struct gl_context *ctx) } -void vbo_exec_vtx_init( struct vbo_exec_context *exec ) +void +vbo_exec_vtx_init(struct vbo_exec_context *exec) { struct gl_context *ctx = exec->ctx; struct vbo_context *vbo = vbo_context(ctx); @@ -1145,6 +1166,7 @@ void vbo_exec_vtx_init( struct vbo_exec_context *exec ) vbo_exec_vtxfmt_init( exec ); _mesa_noop_vtxfmt_init(&exec->vtxfmt_noop); + exec->vtx.enabled = 0; for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) { assert(i < ARRAY_SIZE(exec->vtx.attrsz)); exec->vtx.attrsz[i] = 0; @@ -1158,7 +1180,7 @@ void vbo_exec_vtx_init( struct vbo_exec_context *exec ) assert(i < ARRAY_SIZE(exec->vtx.arrays)); exec->vtx.inputs[i] = &exec->vtx.arrays[i]; } - + { struct gl_client_array *arrays = exec->vtx.arrays; unsigned i; @@ -1192,7 +1214,8 @@ void vbo_exec_vtx_init( struct vbo_exec_context *exec ) } -void vbo_exec_vtx_destroy( struct vbo_exec_context *exec ) +void +vbo_exec_vtx_destroy(struct vbo_exec_context *exec) { /* using a real VBO for vertex data */ struct gl_context *ctx = exec->ctx; @@ -1238,7 +1261,8 @@ void vbo_exec_vtx_destroy( struct vbo_exec_context *exec ) * * \param flags bitmask of FLUSH_STORED_VERTICES, FLUSH_UPDATE_CURRENT */ -void vbo_exec_FlushVertices( struct gl_context *ctx, GLuint flags ) +void +vbo_exec_FlushVertices(struct gl_context *ctx, GLuint flags) { struct vbo_exec_context *exec = &vbo_context(ctx)->exec; @@ -1271,19 +1295,29 @@ void vbo_exec_FlushVertices( struct gl_context *ctx, GLuint flags ) } -static void reset_attrfv( struct vbo_exec_context *exec ) -{ - GLuint i; +/** + * Reset the vertex attribute by setting its size to zero. + */ +static void +vbo_reset_attr(struct vbo_exec_context *exec, GLuint attr) +{ + exec->vtx.attrsz[attr] = 0; + exec->vtx.attrtype[attr] = GL_FLOAT; + exec->vtx.active_sz[attr] = 0; +} - for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) { - exec->vtx.attrsz[i] = 0; - exec->vtx.attrtype[i] = GL_FLOAT; - exec->vtx.active_sz[i] = 0; + +static void +vbo_reset_all_attr(struct vbo_exec_context *exec) +{ + while (exec->vtx.enabled) { + const int i = u_bit_scan64(&exec->vtx.enabled); + vbo_reset_attr(exec, i); } exec->vtx.vertex_size = 0; } - + void GLAPIENTRY _es_Color4f(GLfloat r, GLfloat g, GLfloat b, GLfloat a)