summaryrefslogtreecommitdiffstats
path: root/src/glx/x11/indirect_vertex_array.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/glx/x11/indirect_vertex_array.c')
-rw-r--r--src/glx/x11/indirect_vertex_array.c1878
1 files changed, 1878 insertions, 0 deletions
diff --git a/src/glx/x11/indirect_vertex_array.c b/src/glx/x11/indirect_vertex_array.c
new file mode 100644
index 00000000000..8bc7e449021
--- /dev/null
+++ b/src/glx/x11/indirect_vertex_array.c
@@ -0,0 +1,1878 @@
+/*
+ * (C) Copyright IBM Corporation 2004, 2005
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sub license,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * IBM,
+ * AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
+ * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <inttypes.h>
+#include <assert.h>
+#include "glxclient.h"
+#include "packrender.h"
+#include "indirect.h"
+#include <string.h>
+
+#include <GL/glxproto.h>
+#include "glxextensions.h"
+#include "indirect_vertex_array.h"
+
+/**
+ * \file indirect_vertex_array.c
+ * Implement GLX protocol for vertex arrays and vertex buffer objects.
+ *
+ * The most important function in this fill is \c fill_array_info_cache.
+ * The \c array_state_vector contains a cache of the ARRAY_INFO data sent
+ * in the DrawArrays protocol. Certain operations, such as enabling or
+ * disabling an array, can invalidate this cache. \c fill_array_info_cache
+ * fills-in this data. Additionally, it examines the enabled state and
+ * other factors to determine what "version" of DrawArrays protocoal can be
+ * used.
+ *
+ * Current, only two versions of DrawArrays protocol are implemented. The
+ * first version is the "none" protocol. This is the fallback when the
+ * server does not support GL 1.1 / EXT_vertex_arrays. It is implemented
+ * by sending batches of immediate mode commands that are equivalent to the
+ * DrawArrays protocol.
+ *
+ * The other protocol that is currently implemented is the "old" protocol.
+ * This is the GL 1.1 DrawArrays protocol. The only difference between GL
+ * 1.1 and EXT_vertex_arrays is the opcode used for the DrawArrays command.
+ * This protocol is called "old" because the ARB is in the process of
+ * defining a new protocol, which will probably be called wither "new" or
+ * "vbo", to support multiple texture coordinate arrays, generic attributes,
+ * and vertex buffer objects.
+ *
+ * \author Ian Romanick <[email protected]>
+ */
+
+
+/**
+ * State descriptor for a single array of vertex data.
+ */
+struct array_state {
+ /**
+ * Pointer to the application supplied data.
+ */
+ const void * data;
+
+ /**
+ * Enum representing the type of the application supplied data.
+ */
+ GLenum data_type;
+
+ /**
+ * Stride value supplied by the application. This value is not used
+ * internally. It is only kept so that it can be queried by the
+ * application using glGet*v.
+ */
+ GLsizei user_stride;
+
+ /**
+ * Calculated size, in bytes, of a single element in the array. This
+ * is calculated based on \c count and the size of the data type
+ * represented by \c data_type.
+ */
+ GLsizei element_size;
+
+ /**
+ * Actual byte-stride from one element to the next. This value will
+ * be equal to either \c user_stride or \c element_stride.
+ */
+ GLsizei true_stride;
+
+ /**
+ * Number of data values in each element.
+ */
+ GLint count;
+
+ /**
+ * Pre-calculated GLX protocol command header.
+ */
+ uint32_t header[2];
+
+ /**
+ * Size of the header data. For simple data, like glColorPointerfv,
+ * this is 4. For complex data that requires either a count (e.g.,
+ * glWeightfvARB), an index (e.g., glVertexAttrib1fvARB), or a
+ * selector enum (e.g., glMultiTexCoord2fv) this is 8.
+ */
+ unsigned header_size;
+
+ /**
+ * Set to \c GL_TRUE if this array is enabled. Otherwise, it is set
+ * to \c GL_FALSE.
+ */
+ GLboolean enabled;
+
+ /**
+ * For multi-arrayed data (e.g., texture coordinates, generic vertex
+ * program attributes, etc.), this specifies which array this is.
+ */
+ unsigned index;
+
+ /**
+ * Per-array-type key. For most arrays, this will be the GL enum for
+ * that array (e.g., GL_VERTEX_ARRAY for vertex data, GL_NORMAL_ARRAY
+ * for normal data, GL_TEXTURE_COORD_ARRAY for texture coordinate data,
+ * etc.).
+ */
+ GLenum key;
+
+ /**
+ * If this array can be used with the "classic" \c glDrawArrays protocol,
+ * this is set to \c GL_TRUE. Otherwise, it is set to \c GL_FALSE.
+ */
+ GLboolean old_DrawArrays_possible;
+};
+
+
+/**
+ * Array state that is pushed / poped by \c glPushClientAttrib and
+ * \c glPopClientAttrib.
+ */
+struct array_stack_state {
+ /**
+ * Pointer to the application supplied data.
+ */
+ const void * data;
+
+ /**
+ * Enum representing the type of the application supplied data.
+ */
+ GLenum data_type;
+
+ /**
+ * Stride value supplied by the application. This value is not used
+ * internally. It is only kept so that it can be queried by the
+ * application using glGet*v.
+ */
+ GLsizei user_stride;
+
+ /**
+ * Number of data values in each element.
+ */
+ GLint count;
+
+ /**
+ * Per-array-type key. For most arrays, this will be the GL enum for
+ * that array (e.g., GL_VERTEX_ARRAY for vertex data, GL_NORMAL_ARRAY
+ * for normal data, GL_TEXTURE_COORD_ARRAY for texture coordinate data,
+ * etc.).
+ */
+ GLenum key;
+
+ /**
+ * For multi-arrayed data (e.g., texture coordinates, generic vertex
+ * program attributes, etc.), this specifies which array this is.
+ */
+ unsigned index;
+
+ /**
+ * Set to \c GL_TRUE if this array is enabled. Otherwise, it is set
+ * to \c GL_FALSE.
+ */
+ GLboolean enabled;
+};
+
+
+/**
+ * Collection of all the vertex array state.
+ */
+struct array_state_vector {
+ /**
+ * Number of arrays tracked by \c ::arrays.
+ */
+ size_t num_arrays;
+
+ /**
+ * Array of vertex array state. This array contains all of the valid
+ * vertex arrays. If a vertex array isn't in this array, then it isn't
+ * valid. For example, if an implementation does not support
+ * EXT_fog_coord, there won't be a GL_FOG_COORD_ARRAY entry in this
+ * array.
+ */
+ struct array_state * arrays;
+
+ /**
+ * Number of currently enabled arrays. The value of this field is
+ * only valid if \c array_info_cache_valid is true.
+ */
+ size_t enabled_array_count;
+
+ /**
+ * \name ARRAY_INFO cache.
+ *
+ * These fields track the state of the ARRAY_INFO cache. The
+ * \c array_info_cache_size is the size of the actual data stored in
+ * \c array_info_cache. \c array_info_cache_buffer_size is the size of
+ * the buffer. This will always be greater than or equal to
+ * \c array_info_cache_size.
+ *
+ * \c large_header doesn't completely belong in this group. This is a
+ * pointer to a buffer to hold the header information for DrawArrays in
+ * a RenderLarge command. This buffer is immediately before
+ * \c array_info_cache. The idea is that the header data will be written
+ * to \c large_header and a single call to \c __glXSendLargeChunk can be
+ * made to send the header and the ARRAY_INFO data.
+ *
+ * \note
+ * \c array_info_cache_size and \c array_info_cache_buffer_size do
+ * NOT include the size of \c large_header.
+ */
+ /*@{*/
+ size_t array_info_cache_size;
+ size_t array_info_cache_buffer_size;
+ void * array_info_cache;
+ GLubyte * large_header;
+ /*@}*/
+
+
+ /**
+ * Is the cache of ARRAY_INFO data valid? The cache can become invalid
+ * when one of several state changes occur. Among these chages are
+ * modifying the array settings for an enabled array and enabling /
+ * disabling an array.
+ */
+ GLboolean array_info_cache_valid;
+
+ /**
+ * Is it possible to use the GL 1.1 / EXT_vertex_arrays protocol? Use
+ * of this protocol is disabled with really old servers (i.e., servers
+ * that don't support GL 1.1 or EXT_vertex_arrays) or when an environment
+ * variable is set.
+ *
+ * \todo
+ * GL 1.1 and EXT_vertex_arrays use identical protocol, but have different
+ * opcodes for \c glDrawArrays. For servers that advertise one or the
+ * other, there should be a way to select which opcode to use.
+ */
+ GLboolean old_DrawArrays_possible;
+
+ /**
+ * Is it possible to use the new GL X.X / ARB_vertex_buffer_object
+ * protocol?
+ *
+ * \todo
+ * This protocol has not yet been defined by the ARB, but is currently a
+ * work in progress. This field is a place-holder.
+ */
+ GLboolean new_DrawArrays_possible;
+
+ /**
+ * Active texture unit set by \c glClientActiveTexture.
+ *
+ * \sa __glXGetActiveTextureUnit
+ */
+ unsigned active_texture_unit;
+
+ /**
+ * Number of supported texture units. Even if ARB_multitexture /
+ * GL 1.3 are not supported, this will be at least 1. When multitexture
+ * is supported, this will be the value queried by calling
+ * \c glGetIntegerv with \c GL_MAX_TEXTURE_UNITS.
+ *
+ * \todo
+ * Investigate if this should be the value of \c GL_MAX_TEXTURE_COORDS
+ * instead (if GL 2.0 / ARB_fragment_shader / ARB_fragment_program /
+ * NV_fragment_program are supported).
+ */
+ unsigned num_texture_units;
+
+ /**
+ * \n Methods for implementing various GL functions.
+ *
+ * These method pointers are only valid \c array_info_cache_valid is set.
+ * When each function starts, it much check \c array_info_cache_valid.
+ * If it is not set, it must call \c fill_array_info_cache and call
+ * the new method.
+ *
+ * \sa fill_array_info_cache
+ *
+ * \todo
+ * Write code to plug these functions directly into the dispatch table.
+ */
+ /*@{*/
+ void (*DrawArrays)( GLenum, GLint, GLsizei );
+ void (*DrawElements)( GLenum mode, GLsizei count, GLenum type,
+ const GLvoid *indices );
+ /*@}*/
+
+ struct array_stack_state * stack;
+ unsigned active_texture_unit_stack[ __GL_CLIENT_ATTRIB_STACK_DEPTH ];
+ unsigned stack_index;
+};
+
+
+static void emit_DrawArrays_none( GLenum mode, GLint first, GLsizei count );
+static void emit_DrawArrays_old ( GLenum mode, GLint first, GLsizei count );
+
+static void emit_DrawElements_none( GLenum mode, GLsizei count, GLenum type,
+ const GLvoid *indices );
+static void emit_DrawElements_old ( GLenum mode, GLsizei count, GLenum type,
+ const GLvoid *indices );
+
+
+static GLubyte * emit_element_none( GLubyte * dst,
+ const struct array_state_vector * arrays, unsigned index );
+static GLubyte * emit_element_old( GLubyte * dst,
+ const struct array_state_vector * arrays, unsigned index );
+static struct array_state * get_array_entry(
+ const struct array_state_vector * arrays, GLenum key, unsigned index );
+static void fill_array_info_cache( struct array_state_vector * arrays );
+static GLboolean glx_validate_array_args(__GLXcontext *gc, GLenum mode,
+ GLsizei count);
+
+
+/**
+ * Table of sizes, in bytes, of a GL types. All of the type enums are be in
+ * the range 0x1400 - 0x140F. That includes types added by extensions (i.e.,
+ * \c GL_HALF_FLOAT_NV). This elements of this table correspond to the
+ * type enums masked with 0x0f.
+ *
+ * \notes
+ * \c GL_HALF_FLOAT_NV is not included. Neither are \c GL_2_BYTES,
+ * \c GL_3_BYTES, or \c GL_4_BYTES.
+ */
+const GLuint __glXTypeSize_table[16] = {
+ 1, 1, 2, 2, 4, 4, 4, 0, 0, 0, 8, 0, 0, 0, 0, 0
+};
+
+
+
+/**
+ * Initialize vertex array state of a GLX context.
+ *
+ * \param gc GLX context whose vertex array state is to be initialized.
+ *
+ * \warning
+ * This function may only be called after __GLXcontext::gl_extension_bits,
+ * __GLXcontext::server_minor, and __GLXcontext::server_major have been
+ * initialized. These values are used to determine what vertex arrays are
+ * supported.
+ *
+ * \bug
+ * Return values from malloc are not properly tested.
+ */
+void
+__glXInitVertexArrayState( __GLXcontext * gc )
+{
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays;
+
+ unsigned array_count;
+ unsigned texture_units = 1;
+ unsigned i;
+
+ GLboolean got_fog = GL_FALSE;
+ GLboolean got_secondary_color = GL_FALSE;
+
+
+ arrays = malloc( sizeof( struct array_state_vector ) );
+ state->array_state = arrays;
+ arrays->enabled_array_count = 0;
+ arrays->array_info_cache = NULL;
+ arrays->array_info_cache_size = 0;
+ arrays->array_info_cache_buffer_size = 0;
+ arrays->array_info_cache_valid= GL_FALSE;
+
+ arrays->old_DrawArrays_possible = !state->NoDrawArraysProtocol;
+ arrays->new_DrawArrays_possible = GL_FALSE;
+ arrays->DrawArrays = NULL;
+
+ arrays->active_texture_unit = 0;
+
+
+ /* Determine how many arrays are actually needed. Only arrays that
+ * are supported by the server are create. For example, if the server
+ * supports only 2 texture units, then only 2 texture coordinate arrays
+ * are created.
+ *
+ * At the very least, GL_VERTEX_ARRAY, GL_NORMAL_ARRAY,
+ * GL_COLOR_ARRAY, GL_INDEX_ARRAY, GL_TEXTURE_COORD_ARRAY, and
+ * GL_EDGE_FLAG_ARRAY are supported.
+ */
+
+ array_count = 5;
+
+ if ( __glExtensionBitIsEnabled( gc, GL_EXT_fog_coord_bit )
+ || (gc->server_major > 1) || (gc->server_minor >= 4) ) {
+ got_fog = GL_TRUE;
+ array_count++;
+ }
+
+ if ( __glExtensionBitIsEnabled( gc, GL_EXT_secondary_color_bit )
+ || (gc->server_major > 1) || (gc->server_minor >= 4) ) {
+ got_secondary_color = GL_TRUE;
+ array_count++;
+ }
+
+ if ( __glExtensionBitIsEnabled( gc, GL_ARB_multitexture_bit )
+ || (gc->server_major > 1) || (gc->server_minor >= 3) ) {
+ glGetIntegerv( GL_MAX_TEXTURE_UNITS, & texture_units );
+ }
+ else {
+ texture_units = 1;
+ }
+
+ arrays->num_texture_units = texture_units;
+ array_count += texture_units;
+ arrays->num_arrays = array_count;
+ arrays->arrays = malloc( sizeof( struct array_state ) * array_count );
+
+ (void) memset( arrays->arrays, 0,
+ sizeof( struct array_state ) * array_count );
+
+
+ arrays->arrays[0].data_type = GL_FLOAT;
+ arrays->arrays[0].count = 3;
+ arrays->arrays[0].key = GL_NORMAL_ARRAY;
+ arrays->arrays[0].old_DrawArrays_possible = GL_TRUE;
+
+ arrays->arrays[1].data_type = GL_FLOAT;
+ arrays->arrays[1].count = 4;
+ arrays->arrays[1].key = GL_COLOR_ARRAY;
+ arrays->arrays[1].old_DrawArrays_possible = GL_TRUE;
+
+ arrays->arrays[2].data_type = GL_FLOAT;
+ arrays->arrays[2].count = 1;
+ arrays->arrays[2].key = GL_INDEX_ARRAY;
+ arrays->arrays[2].old_DrawArrays_possible = GL_TRUE;
+
+ arrays->arrays[3].data_type = GL_UNSIGNED_BYTE;
+ arrays->arrays[3].count = 1;
+ arrays->arrays[3].key = GL_EDGE_FLAG_ARRAY;
+ arrays->arrays[3].old_DrawArrays_possible = GL_TRUE;
+
+ for ( i = 0 ; i < texture_units ; i++ ) {
+ arrays->arrays[4 + i].data_type = GL_FLOAT;
+ arrays->arrays[4 + i].count = 4;
+ arrays->arrays[4 + i].key = GL_TEXTURE_COORD_ARRAY;
+
+ arrays->arrays[4 + i].old_DrawArrays_possible = (i == 0);
+ arrays->arrays[4 + i].index = i;
+
+ arrays->arrays[4 + i].header[1] = i + GL_TEXTURE0;
+ }
+
+ i = 4 + texture_units;
+
+ if ( got_fog ) {
+ arrays->arrays[i].data_type = GL_FLOAT;
+ arrays->arrays[i].count = 1;
+ arrays->arrays[i].key = GL_FOG_COORDINATE_ARRAY;
+ arrays->arrays[i].old_DrawArrays_possible = GL_TRUE;
+ i++;
+ }
+
+ if ( got_secondary_color ) {
+ arrays->arrays[i].data_type = GL_FLOAT;
+ arrays->arrays[i].count = 3;
+ arrays->arrays[i].key = GL_SECONDARY_COLOR_ARRAY;
+ arrays->arrays[i].old_DrawArrays_possible = GL_TRUE;
+ i++;
+ }
+
+
+ /* Vertex array *must* be last becuase of the way that
+ * emit_DrawArrays_none works.
+ */
+
+ arrays->arrays[i].data_type = GL_FLOAT;
+ arrays->arrays[i].count = 4;
+ arrays->arrays[i].key = GL_VERTEX_ARRAY;
+ arrays->arrays[i].old_DrawArrays_possible = GL_TRUE;
+
+
+ arrays->stack_index = 0;
+ arrays->stack = malloc( sizeof( struct array_stack_state )
+ * arrays->num_arrays );
+}
+
+
+/**
+ * Calculate the size of a single vertex for the "none" protocol. This is
+ * essentially the size of all the immediate-mode commands required to
+ * implement the enabled vertex arrays.
+ */
+static size_t
+calculate_single_vertex_size_none( const struct array_state_vector * arrays )
+{
+ size_t single_vertex_size = 0;
+ unsigned i;
+
+
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ if ( arrays->arrays[i].enabled ) {
+ single_vertex_size += __GLX_PAD(arrays->arrays[i].element_size)
+ + arrays->arrays[i].header_size;
+ }
+ }
+
+ return single_vertex_size;
+}
+
+
+/**
+ * Emit a single element using non-DrawArrays protocol.
+ */
+GLubyte *
+emit_element_none( GLubyte * dst,
+ const struct array_state_vector * arrays,
+ unsigned index )
+{
+ unsigned i;
+
+
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ if ( arrays->arrays[i].enabled ) {
+ const size_t offset = index * arrays->arrays[i].true_stride;
+
+ (void) memcpy( dst, arrays->arrays[i].header,
+ arrays->arrays[i].header_size );
+
+ dst += arrays->arrays[i].header_size;
+
+ (void) memcpy( dst, ((GLubyte *) arrays->arrays[i].data) + offset,
+ arrays->arrays[i].element_size );
+
+ dst += __GLX_PAD( arrays->arrays[i].element_size );
+ }
+ }
+
+ return dst;
+}
+
+
+/**
+ * Emit a single element using "old" DrawArrays protocol from
+ * EXT_vertex_arrays / OpenGL 1.1.
+ */
+GLubyte *
+emit_element_old( GLubyte * dst,
+ const struct array_state_vector * arrays,
+ unsigned index )
+{
+ unsigned i;
+
+
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ if ( arrays->arrays[i].enabled ) {
+ const size_t offset = index * arrays->arrays[i].true_stride;
+
+ (void) memcpy( dst, ((GLubyte *) arrays->arrays[i].data) + offset,
+ arrays->arrays[i].element_size );
+
+ dst += __GLX_PAD( arrays->arrays[i].element_size );
+ }
+ }
+
+ return dst;
+}
+
+
+struct array_state *
+get_array_entry( const struct array_state_vector * arrays,
+ GLenum key, unsigned index )
+{
+ unsigned i;
+
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ if ( (arrays->arrays[i].key == key)
+ && (arrays->arrays[i].index == index) ) {
+ return & arrays->arrays[i];
+ }
+ }
+
+ return NULL;
+}
+
+
+static GLboolean
+allocate_array_info_cache( struct array_state_vector * arrays,
+ size_t required_size )
+{
+ if ( arrays->array_info_cache_buffer_size < required_size ) {
+ GLubyte * temp = realloc( arrays->array_info_cache, required_size + 20 );
+
+ if ( temp == NULL ) {
+ return GL_FALSE;
+ }
+
+ arrays->large_header = temp;
+ arrays->array_info_cache = temp + 20;
+ arrays->array_info_cache_buffer_size = required_size;
+ }
+
+ arrays->array_info_cache_size = required_size;
+ return GL_TRUE;
+}
+
+
+/**
+ */
+void
+fill_array_info_cache( struct array_state_vector * arrays )
+{
+ GLboolean old_DrawArrays_possible;
+ unsigned i;
+
+
+ /* Determine how many arrays are enabled.
+ */
+
+ arrays->enabled_array_count = 0;
+ old_DrawArrays_possible = arrays->old_DrawArrays_possible;
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ if ( arrays->arrays[i].enabled ) {
+ arrays->enabled_array_count++;
+ old_DrawArrays_possible &= arrays->arrays[i].old_DrawArrays_possible;
+ }
+ }
+
+
+ if ( arrays->new_DrawArrays_possible ) {
+ assert( ! arrays->new_DrawArrays_possible );
+ }
+ else if ( old_DrawArrays_possible ) {
+ const size_t required_size = arrays->enabled_array_count * 12;
+ uint32_t * info;
+
+
+ if ( ! allocate_array_info_cache( arrays, required_size ) ) {
+ return;
+ }
+
+
+ info = (uint32_t *) arrays->array_info_cache;
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ if ( arrays->arrays[i].enabled ) {
+ *(info++) = arrays->arrays[i].data_type;
+ *(info++) = arrays->arrays[i].count;
+ *(info++) = arrays->arrays[i].key;
+ }
+ }
+
+ arrays->array_info_cache_valid = GL_TRUE;
+ arrays->DrawArrays = emit_DrawArrays_old;
+ arrays->DrawElements = emit_DrawElements_old;
+ }
+ else {
+ arrays->DrawArrays = emit_DrawArrays_none;
+ arrays->DrawElements = emit_DrawElements_none;
+ }
+}
+
+
+/**
+ * Emit a \c glDrawArrays command using the "none" protocol. That is,
+ * emit immediate-mode commands that are equivalent to the requiested
+ * \c glDrawArrays command. This is used with servers that don't support
+ * the OpenGL 1.1 / EXT_vertex_arrays DrawArrays protocol or in cases where
+ * vertex state is enabled that is not compatible with that protocol.
+ */
+void
+emit_DrawArrays_none( GLenum mode, GLint first, GLsizei count )
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+
+ size_t single_vertex_size;
+ GLubyte * pc;
+ unsigned i;
+ static const uint16_t begin_cmd[2] = { 8, X_GLrop_Begin };
+ static const uint16_t end_cmd[2] = { 4, X_GLrop_End };
+
+
+ single_vertex_size = calculate_single_vertex_size_none( arrays );
+
+ pc = gc->pc;
+
+ (void) memcpy( pc, begin_cmd, 4 );
+ *(int *)(pc + 4) = mode;
+
+ pc += 8;
+
+ for ( i = 0 ; i < count ; i++ ) {
+ if ( (pc + single_vertex_size) >= gc->bufEnd ) {
+ pc = __glXFlushRenderBuffer(gc, gc->pc);
+ }
+
+ pc = emit_element_none( pc, arrays, first + i );
+ }
+
+ if ( (pc + 4) >= gc->bufEnd ) {
+ pc = __glXFlushRenderBuffer(gc, gc->pc);
+ }
+
+ (void) memcpy( pc, end_cmd, 4 );
+ pc += 4;
+
+ gc->pc = pc;
+ if ( gc->pc > gc->limit ) {
+ (void) __glXFlushRenderBuffer(gc, gc->pc);
+ }
+}
+
+
+/**
+ * Emit the header data for the GL 1.1 / EXT_vertex_arrays DrawArrays
+ * protocol.
+ *
+ * \param gc GLX context.
+ * \param arrays Array state.
+ * \param elements_per_request Location to store the number of elements that
+ * can fit in a single Render / RenderLarge
+ * command.
+ * \param total_request Total number of requests for a RenderLarge
+ * command. If a Render command is used, this
+ * will be zero.
+ * \param mode Drawing mode.
+ * \param count Number of vertices.
+ *
+ * \returns
+ * A pointer to the buffer for array data.
+ */
+static GLubyte *
+emit_DrawArrays_header_old( __GLXcontext * gc,
+ struct array_state_vector * arrays,
+ size_t * elements_per_request,
+ size_t * total_requests,
+ GLenum mode, GLsizei count )
+{
+ size_t command_size;
+ size_t single_vertex_size;
+ const unsigned header_size = 16;
+ unsigned i;
+ GLubyte * pc;
+
+
+ /* Determine the size of the whole command. This includes the header,
+ * the ARRAY_INFO data and the array data. Once this size is calculated,
+ * it will be known whether a Render or RenderLarge command is needed.
+ */
+
+ single_vertex_size = 0;
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ if ( arrays->arrays[i].enabled ) {
+ single_vertex_size += __GLX_PAD( arrays->arrays[i].element_size );
+ }
+ }
+
+ command_size = arrays->array_info_cache_size + header_size
+ + (single_vertex_size * count);
+
+
+ /* Write the header for either a Render command or a RenderLarge
+ * command. After the header is written, write the ARRAY_INFO data.
+ */
+
+ if ( command_size > gc->maxSmallRenderCommandSize ) {
+ /* maxSize is the maximum amount of data can be stuffed into a single
+ * packet. sz_xGLXRenderReq is added because bufSize is the maximum
+ * packet size minus sz_xGLXRenderReq.
+ */
+ const size_t maxSize = (gc->bufSize + sz_xGLXRenderReq)
+ - sz_xGLXRenderLargeReq;
+ unsigned vertex_requests;
+
+
+ /* Calculate the number of data packets that will be required to send
+ * the whole command. To do this, the number of verticies that
+ * will fit in a single buffer must be calculated.
+ *
+ * The important value here is elements_per_request. This is the
+ * number of complete array elements that will fit in a single
+ * buffer. There may be some wasted space at the end of the buffer,
+ * but splitting elements across buffer boundries would be painful.
+ */
+
+ elements_per_request[0] = maxSize / single_vertex_size;
+
+ vertex_requests = (count + elements_per_request[0] - 1)
+ / elements_per_request[0];
+
+ *total_requests = vertex_requests + 1;
+
+
+ __glXFlushRenderBuffer(gc, gc->pc);
+
+ command_size += 4;
+
+ pc = arrays->large_header;
+ *(uint32_t *)(pc + 0) = command_size;
+ *(uint32_t *)(pc + 4) = X_GLrop_DrawArrays;
+ *(uint32_t *)(pc + 8) = count;
+ *(uint32_t *)(pc + 12) = arrays->enabled_array_count;
+ *(uint32_t *)(pc + 16) = mode;
+
+ __glXSendLargeChunk( gc, 1, *total_requests, pc,
+ header_size + 4 + arrays->array_info_cache_size );
+
+ pc = gc->pc;
+ }
+ else {
+ if ( (gc->pc + command_size) >= gc->bufEnd ) {
+ (void) __glXFlushRenderBuffer(gc, gc->pc);
+ }
+
+ pc = gc->pc;
+ *(uint16_t *)(pc + 0) = command_size;
+ *(uint16_t *)(pc + 2) = X_GLrop_DrawArrays;
+ *(uint32_t *)(pc + 4) = count;
+ *(uint32_t *)(pc + 8) = arrays->enabled_array_count;
+ *(uint32_t *)(pc + 12) = mode;
+
+ pc += header_size;
+
+ (void) memcpy( pc, arrays->array_info_cache,
+ arrays->array_info_cache_size );
+ pc += arrays->array_info_cache_size;
+
+ *elements_per_request = count;
+ *total_requests = 0;
+ }
+
+
+ return pc;
+}
+
+
+/**
+ */
+void
+emit_DrawArrays_old( GLenum mode, GLint first, GLsizei count )
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+
+ GLubyte * pc;
+ size_t elements_per_request;
+ unsigned total_requests = 0;
+ unsigned i;
+ size_t total_sent = 0;
+
+
+ pc = emit_DrawArrays_header_old( gc, arrays, & elements_per_request,
+ & total_requests, mode, count);
+
+
+ /* Write the arrays.
+ */
+
+ if ( total_requests == 0 ) {
+ assert( elements_per_request >= count );
+
+ for ( i = 0 ; i < count ; i++ ) {
+ pc = emit_element_old( pc, arrays, i + first );
+ }
+
+ assert( pc <= gc->bufEnd );
+
+ gc->pc = pc;
+ if ( gc->pc > gc->limit ) {
+ (void) __glXFlushRenderBuffer(gc, gc->pc);
+ }
+ }
+ else {
+ unsigned req;
+
+
+ for ( req = 2 ; req <= total_requests ; req++ ) {
+ if ( count < elements_per_request ) {
+ elements_per_request = count;
+ }
+
+ pc = gc->pc;
+ for ( i = 0 ; i < elements_per_request ; i++ ) {
+ pc = emit_element_old( pc, arrays, i + first );
+ }
+
+ first += elements_per_request;
+
+ total_sent += (size_t) (pc - gc->pc);
+ __glXSendLargeChunk( gc, req, total_requests, gc->pc,
+ pc - gc->pc );
+
+ count -= elements_per_request;
+ }
+ }
+}
+
+
+void
+emit_DrawElements_none( GLenum mode, GLsizei count, GLenum type,
+ const GLvoid *indices )
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ static const uint16_t begin_cmd[2] = { 8, X_GLrop_Begin };
+ static const uint16_t end_cmd[2] = { 4, X_GLrop_End };
+
+ GLubyte * pc;
+ size_t single_vertex_size;
+ unsigned i;
+
+
+ single_vertex_size = calculate_single_vertex_size_none( arrays );
+
+
+ if ( (gc->pc + single_vertex_size) >= gc->bufEnd ) {
+ gc->pc = __glXFlushRenderBuffer(gc, gc->pc);
+ }
+
+ pc = gc->pc;
+
+ (void) memcpy( pc, begin_cmd, 4 );
+ *(int *)(pc + 4) = mode;
+
+ pc += 8;
+
+ for ( i = 0 ; i < count ; i++ ) {
+ unsigned index;
+
+ if ( (pc + single_vertex_size) >= gc->bufEnd ) {
+ pc = __glXFlushRenderBuffer(gc, gc->pc);
+ }
+
+ switch( type ) {
+ case GL_UNSIGNED_INT:
+ index = (unsigned) (((GLuint *) indices)[i]);
+ break;
+ case GL_UNSIGNED_SHORT:
+ index = (unsigned) (((GLushort *) indices)[i]);
+ break;
+ case GL_UNSIGNED_BYTE:
+ index = (unsigned) (((GLubyte *) indices)[i]);
+ break;
+ }
+ pc = emit_element_none( pc, arrays, index );
+ }
+
+ if ( (pc + 4) >= gc->bufEnd ) {
+ pc = __glXFlushRenderBuffer(gc, gc->pc);
+ }
+
+ (void) memcpy( pc, end_cmd, 4 );
+ pc += 4;
+
+ gc->pc = pc;
+ if ( gc->pc > gc->limit ) {
+ (void) __glXFlushRenderBuffer(gc, gc->pc);
+ }
+}
+
+
+/**
+ */
+void
+emit_DrawElements_old( GLenum mode, GLsizei count, GLenum type,
+ const GLvoid *indices )
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+
+ GLubyte * pc;
+ size_t elements_per_request;
+ unsigned total_requests = 0;
+ unsigned i;
+ unsigned req;
+ const GLuint * ui_ptr = (const GLuint *) indices;
+ const GLushort * us_ptr = (const GLushort *) indices;
+ const GLubyte * ub_ptr = (const GLubyte *) indices;
+
+
+ pc = emit_DrawArrays_header_old( gc, arrays, & elements_per_request,
+ & total_requests, mode, count);
+
+
+ /* Write the arrays.
+ */
+
+ req = 2;
+ while ( count > 0 ) {
+ if ( count < elements_per_request ) {
+ elements_per_request = count;
+ }
+
+ switch( type ) {
+ case GL_UNSIGNED_INT:
+ for ( i = 0 ; i < elements_per_request ; i++ ) {
+ const GLint index = (GLint) *(ui_ptr++);
+ pc = emit_element_old( pc, arrays, index );
+ }
+ break;
+ case GL_UNSIGNED_SHORT:
+ for ( i = 0 ; i < elements_per_request ; i++ ) {
+ const GLint index = (GLint) *(us_ptr++);
+ pc = emit_element_old( pc, arrays, index );
+ }
+ break;
+ case GL_UNSIGNED_BYTE:
+ for ( i = 0 ; i < elements_per_request ; i++ ) {
+ const GLint index = (GLint) *(ub_ptr++);
+ pc = emit_element_old( pc, arrays, index );
+ }
+ break;
+ }
+
+ if ( total_requests != 0 ) {
+ __glXSendLargeChunk( gc, req, total_requests, gc->pc,
+ pc - gc->pc );
+ pc = gc->pc;
+ req++;
+ }
+
+ count -= elements_per_request;
+ }
+
+
+ assert( (total_requests == 0) || ((req - 1) == total_requests) );
+
+ if ( total_requests == 0 ) {
+ assert( pc <= gc->bufEnd );
+
+ gc->pc = pc;
+ if ( gc->pc > gc->limit ) {
+ (void) __glXFlushRenderBuffer(gc, gc->pc);
+ }
+ }
+}
+
+
+/**
+ * Validate that the \c mode parameter to \c glDrawArrays, et. al. is valid.
+ * If it is not valid, then an error code is set in the GLX context.
+ *
+ * \returns
+ * \c GL_TRUE if the argument is valid, \c GL_FALSE if is not.
+ */
+static GLboolean
+validate_mode(__GLXcontext *gc, GLenum mode)
+{
+ switch(mode) {
+ case GL_POINTS:
+ case GL_LINE_STRIP:
+ case GL_LINE_LOOP:
+ case GL_LINES:
+ case GL_TRIANGLE_STRIP:
+ case GL_TRIANGLE_FAN:
+ case GL_TRIANGLES:
+ case GL_QUAD_STRIP:
+ case GL_QUADS:
+ case GL_POLYGON:
+ break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return GL_FALSE;
+ }
+
+ return GL_TRUE;
+}
+
+
+/**
+ * Validate that the \c count parameter to \c glDrawArrays, et. al. is valid.
+ * A value less than zero is invalid and will result in \c GL_INVALID_VALUE
+ * being set. A value of zero will not result in an error being set, but
+ * will result in \c GL_FALSE being returned.
+ *
+ * \returns
+ * \c GL_TRUE if the argument is valid, \c GL_FALSE if it is not.
+ */
+static GLboolean
+validate_count(__GLXcontext *gc, GLsizei count)
+{
+ if (count < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ }
+
+ return (count > 0);
+}
+
+
+void __indirect_glDrawArrays(GLenum mode, GLint first, GLsizei count)
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+
+
+ if ( validate_mode(gc, mode) && validate_count(gc, count) ) {
+ if ( ! arrays->array_info_cache_valid ) {
+ fill_array_info_cache( arrays );
+ }
+
+ arrays->DrawArrays(mode, first, count);
+ }
+}
+
+
+void __indirect_glArrayElement(GLint index)
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+
+ size_t single_vertex_size;
+
+
+ single_vertex_size = calculate_single_vertex_size_none( arrays );
+
+ if ( (gc->pc + single_vertex_size) >= gc->bufEnd ) {
+ gc->pc = __glXFlushRenderBuffer(gc, gc->pc);
+ }
+
+ gc->pc = emit_element_none( gc->pc, arrays, index );
+
+ if ( gc->pc > gc->limit ) {
+ (void) __glXFlushRenderBuffer(gc, gc->pc);
+ }
+}
+
+
+void __indirect_glDrawElements(GLenum mode, GLsizei count, GLenum type,
+ const GLvoid *indices)
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+
+
+ if ( validate_mode(gc, mode) && validate_count(gc, count) ) {
+ if ( ! arrays->array_info_cache_valid ) {
+ fill_array_info_cache( arrays );
+ }
+
+ arrays->DrawElements(mode, count, type, indices);
+ }
+}
+
+
+void __indirect_glDrawRangeElements(GLenum mode, GLuint start, GLuint end,
+ GLsizei count, GLenum type,
+ const GLvoid *indices)
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+
+
+ if ( validate_mode(gc, mode) && validate_count(gc, count) ) {
+ if (end < start) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+ if ( ! arrays->array_info_cache_valid ) {
+ fill_array_info_cache( arrays );
+ }
+
+ arrays->DrawElements(mode, count, type, indices);
+ }
+}
+
+
+void __indirect_glMultiDrawArraysEXT(GLenum mode, GLint *first, GLsizei *count,
+ GLsizei primcount)
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ GLsizei i;
+
+
+ if ( validate_mode(gc, mode) ) {
+ if ( ! arrays->array_info_cache_valid ) {
+ fill_array_info_cache( arrays );
+ }
+
+ for ( i = 0 ; i < primcount ; i++ ) {
+ if ( validate_count( gc, count[i] ) ) {
+ arrays->DrawArrays(mode, first[i], count[i]);
+ }
+ }
+ }
+}
+
+
+void __indirect_glMultiDrawElementsEXT(GLenum mode, const GLsizei *count,
+ GLenum type, const GLvoid ** indices,
+ GLsizei primcount)
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ const __GLXattribute * state =
+ (const __GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ GLsizei i;
+
+
+ if ( validate_mode(gc, mode) ) {
+ if ( ! arrays->array_info_cache_valid ) {
+ fill_array_info_cache( arrays );
+ }
+
+ for ( i = 0 ; i < primcount ; i++ ) {
+ if ( validate_count( gc, count[i] ) ) {
+ arrays->DrawElements(mode, count[i], type, indices[i]);
+ }
+ }
+ }
+}
+
+
+#define COMMON_ARRAY_DATA_INIT(a, PTR, TYPE, STRIDE, COUNT, HDR_SIZE, OPCODE) \
+ do { \
+ (a)->data = PTR; \
+ (a)->data_type = TYPE; \
+ (a)->user_stride = STRIDE; \
+ (a)->count = COUNT; \
+ \
+ (a)->element_size = __glXTypeSize( TYPE ) * COUNT; \
+ (a)->true_stride = (STRIDE == 0) \
+ ? (a)->element_size : STRIDE; \
+ \
+ (a)->header_size = HDR_SIZE; \
+ ((uint16_t *) (a)->header)[0] = __GLX_PAD((a)->header_size + (a)->element_size); \
+ ((uint16_t *) (a)->header)[1] = OPCODE; \
+ } while(0)
+
+
+void __indirect_glVertexPointer( GLint size, GLenum type, GLsizei stride,
+ const GLvoid * pointer )
+{
+ static const uint16_t short_ops[5] = {
+ 0, 0, X_GLrop_Vertex2sv, X_GLrop_Vertex3sv, X_GLrop_Vertex4sv
+ };
+ static const uint16_t int_ops[5] = {
+ 0, 0, X_GLrop_Vertex2iv, X_GLrop_Vertex3iv, X_GLrop_Vertex4iv
+ };
+ static const uint16_t float_ops[5] = {
+ 0, 0, X_GLrop_Vertex2fv, X_GLrop_Vertex3fv, X_GLrop_Vertex4fv
+ };
+ static const uint16_t double_ops[5] = {
+ 0, 0, X_GLrop_Vertex2dv, X_GLrop_Vertex3dv, X_GLrop_Vertex4dv
+ };
+ uint16_t opcode;
+ __GLXcontext *gc = __glXGetCurrentContext();
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+
+
+ if (size < 2 || size > 4 || stride < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+ switch ( type ) {
+ case GL_SHORT: opcode = short_ops[size]; break;
+ case GL_INT: opcode = int_ops[size]; break;
+ case GL_FLOAT: opcode = float_ops[size]; break;
+ case GL_DOUBLE: opcode = double_ops[size]; break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ a = get_array_entry( arrays, GL_VERTEX_ARRAY, 0 );
+ assert( a != NULL );
+ COMMON_ARRAY_DATA_INIT( a, pointer, type, stride, size, 4, opcode );
+
+ if ( a->enabled ) {
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+}
+
+
+void __indirect_glNormalPointer( GLenum type, GLsizei stride,
+ const GLvoid * pointer )
+{
+ uint16_t opcode;
+ __GLXcontext *gc = __glXGetCurrentContext();
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+
+
+ if (stride < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+ switch ( type ) {
+ case GL_BYTE: opcode = X_GLrop_Normal3bv; break;
+ case GL_SHORT: opcode = X_GLrop_Normal3sv; break;
+ case GL_INT: opcode = X_GLrop_Normal3iv; break;
+ case GL_FLOAT: opcode = X_GLrop_Normal3fv; break;
+ case GL_DOUBLE: opcode = X_GLrop_Normal3dv; break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ a = get_array_entry( arrays, GL_NORMAL_ARRAY, 0 );
+ assert( a != NULL );
+ COMMON_ARRAY_DATA_INIT( a, pointer, type, stride, 3, 4, opcode );
+
+ if ( a->enabled ) {
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+}
+
+
+void __indirect_glColorPointer( GLint size, GLenum type, GLsizei stride,
+ const GLvoid * pointer )
+{
+ static const uint16_t byte_ops[5] = {
+ 0, 0, 0, X_GLrop_Color3bv, X_GLrop_Color4bv
+ };
+ static const uint16_t ubyte_ops[5] = {
+ 0, 0, 0, X_GLrop_Color3ubv, X_GLrop_Color4ubv
+ };
+ static const uint16_t short_ops[5] = {
+ 0, 0, 0, X_GLrop_Color3sv, X_GLrop_Color4sv
+ };
+ static const uint16_t ushort_ops[5] = {
+ 0, 0, 0, X_GLrop_Color3usv, X_GLrop_Color4usv
+ };
+ static const uint16_t int_ops[5] = {
+ 0, 0, 0, X_GLrop_Color3iv, X_GLrop_Color4iv
+ };
+ static const uint16_t uint_ops[5] = {
+ 0, 0, 0, X_GLrop_Color3uiv, X_GLrop_Color4uiv
+ };
+ static const uint16_t float_ops[5] = {
+ 0, 0, 0, X_GLrop_Color3fv, X_GLrop_Color4fv
+ };
+ static const uint16_t double_ops[5] = {
+ 0, 0, 0, X_GLrop_Color3dv, X_GLrop_Color4dv
+ };
+ uint16_t opcode;
+ __GLXcontext *gc = __glXGetCurrentContext();
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+
+
+ if (size < 3 || size > 4 || stride < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+ switch ( type ) {
+ case GL_BYTE: opcode = byte_ops[size]; break;
+ case GL_UNSIGNED_BYTE: opcode = ubyte_ops[size]; break;
+ case GL_SHORT: opcode = short_ops[size]; break;
+ case GL_UNSIGNED_SHORT: opcode = ushort_ops[size]; break;
+ case GL_INT: opcode = int_ops[size]; break;
+ case GL_UNSIGNED_INT: opcode = uint_ops[size]; break;
+ case GL_FLOAT: opcode = float_ops[size]; break;
+ case GL_DOUBLE: opcode = double_ops[size]; break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ a = get_array_entry( arrays, GL_COLOR_ARRAY, 0 );
+ assert( a != NULL );
+ COMMON_ARRAY_DATA_INIT( a, pointer, type, stride, size, 4, opcode );
+
+ if ( a->enabled ) {
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+}
+
+
+void __indirect_glIndexPointer( GLenum type, GLsizei stride,
+ const GLvoid * pointer )
+{
+ uint16_t opcode;
+ __GLXcontext *gc = __glXGetCurrentContext();
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+
+
+ if (stride < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+ switch ( type ) {
+ case GL_UNSIGNED_BYTE: opcode = X_GLrop_Indexubv; break;
+ case GL_SHORT: opcode = X_GLrop_Indexsv; break;
+ case GL_INT: opcode = X_GLrop_Indexiv; break;
+ case GL_FLOAT: opcode = X_GLrop_Indexfv; break;
+ case GL_DOUBLE: opcode = X_GLrop_Indexdv; break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ a = get_array_entry( arrays, GL_INDEX_ARRAY, 0 );
+ assert( a != NULL );
+ COMMON_ARRAY_DATA_INIT( a, pointer, type, stride, 1, 4, opcode );
+
+ if ( a->enabled ) {
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+}
+
+
+void __indirect_glEdgeFlagPointer( GLsizei stride, const GLvoid * pointer )
+{
+ __GLXcontext *gc = __glXGetCurrentContext();
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+
+
+ if (stride < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+
+ a = get_array_entry( arrays, GL_EDGE_FLAG_ARRAY, 0 );
+ assert( a != NULL );
+ COMMON_ARRAY_DATA_INIT( a, pointer, GL_UNSIGNED_BYTE, stride, 1, 4, X_GLrop_EdgeFlagv );
+
+ if ( a->enabled ) {
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+}
+
+
+void __indirect_glTexCoordPointer( GLint size, GLenum type, GLsizei stride,
+ const GLvoid * pointer )
+{
+ static const uint16_t short_ops[5] = {
+ 0, X_GLrop_TexCoord1sv, X_GLrop_TexCoord2sv, X_GLrop_TexCoord3sv, X_GLrop_TexCoord4sv
+ };
+ static const uint16_t int_ops[5] = {
+ 0, X_GLrop_TexCoord1iv, X_GLrop_TexCoord2iv, X_GLrop_TexCoord3iv, X_GLrop_TexCoord4iv
+ };
+ static const uint16_t float_ops[5] = {
+ 0, X_GLrop_TexCoord1dv, X_GLrop_TexCoord2fv, X_GLrop_TexCoord3fv, X_GLrop_TexCoord4fv
+ };
+ static const uint16_t double_ops[5] = {
+ 0, X_GLrop_TexCoord1dv, X_GLrop_TexCoord2dv, X_GLrop_TexCoord3dv, X_GLrop_TexCoord4dv
+ };
+
+ static const uint16_t mshort_ops[5] = {
+ 0, X_GLrop_MultiTexCoord1svARB, X_GLrop_MultiTexCoord2svARB, X_GLrop_MultiTexCoord3svARB, X_GLrop_MultiTexCoord4svARB
+ };
+ static const uint16_t mint_ops[5] = {
+ 0, X_GLrop_MultiTexCoord1ivARB, X_GLrop_MultiTexCoord2ivARB, X_GLrop_MultiTexCoord3ivARB, X_GLrop_MultiTexCoord4ivARB
+ };
+ static const uint16_t mfloat_ops[5] = {
+ 0, X_GLrop_MultiTexCoord1dvARB, X_GLrop_MultiTexCoord2fvARB, X_GLrop_MultiTexCoord3fvARB, X_GLrop_MultiTexCoord4fvARB
+ };
+ static const uint16_t mdouble_ops[5] = {
+ 0, X_GLrop_MultiTexCoord1dvARB, X_GLrop_MultiTexCoord2dvARB, X_GLrop_MultiTexCoord3dvARB, X_GLrop_MultiTexCoord4dvARB
+ };
+
+ uint16_t opcode;
+ __GLXcontext *gc = __glXGetCurrentContext();
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+ unsigned header_size;
+ unsigned index;
+
+
+ if (size < 1 || size > 4 || stride < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+ index = arrays->active_texture_unit;
+ if ( index == 0 ) {
+ switch ( type ) {
+ case GL_SHORT: opcode = short_ops[size]; break;
+ case GL_INT: opcode = int_ops[size]; break;
+ case GL_FLOAT: opcode = float_ops[size]; break;
+ case GL_DOUBLE: opcode = double_ops[size]; break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ header_size = 4;
+ }
+ else {
+ switch ( type ) {
+ case GL_SHORT: opcode = mshort_ops[size]; break;
+ case GL_INT: opcode = mint_ops[size]; break;
+ case GL_FLOAT: opcode = mfloat_ops[size]; break;
+ case GL_DOUBLE: opcode = mdouble_ops[size]; break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ header_size = 8;
+ }
+
+ a = get_array_entry( arrays, GL_TEXTURE_COORD_ARRAY, index );
+ assert( a != NULL );
+ COMMON_ARRAY_DATA_INIT( a, pointer, type, stride, size, header_size, opcode );
+
+ if ( a->enabled ) {
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+}
+
+
+void __indirect_glSecondaryColorPointerEXT( GLint size, GLenum type, GLsizei stride,
+ const GLvoid * pointer )
+{
+ uint16_t opcode;
+ __GLXcontext *gc = __glXGetCurrentContext();
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+
+
+ if (size != 3 || stride < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+ switch ( type ) {
+ case GL_BYTE: opcode = 4126; break;
+ case GL_UNSIGNED_BYTE: opcode = 4131; break;
+ case GL_SHORT: opcode = 4127; break;
+ case GL_UNSIGNED_SHORT: opcode = 4132; break;
+ case GL_INT: opcode = 4128; break;
+ case GL_UNSIGNED_INT: opcode = 4133; break;
+ case GL_FLOAT: opcode = 4129; break;
+ case GL_DOUBLE: opcode = 4130; break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ a = get_array_entry( arrays, GL_SECONDARY_COLOR_ARRAY, 0 );
+ if ( a == NULL ) {
+ __glXSetError(gc, GL_INVALID_OPERATION);
+ return;
+ }
+
+ COMMON_ARRAY_DATA_INIT( a, pointer, type, stride, size, 4, opcode );
+
+ if ( a->enabled ) {
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+}
+
+
+void __indirect_glFogCoordPointerEXT( GLenum type, GLsizei stride,
+ const GLvoid * pointer )
+{
+ uint16_t opcode;
+ __GLXcontext *gc = __glXGetCurrentContext();
+ __GLXattribute * state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+
+
+ if (stride < 0) {
+ __glXSetError(gc, GL_INVALID_VALUE);
+ return;
+ }
+
+ switch ( type ) {
+ case GL_FLOAT: opcode = 4124; break;
+ case GL_DOUBLE: opcode = 4125; break;
+ default:
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ a = get_array_entry( arrays, GL_FOG_COORD_ARRAY, 0 );
+ if ( a == NULL ) {
+ __glXSetError(gc, GL_INVALID_OPERATION);
+ return;
+ }
+
+ COMMON_ARRAY_DATA_INIT( a, pointer, type, stride, 1, 4, opcode );
+
+ if ( a->enabled ) {
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+}
+
+
+void __indirect_glClientActiveTextureARB(GLenum texture)
+{
+ __GLXcontext * const gc = __glXGetCurrentContext();
+ __GLXattribute * const state = (__GLXattribute *)(gc->client_state_private);
+ struct array_state_vector * const arrays = state->array_state;
+ const GLint unit = (GLint) texture - GL_TEXTURE0;
+
+
+ if ( (unit < 0) || (unit > arrays->num_texture_units) ) {
+ __glXSetError(gc, GL_INVALID_ENUM);
+ return;
+ }
+
+ arrays->active_texture_unit = unit;
+}
+
+
+/**
+ */
+GLboolean
+__glXSetArrayEnable( __GLXattribute * state,
+ GLenum key, unsigned index, GLboolean enable )
+{
+ struct array_state_vector * arrays = state->array_state;
+ struct array_state * a;
+
+
+ if ( key == GL_TEXTURE_COORD_ARRAY ) {
+ index = arrays->active_texture_unit;
+ }
+
+ a = get_array_entry( arrays, key, index );
+
+ if ( (a != NULL) && (a->enabled != enable) ) {
+ a->enabled = enable;
+ arrays->array_info_cache_valid = GL_FALSE;
+ }
+
+ return (a != NULL);
+}
+
+
+void
+__glXArrayDisableAll( __GLXattribute * state )
+{
+ struct array_state_vector * arrays = state->array_state;
+ unsigned i;
+
+
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ arrays->arrays[i].enabled = GL_FALSE;
+ }
+
+ arrays->array_info_cache_valid = GL_FALSE;
+}
+
+
+/**
+ */
+GLboolean
+__glXGetArrayEnable( const __GLXattribute * const state,
+ GLenum key, unsigned index, GLintptr * dest )
+{
+ const struct array_state_vector * arrays = state->array_state;
+ const struct array_state * a = get_array_entry( (struct array_state_vector *) arrays,
+ key, index );
+
+ if ( a != NULL ) {
+ *dest = (GLintptr) a->enabled;
+ }
+
+ return (a != NULL);
+}
+
+
+/**
+ */
+GLboolean
+__glXGetArrayType( const __GLXattribute * const state,
+ GLenum key, unsigned index, GLintptr * dest )
+{
+ const struct array_state_vector * arrays = state->array_state;
+ const struct array_state * a = get_array_entry( (struct array_state_vector *) arrays,
+ key, index );
+
+ if ( a != NULL ) {
+ *dest = (GLintptr) a->enabled;
+ }
+
+ return (a != NULL);
+}
+
+
+/**
+ */
+GLboolean
+__glXGetArraySize( const __GLXattribute * const state,
+ GLenum key, unsigned index, GLintptr * dest )
+{
+ const struct array_state_vector * arrays = state->array_state;
+ const struct array_state * a = get_array_entry( (struct array_state_vector *) arrays,
+ key, index );
+
+ if ( a != NULL ) {
+ *dest = (GLintptr) a->count;
+ }
+
+ return (a != NULL);
+}
+
+
+/**
+ */
+GLboolean
+__glXGetArrayStride( const __GLXattribute * const state,
+ GLenum key, unsigned index, GLintptr * dest )
+{
+ const struct array_state_vector * arrays = state->array_state;
+ const struct array_state * a = get_array_entry( (struct array_state_vector *) arrays,
+ key, index );
+
+ if ( a != NULL ) {
+ *dest = (GLintptr) a->user_stride;
+ }
+
+ return (a != NULL);
+}
+
+
+/**
+ */
+GLboolean
+__glXGetArrayPointer( const __GLXattribute * const state,
+ GLenum key, unsigned index, void ** dest )
+{
+ const struct array_state_vector * arrays = state->array_state;
+ const struct array_state * a = get_array_entry( (struct array_state_vector *) arrays,
+ key, index );
+
+
+ if ( a != NULL ) {
+ *dest = a->data;
+ }
+
+ return (a != NULL);
+}
+
+
+/**
+ */
+GLuint
+__glXGetActiveTextureUnit( const __GLXattribute * const state )
+{
+ return state->array_state->active_texture_unit;
+}
+
+
+void
+__glXPushArrayState( __GLXattribute * state )
+{
+ struct array_state_vector * arrays = state->array_state;
+ struct array_stack_state * stack = & arrays->stack[ (arrays->stack_index * arrays->num_arrays)];
+ unsigned i;
+
+
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ stack[i].data = arrays->arrays[i].data;
+ stack[i].data_type = arrays->arrays[i].data_type;
+ stack[i].user_stride = arrays->arrays[i].user_stride;
+ stack[i].count = arrays->arrays[i].count;
+ stack[i].key = arrays->arrays[i].key;
+ stack[i].enabled = arrays->arrays[i].enabled;
+ }
+
+ arrays->active_texture_unit_stack[ arrays->stack_index ] =
+ arrays->active_texture_unit;
+
+ arrays->stack_index++;
+}
+
+
+void
+__glXPopArrayState( __GLXattribute * state )
+{
+ struct array_state_vector * arrays = state->array_state;
+ struct array_stack_state * stack;
+ unsigned i;
+
+
+ arrays->stack_index--;
+ stack = & arrays->stack[ (arrays->stack_index * arrays->num_arrays) ];
+
+ for ( i = 0 ; i < arrays->num_arrays ; i++ ) {
+ switch ( stack[i].key ) {
+ case GL_NORMAL_ARRAY:
+ __indirect_glNormalPointer( stack[i].data_type,
+ stack[i].user_stride,
+ stack[i].data );
+ break;
+ case GL_COLOR_ARRAY:
+ __indirect_glColorPointer( stack[i].count,
+ stack[i].data_type,
+ stack[i].user_stride,
+ stack[i].data );
+ break;
+ case GL_INDEX_ARRAY:
+ __indirect_glIndexPointer( stack[i].data_type,
+ stack[i].user_stride,
+ stack[i].data );
+ break;
+ case GL_EDGE_FLAG_ARRAY:
+ __indirect_glEdgeFlagPointer( stack[i].user_stride,
+ stack[i].data );
+ break;
+ case GL_TEXTURE_COORD_ARRAY:
+ arrays->active_texture_unit = stack[i].index;
+ __indirect_glTexCoordPointer( stack[i].count,
+ stack[i].data_type,
+ stack[i].user_stride,
+ stack[i].data );
+ break;
+ case GL_SECONDARY_COLOR_ARRAY:
+ __indirect_glSecondaryColorPointerEXT( stack[i].count,
+ stack[i].data_type,
+ stack[i].user_stride,
+ stack[i].data );
+ break;
+ case GL_FOG_COORDINATE_ARRAY:
+ __indirect_glFogCoordPointerEXT( stack[i].data_type,
+ stack[i].user_stride,
+ stack[i].data );
+ break;
+
+ }
+
+ __glXSetArrayEnable( state, stack[i].key, stack[i].index,
+ stack[i].enabled );
+ }
+
+ arrays->active_texture_unit = arrays->active_texture_unit_stack[ arrays->stack_index ];
+}