Clean up submission code

GL/draw.c

@@ -1170,6 +1170,8 @@ void _glInitSubmissionTarget() {
 
 
 GL_FORCE_INLINE void submitVertices(GLenum mode, GLsizei first, GLuint count, GLenum type, const GLvoid* indices) {
+    gl_assert(mode != GL_POLYGON);
+
     SubmissionTarget* const target = &SUBMISSION_TARGET;
     AlignedVector* const extras = target->extras;
 
@@ -1185,37 +1187,8 @@ GL_FORCE_INLINE void submitVertices(GLenum mode, GLsizei first, GLuint count, GL
         return;
     }
 
-    /* Polygons are treated as triangle fans, the only time this would be a
-     * problem is if we supported glPolygonMode(..., GL_LINE) but we don't.
-     * We optimise the triangle and quad cases.
-     */
-    if(mode == GL_POLYGON) {
-        switch(count) {
-            case 2:
-                mode = GL_LINES;
-            break;
-            case 3:
-                mode = GL_TRIANGLES;
-            break;
-            case 4:
-                mode = GL_QUADS;
-            break;
-            default:
-                mode = GL_TRIANGLE_FAN;
-        }
-    }
-
-    if(mode == GL_LINE_STRIP || mode == GL_LINES) {
-        fprintf(stderr, "Line drawing is currently unsupported\n");
-        return;
-    }
-
     GLboolean header_required = (target->output->vector.size == 0) || _glGPUStateIsDirty();
 
-
-    // We don't handle this any further, so just make sure we never pass it down */
-    gl_assert(mode != GL_POLYGON);
-
     target->output = _glActivePolyList();
     target->count = (mode == GL_TRIANGLE_FAN) ? ((count - 2) * 3) : count;
     target->header_offset = target->output->vector.size;
@@ -1308,6 +1281,35 @@ GL_FORCE_INLINE void submitVertices(GLenum mode, GLsizei first, GLuint count, GL
     // }
 }
 
+static GLenum convertModeIfNecessary(GLenum mode, GLsizei count) {
+    /* Polygons are treated as triangle fans, the only time this would be a
+     * problem is if we supported glPolygonMode(..., GL_LINE) but we don't.
+     * We optimise the triangle and quad cases.
+     */
+    if(mode == GL_POLYGON) {
+        switch(count) {
+            case 2:
+                mode = GL_LINES;
+            break;
+            case 3:
+                mode = GL_TRIANGLES;
+            break;
+            case 4:
+                mode = GL_QUADS;
+            break;
+            default:
+                mode = GL_TRIANGLE_FAN;
+        }
+    }
+
+    if(mode == GL_LINE_STRIP || mode == GL_LINES) {
+        _glKosThrowError(GL_INVALID_VALUE, __func__);
+        return GL_NONE;
+    }
+
+    return mode;
+}
+
 void APIENTRY glDrawElements(GLenum mode, GLsizei count, GLenum type, const GLvoid* indices) {
     TRACE();
 
@@ -1315,6 +1317,11 @@ void APIENTRY glDrawElements(GLenum mode, GLsizei count, GLenum type, const GLvo
         return;
     }
 
+    mode = convertModeIfNecessary(mode, count);
+    if(mode == GL_NONE) {
+        return;
+    }
+
     submitVertices(mode, 0, count, type, indices);
 }
 
@@ -1325,9 +1332,169 @@ void APIENTRY glDrawArrays(GLenum mode, GLint first, GLsizei count) {
         return;
     }
 
+    mode = convertModeIfNecessary(mode, count);
+    if(mode == GL_NONE) {
+        return;
+    }
+
     submitVertices(mode, first, count, GL_UNSIGNED_INT, NULL);
 }
 
+static void glDrawPVRArrays(GLenum mode, GLsizei stride, GLint first, GLsizei count, void* data) {
+    GLint validTypes[] = {
+        GL_TRIANGLES,
+        GL_QUADS,
+        GL_TRIANGLE_STRIP,
+        0
+    };
+
+    if(_glCheckValidEnum(mode, validTypes, __func__) != 0) {
+        return;
+    }
+
+    SubmissionTarget* const target = &SUBMISSION_TARGET;
+    AlignedVector* const extras = target->extras;
+
+    GLboolean header_required = (target->output->vector.size == 0) || _glGPUStateIsDirty();
+
+    target->output = _glActivePolyList();
+    target->count = (mode == GL_TRIANGLE_FAN) ? ((count - 2) * 3) : count;
+    target->header_offset = target->output->vector.size;
+    target->start_offset = target->header_offset + (header_required);
+
+    gl_assert(target->count);
+
+    /* Make sure we have enough room for all the "extra" data */
+    aligned_vector_resize(extras, target->count);
+
+    /* Make room for the vertices and header */
+    aligned_vector_extend(&target->output->vector, target->count + (header_required));
+
+    if(header_required) {
+        apply_poly_header(_glSubmissionTargetHeader(target), GL_FALSE, target->output, 0);
+        _glGPUStateMarkClean();
+    }
+
+    PREFETCH(data + (first * 32));
+
+    if(_glIsLightingEnabled()) {
+        _glMatrixLoadModelView();
+    } else {
+        _glMatrixLoadModelViewProjection();
+    }
+
+#define DO_TRANSFORM() \
+    do { \
+        Vertex* v = _glSubmissionTargetStart(target); \
+        VertexExtra* ve = aligned_vector_at(target->extras, 0); \
+        float w = 1.0f; \
+        if(mode == GL_QUADS) { \
+            const static uint32_t flags [] = { \
+                GPU_CMD_VERTEX, GPU_CMD_VERTEX, GPU_CMD_VERTEX_EOL, GPU_CMD_VERTEX \
+            }; \
+            for(int_fast32_t i = 0; i < count; ++i, ++v) { \
+                v->flags = flags[(i % 4)]; \
+                TransformVertex(v->xyz, &w, v->xyz, &v->w); \
+            } \
+            v = _glSubmissionTargetStart(target); \
+            Vertex* prev = (v + 2); \
+            v += 3; \
+            for(int_fast32_t i = 0; i < count; i +=4, v +=4, prev += 4) { \
+                const Vertex t = (*prev); \
+                *(prev) = *((v)); \
+                *((v)) = t; \
+            } \
+        } else if(mode == GL_TRIANGLES) { \
+            const static uint32_t flags [] = { \
+                GPU_CMD_VERTEX, GPU_CMD_VERTEX, GPU_CMD_VERTEX_EOL \
+            }; \
+            for(int_fast32_t i = 0; i < count; ++i, ++v) { \
+                PREFETCH(v + 3); \
+                TransformVertex(v->xyz, &w, v->xyz, &v->w); \
+                v->flags = flags[(i % 3)]; \
+            } \
+        } else { \
+            for(int_fast32_t i = 0; i < count; ++i, ++v) { \
+                PREFETCH(v + 6); \
+                TransformVertex(v->xyz, &w, v->xyz, &v->w); \
+                v->flags = GPU_CMD_VERTEX; \
+            } \
+            (v - 1)->flags = GPU_CMD_VERTEX_EOL; \
+        } \
+    } while(0)
+
+
+    if(stride == 32) {
+        uint8_t* src = (uint8_t*) data;
+        /* Copy the data directly */
+        sq_cpy(_glSubmissionTargetStart(target), src + (first * 32), count * 32);
+
+        DO_TRANSFORM();
+    } else {
+        assert(stride == 64);
+
+        struct {
+            uint32_t data[8];
+        }* src = data + (first * 32);
+
+        Vertex* dst = aligned_vector_at(&target->output->vector, target->start_offset);
+        VertexExtra* dst2 = aligned_vector_at(target->extras, 0);
+
+        for(int i = 0; i < count - 6; ++i) {
+            PREFETCH(src + 2);
+            *dst = *((Vertex*) src++);
+            dst++;
+            *dst2 = *((VertexExtra*) src++);
+            dst2++;
+        }
+
+        PREFETCH(_glSubmissionTargetStart(target));
+        PREFETCH(aligned_vector_at(target->extras, 0));
+
+        for(int i = count - 6; i < count; ++i) {
+            *dst = *((Vertex*) src++);
+            dst++;
+            *dst2 = *((VertexExtra*) src++);
+            dst2++;
+        }
+
+        DO_TRANSFORM();
+    }
+}
+
+
+void APIENTRY glDrawPVRArrays32KOS(GLenum mode, GLint first, GLsizei count, void* data) {
+    TRACE();
+
+    if(_glCheckImmediateModeInactive(__func__)) {
+        return;
+    }
+
+    mode = convertModeIfNecessary(mode, count);
+    if(mode == GL_NONE) {
+        _glKosThrowError(GL_INVALID_VALUE, __func__);
+        return;
+    }
+
+    glDrawPVRArrays(mode, 32, first, count, data);
+}
+
+void APIENTRY glDrawPVRArrays64KOS(GLenum mode, GLint first, GLsizei count, void* data) {
+    TRACE();
+
+    if(_glCheckImmediateModeInactive(__func__)) {
+        return;
+    }
+
+    mode = convertModeIfNecessary(mode, count);
+    if(mode == GL_NONE) {
+        _glKosThrowError(GL_INVALID_VALUE, __func__);
+        return;
+    }
+
+    glDrawPVRArrays(mode, 64, first, count, data);
+}
+
 void APIENTRY glEnableClientState(GLenum cap) {
     TRACE();
 

GL/immediate.c

@@ -12,8 +12,6 @@
 
 #include "private.h"
 
-extern inline GLuint _glRecalcFastPath();
-
 GLboolean IMMEDIATE_MODE_ACTIVE = GL_FALSE;
 static GLenum ACTIVE_POLYGON_MODE = GL_TRIANGLES;
 
@@ -23,57 +21,28 @@ static GLfloat UV_COORD[2] = {0.0f, 0.0f};
 static GLfloat ST_COORD[2] = {0.0f, 0.0f};
 
 static AlignedVector VERTICES;
-static AttribPointerList IM_ATTRIBS;
-
-/* We store the list of attributes that have been "enabled" by a call to
-  glColor, glNormal, glTexCoord etc. otherwise we already have defaults that
-  can be applied faster */
-static GLuint IM_ENABLED_VERTEX_ATTRIBUTES = 0;
 
 typedef struct {
+    uint32_t padding;
     GLfloat x;
     GLfloat y;
     GLfloat z;
     GLfloat u;
     GLfloat v;
-    GLfloat s;
-    GLfloat t;
     GLubyte bgra[4];
+    GLubyte obgra[4];
     GLfloat nx;
     GLfloat ny;
     GLfloat nz;
-    GLuint padding[5];
+    GLfloat s;
+    GLfloat t;
+    GLuint padding2[3];
 } IMVertex;
 
 
 void _glInitImmediateMode(GLuint initial_size) {
     aligned_vector_init(&VERTICES, sizeof(IMVertex));
     aligned_vector_reserve(&VERTICES, initial_size);
-
-    IM_ATTRIBS.vertex.ptr = VERTICES.data;
-    IM_ATTRIBS.vertex.size = 3;
-    IM_ATTRIBS.vertex.type = GL_FLOAT;
-    IM_ATTRIBS.vertex.stride = sizeof(IMVertex);
-
-    IM_ATTRIBS.uv.ptr = IM_ATTRIBS.vertex.ptr + (sizeof(GLfloat) * 3);
-    IM_ATTRIBS.uv.stride = sizeof(IMVertex);
-    IM_ATTRIBS.uv.type = GL_FLOAT;
-    IM_ATTRIBS.uv.size = 2;
-
-    IM_ATTRIBS.st.ptr = IM_ATTRIBS.vertex.ptr + (sizeof(GLfloat) * 5);
-    IM_ATTRIBS.st.stride = sizeof(IMVertex);
-    IM_ATTRIBS.st.type = GL_FLOAT;
-    IM_ATTRIBS.st.size = 2;
-
-    IM_ATTRIBS.colour.ptr = IM_ATTRIBS.vertex.ptr + (sizeof(GLfloat) * 7);
-    IM_ATTRIBS.colour.size = GL_BGRA;  /* Flipped color order */
-    IM_ATTRIBS.colour.type = GL_UNSIGNED_BYTE;
-    IM_ATTRIBS.colour.stride = sizeof(IMVertex);
-
-    IM_ATTRIBS.normal.ptr = IM_ATTRIBS.vertex.ptr + (sizeof(GLfloat) * 7) + sizeof(uint32_t);
-    IM_ATTRIBS.normal.stride = sizeof(IMVertex);
-    IM_ATTRIBS.normal.type = GL_FLOAT;
-    IM_ATTRIBS.normal.size = 3;
 }
 
 void APIENTRY glBegin(GLenum mode) {
@@ -87,8 +56,6 @@ void APIENTRY glBegin(GLenum mode) {
 }
 
 void APIENTRY glColor4f(GLfloat r, GLfloat g, GLfloat b, GLfloat a) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= DIFFUSE_ENABLED_FLAG;
-
     COLOR[A8IDX] = (GLubyte)(a * 255.0f);
     COLOR[R8IDX] = (GLubyte)(r * 255.0f);
     COLOR[G8IDX] = (GLubyte)(g * 255.0f);
@@ -96,8 +63,6 @@ void APIENTRY glColor4f(GLfloat r, GLfloat g, GLfloat b, GLfloat a) {
 }
 
 void APIENTRY glColor4ub(GLubyte r, GLubyte  g, GLubyte b, GLubyte a) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= DIFFUSE_ENABLED_FLAG;
-
     COLOR[A8IDX] = a;
     COLOR[R8IDX] = r;
     COLOR[G8IDX] = g;
@@ -105,8 +70,6 @@ void APIENTRY glColor4ub(GLubyte r, GLubyte  g, GLubyte b, GLubyte a) {
 }
 
 void APIENTRY glColor4ubv(const GLubyte *v) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= DIFFUSE_ENABLED_FLAG;
-
     COLOR[A8IDX] = v[3];
     COLOR[R8IDX] = v[0];
     COLOR[G8IDX] = v[1];
@@ -114,8 +77,6 @@ void APIENTRY glColor4ubv(const GLubyte *v) {
 }
 
 void APIENTRY glColor4fv(const GLfloat* v) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= DIFFUSE_ENABLED_FLAG;
-
     COLOR[B8IDX] = (GLubyte)(v[2] * 255);
     COLOR[G8IDX] = (GLubyte)(v[1] * 255);
     COLOR[R8IDX] = (GLubyte)(v[0] * 255);
@@ -123,8 +84,6 @@ void APIENTRY glColor4fv(const GLfloat* v) {
 }
 
 void APIENTRY glColor3f(GLfloat r, GLfloat g, GLfloat b) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= DIFFUSE_ENABLED_FLAG;
-
     COLOR[B8IDX] = (GLubyte)(b * 255.0f);
     COLOR[G8IDX] = (GLubyte)(g * 255.0f);
     COLOR[R8IDX] = (GLubyte)(r * 255.0f);
@@ -132,8 +91,6 @@ void APIENTRY glColor3f(GLfloat r, GLfloat g, GLfloat b) {
 }
 
 void APIENTRY glColor3ub(GLubyte red, GLubyte green, GLubyte blue) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= DIFFUSE_ENABLED_FLAG;
-
     COLOR[A8IDX] = 255;
     COLOR[R8IDX] = red;
     COLOR[G8IDX] = green;
@@ -141,8 +98,6 @@ void APIENTRY glColor3ub(GLubyte red, GLubyte green, GLubyte blue) {
 }
 
 void APIENTRY glColor3ubv(const GLubyte *v) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= DIFFUSE_ENABLED_FLAG;
-
     COLOR[A8IDX] = 255;
     COLOR[R8IDX] = v[0];
     COLOR[G8IDX] = v[1];
@@ -150,8 +105,6 @@ void APIENTRY glColor3ubv(const GLubyte *v) {
 }
 
 void APIENTRY glColor3fv(const GLfloat* v) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= DIFFUSE_ENABLED_FLAG;
-
     COLOR[A8IDX] = 255;
     COLOR[R8IDX] = (GLubyte)(v[0] * 255);
     COLOR[G8IDX] = (GLubyte)(v[1] * 255);
@@ -159,20 +112,8 @@ void APIENTRY glColor3fv(const GLfloat* v) {
 }
 
 void APIENTRY glVertex3f(GLfloat x, GLfloat y, GLfloat z) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= VERTEX_ENABLED_FLAG;
-
-    unsigned int cap = VERTICES.capacity;
     IMVertex* vert = aligned_vector_extend(&VERTICES, 1);
 
-    if(cap != VERTICES.capacity) {
-        /* Resizing could've invalidated the pointers */
-        IM_ATTRIBS.vertex.ptr = VERTICES.data;
-        IM_ATTRIBS.uv.ptr = IM_ATTRIBS.vertex.ptr + (sizeof(GLfloat) * 3);
-        IM_ATTRIBS.st.ptr = IM_ATTRIBS.vertex.ptr + (sizeof(GLfloat) * 5);
-        IM_ATTRIBS.colour.ptr = IM_ATTRIBS.vertex.ptr + (sizeof(GLfloat) * 7);
-        IM_ATTRIBS.normal.ptr = IM_ATTRIBS.vertex.ptr + (sizeof(GLfloat) * 7) + sizeof(uint32_t);
-    }
-
     vert->x = x;
     vert->y = y;
     vert->z = z;
@@ -211,11 +152,9 @@ void APIENTRY glVertex4fv(const GLfloat* v) {
 
 void APIENTRY glMultiTexCoord2fARB(GLenum target, GLfloat s, GLfloat t) {
     if(target == GL_TEXTURE0) {
-        IM_ENABLED_VERTEX_ATTRIBUTES |= UV_ENABLED_FLAG;
         UV_COORD[0] = s;
         UV_COORD[1] = t;
     } else if(target == GL_TEXTURE1) {
-        IM_ENABLED_VERTEX_ATTRIBUTES |= ST_ENABLED_FLAG;
         ST_COORD[0] = s;
         ST_COORD[1] = t;
     } else {
@@ -225,7 +164,6 @@ void APIENTRY glMultiTexCoord2fARB(GLenum target, GLfloat s, GLfloat t) {
 }
 
 void APIENTRY glTexCoord1f(GLfloat u) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= UV_ENABLED_FLAG;
     UV_COORD[0] = u;
     UV_COORD[1] = 0.0f;
 }
@@ -235,7 +173,6 @@ void APIENTRY glTexCoord1fv(const GLfloat* v) {
 }
 
 void APIENTRY glTexCoord2f(GLfloat u, GLfloat v) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= UV_ENABLED_FLAG;
     UV_COORD[0] = u;
     UV_COORD[1] = v;
 }
@@ -245,7 +182,6 @@ void APIENTRY glTexCoord2fv(const GLfloat* v) {
 }
 
 void APIENTRY glNormal3f(GLfloat x, GLfloat y, GLfloat z) {
-    IM_ENABLED_VERTEX_ATTRIBUTES |= NORMAL_ENABLED_FLAG;
     NORMAL[0] = x;
     NORMAL[1] = y;
     NORMAL[2] = z;
@@ -258,38 +194,9 @@ void APIENTRY glNormal3fv(const GLfloat* v) {
 void APIENTRY glEnd() {
     IMMEDIATE_MODE_ACTIVE = GL_FALSE;
 
-    GLuint* attrs = &ENABLED_VERTEX_ATTRIBUTES;
-
-    /* Redirect attrib pointers */
-    AttribPointerList stashed_attrib_pointers = ATTRIB_POINTERS;
-    ATTRIB_POINTERS = IM_ATTRIBS;
-
-    GLuint prevAttrs = *attrs;
-
-    *attrs = IM_ENABLED_VERTEX_ATTRIBUTES;
-
-    /* Store the fast path enabled setting so we can restore it
-     * after drawing */
-    const GLboolean fp_was_enabled = FAST_PATH_ENABLED;
-
-#ifndef NDEBUG
-    // Immediate mode should always activate the fast path
-    GLuint fastPathEnabled = _glRecalcFastPath();
-    gl_assert(fastPathEnabled);
-#else
-    /* If we're not debugging, set to true - we assume we haven't broken it! */
-    FAST_PATH_ENABLED = GL_TRUE;
-#endif
-
-    glDrawArrays(ACTIVE_POLYGON_MODE, 0, VERTICES.size);
-
-    ATTRIB_POINTERS = stashed_attrib_pointers;
-
-    *attrs = prevAttrs;
+    glDrawPVRArrays64KOS(ACTIVE_POLYGON_MODE, 0, VERTICES.size, VERTICES.data);
 
     aligned_vector_clear(&VERTICES);
-
-    FAST_PATH_ENABLED = fp_was_enabled;
 }
 
 void APIENTRY glRectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2) {

GL/platforms/sh4.c

@@ -72,7 +72,7 @@ GL_FORCE_INLINE void _glPerspectiveDivideVertex(Vertex* vertex, const float h) {
     vertex->xyz[2] = (vertex->w == 1.0f) ? _glFastInvert(1.0001f + vertex->xyz[2]) : f;
 }
 
-GL_FORCE_INLINE void _glSubmitHeaderOrVertex(uint32_t* d, const Vertex* v) {
+GL_FORCE_INLINE void _glSubmitHeaderOrVertex(uint32_t* d, const uint32_t* s) {
 #ifndef NDEBUG
     gl_assert(!isnan(v->xyz[2]));
     gl_assert(!isnan(v->w));
@@ -81,8 +81,6 @@ GL_FORCE_INLINE void _glSubmitHeaderOrVertex(uint32_t* d, const Vertex* v) {
 #if CLIP_DEBUG
     printf("Submitting: %x (%x)\n", v, v->flags);
 #endif
-
-    uint32_t *s = (uint32_t*) v;
     __asm__("pref @%0" : : "r"(s + 8));  /* prefetch 32 bytes for next loop */
     d[0] = *(s++);
     d[1] = *(s++);
@@ -172,37 +170,28 @@ GL_FORCE_INLINE void ShiftRotateTriangle() {
 
 #define SPAN_SORT_CFG 0x005F8030
 
-void SceneListSubmit(void* src, int n) {
+static inline void submit_unclipped(uint32_t* d, Vertex* vertex, int n) {
     const float h = GetVideoMode()->height;
 
-    PVR_SET(SPAN_SORT_CFG, 0x0);
-
-    uint32_t *d = (uint32_t*) SQ_BASE_ADDRESS;
-    *PVR_LMMODE0 = 0x0; /* Enable 64bit mode */
-
-    Vertex __attribute__((aligned(32))) tmp;
-
-    /* Perform perspective divide on each vertex */
-    Vertex* vertex = (Vertex*) src;
-
-    if(!_glNearZClippingEnabled()) {
-        /* Prep store queues */
-
-        for(int i = 0; i < n; ++i, ++vertex) {
-            PREFETCH(vertex + 1);
-            if(glIsVertex(vertex->flags)) {
-                _glPerspectiveDivideVertex(vertex, h);
-            }
-            _glSubmitHeaderOrVertex(d, vertex);
+    for(int i = 0; i < n; ++i, ++vertex) {
+        PREFETCH(vertex + 1);
+        if(glIsVertex(vertex->flags)) {
+            _glPerspectiveDivideVertex(vertex, h);
         }
-
-        /* Wait for both store queues to complete */
-        d = (uint32_t *) SQ_BASE_ADDRESS;
-        d[0] = d[8] = 0;
-
-        return;
+        _glSubmitHeaderOrVertex(d, (const uint32_t*) vertex);
     }
 
+    /* Wait for both store queues to complete */
+    d = (uint32_t *) SQ_BASE_ADDRESS;
+    d[0] = d[8] = 0;
+
+    return;
+}
+
+static inline void submit_clipped(uint32_t* d, Vertex* vertex, int n) {
+    static Vertex __attribute__((aligned(32))) tmp;
+    const float h = GetVideoMode()->height;
+
     tri_count = 0;
     strip_count = 0;
 
@@ -226,7 +215,7 @@ void SceneListSubmit(void* src, int n) {
                 /* We hit a header */
                 tri_count = 0;
                 strip_count = 0;
-                _glSubmitHeaderOrVertex(d, vertex);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) vertex);
                 continue;
             }
         }
@@ -257,11 +246,11 @@ void SceneListSubmit(void* src, int n) {
         tmp = *(vertex - 2); \
         /* If we had triangles ahead of this one, submit and finalize */ \
         _glPerspectiveDivideVertex(&tmp, h); \
-        _glSubmitHeaderOrVertex(d, &tmp); \
+        _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp); \
         tmp = *(vertex - 1); \
         tmp.flags = GPU_CMD_VERTEX_EOL; \
         _glPerspectiveDivideVertex(&tmp, h); \
-        _glSubmitHeaderOrVertex(d, &tmp); \
+        _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp); \
     }
 
         bool is_last_in_strip = glIsLastVertex(vertex->flags);
@@ -273,17 +262,17 @@ void SceneListSubmit(void* src, int n) {
                 tmp = *triangle[0].v;
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[0].v, triangle[1].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[2].v, triangle[0].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX_EOL;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
             } break;
             case 2: {
                 SUBMIT_QUEUED();
@@ -291,17 +280,17 @@ void SceneListSubmit(void* src, int n) {
                 _glClipEdge(triangle[0].v, triangle[1].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 tmp = *triangle[1].v;
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[1].v, triangle[2].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX_EOL;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
             } break;
             case 3: {
                 SUBMIT_QUEUED();
@@ -309,22 +298,22 @@ void SceneListSubmit(void* src, int n) {
                 tmp = *triangle[0].v;
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 tmp = *triangle[1].v;
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[2].v, triangle[0].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[1].v, triangle[2].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX_EOL;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
             } break;
             case 4: {
                 SUBMIT_QUEUED();
@@ -332,17 +321,17 @@ void SceneListSubmit(void* src, int n) {
                 _glClipEdge(triangle[1].v, triangle[2].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 tmp = *triangle[2].v;
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[2].v, triangle[0].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX_EOL;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
             } break;
             case 5: {
                 SUBMIT_QUEUED();
@@ -350,22 +339,22 @@ void SceneListSubmit(void* src, int n) {
                 tmp = *triangle[0].v;
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[0].v, triangle[1].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 tmp = *triangle[2].v;
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[1].v, triangle[2].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX_EOL;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
             } break;
             case 6: {
                 SUBMIT_QUEUED();
@@ -373,33 +362,33 @@ void SceneListSubmit(void* src, int n) {
                 _glClipEdge(triangle[0].v, triangle[1].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 tmp = *triangle[1].v;
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 _glClipEdge(triangle[2].v, triangle[0].v, &tmp);
                 tmp.flags = GPU_CMD_VERTEX;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
 
                 tmp = *triangle[2].v;
                 tmp.flags = GPU_CMD_VERTEX_EOL;
                 _glPerspectiveDivideVertex(&tmp, h);
-                _glSubmitHeaderOrVertex(d, &tmp);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) &tmp);
             } break;
             case 7: {
                 /* All the vertices are visible! We divide and submit v0, then shift */
                 _glPerspectiveDivideVertex(vertex - 2, h);
-                _glSubmitHeaderOrVertex(d, vertex - 2);
+                _glSubmitHeaderOrVertex(d, (const uint32_t*) (vertex - 2));
 
                 if(is_last_in_strip) {
                     _glPerspectiveDivideVertex(vertex - 1, h);
-                    _glSubmitHeaderOrVertex(d, vertex - 1);
+                    _glSubmitHeaderOrVertex(d, (const uint32_t*) (vertex - 1));
                     _glPerspectiveDivideVertex(vertex, h);
-                    _glSubmitHeaderOrVertex(d, vertex);
+                    _glSubmitHeaderOrVertex(d, (const uint32_t*) vertex);
                     tri_count = 0;
                     strip_count = 0;
                 }
@@ -428,6 +417,25 @@ void SceneListSubmit(void* src, int n) {
     d[0] = d[8] = 0;
 }
 
+void SceneListSubmit(void* src, int n) {
+    PVR_SET(SPAN_SORT_CFG, 0x0);
+
+    uint32_t *d = (uint32_t*) SQ_BASE_ADDRESS;
+
+    *PVR_LMMODE0 = 0x0; /* Enable 64bit mode */
+    *((volatile int *)0xA05F6888) = 1;
+
+    /* Perform perspective divide on each vertex */
+    Vertex* vertex = (Vertex*) src;
+
+    if(!_glNearZClippingEnabled()) {
+        /* Prep store queues */
+        submit_unclipped(d, vertex, n);
+    } else {
+        submit_clipped(d, vertex, n);
+    }
+}
+
 void SceneListFinish() {
     pvr_list_finish();
 }

GL/private.h

@@ -245,6 +245,8 @@ do {                 \
 typedef struct {
     float nxyz[3];
     float st[2];
+    float w;
+    uint32_t padding[2];
 } VertexExtra;
 
 /* Generating PVR vertices from the user-submitted data gets complicated, particularly

include/GL/glkos.h

@@ -193,5 +193,60 @@ GLAPI GLvoid APIENTRY glDefragmentTextureMemory_KOS(void);
 //for palette internal format (glfcConfig)
 #define GL_RGB565_KOS                               0xEF40
 
+/**
+ * This is a custom extension to OpenGL that allows submitting data
+ * directly in the expected format and alignment of the Dreamcast PowerVR
+ * chip.
+ *
+ * Vertices sent to via this API should be 4-byte aligned (ideally 32) and
+ * use the following layout:
+ *
+ * uint32_t padding; // (Space for the PVR command header)
+ * float xyz[3]; // Pos
+ * float uv[2];  // Tex coord
+ * uint8_t bgra[4]; // Vertex color
+ * uint8_t obgra[4]; // Offset color (currently unused)
+ *
+ * This layout is 32 bytes in size and is copied directly into the PVR command stream.
+ *
+ * However, GL also allows for other attributes: normals, and secondary texture coordinates.
+ *
+ * For that reason, you can use this API to submit vertices in an "extended" format which
+ * is as follows:
+ *
+ * uint32_t header; // PVR command header
+ * float xyz[3]; // Pos
+ * float uv[2];  // Tex coord
+ * uint8_t bgra[4]; // Vertex color
+ * uint8_t obgra[4]; // Offset color (currently unused)
+ * float nxyz[3]; // Normal
+ * float st[2]; // Secondary texture coords
+ * uint32_t padding[3]; // Internal use.
+ */
+
+/* Draw polygons by directly copying the vertices into the PVR command stream. Transformation
+   will happen as usual. Enabled client state *is NOT* respected.
+   ERRORS:
+
+    - GL_INVALID_ENUM is generated if mode is not an accepted value (GL_TRIANGLES, GL_QUADS, GL_TRIANGLE_STRIP are accepted)
+    - GL_INVALID_VALUE is generated if count is negative
+    - GL_INVALID_VALUE is generated if data is NULL
+*/
+void glDrawPVRArrays32KOS(GLenum mode, GLint first, GLsizei count, void* data);
+
+/* Draw polygons by directly copying the vertices into the PVR command stream. Transformation
+   will happen as usual. Enabled client state *is NOT* respected. This uses the extended vertex format which includes
+   normals and secondary texture coordinates. These will only take effect if you've enabled secondary texture coordinates
+   with glEnable (for ST) or enabled lighting (for normals). GL_NORMALIZE is not respected.
+
+   ERRORS:
+
+    - GL_INVALID_ENUM is generated if mode is not an accepted value (GL_TRIANGLES, GL_QUADS, GL_TRIANGLE_STRIP are accepted)
+    - GL_INVALID_VALUE is generated if count is negative
+    - GL_INVALID_VALUE is generated if data is NULL
+*/
+void glDrawPVRArrays64KOS(GLenum mode, GLint first, GLsizei count, void* data);
+
+
 __END_DECLS
 

samples/quadmark/main.c

@@ -69,7 +69,11 @@ int check_start() {
 
 void setup() {
     //PVR needs to warm up for a frame, or results will be low
-    glKosInit();
+    GLdcConfig config;
+    glKosInitConfig(&config);
+    config.initial_op_capacity = 7000 * 4;
+    config.initial_immediate_capacity = 7000 * 4;
+    glKosInitEx(&config);
     glMatrixMode(GL_MODELVIEW);
     glLoadIdentity();
     glOrtho(0, 640, 0, 480, -100, 100);

samples/terrain/main.c

@@ -9,7 +9,7 @@
 #include "GL/glu.h"
 #include "GL/glkos.h"
 
-#define TERRAIN_SIZE 100
+#define TERRAIN_SIZE 75
 #define TERRAIN_SCALE 1.0f
 #define TERRAIN_HEIGHT_SCALE 1.0f
 
@@ -155,11 +155,29 @@ int main(int argc, char **argv)
     InitGL(640, 480);
     ReSizeGLScene(640, 480);
 
+    uint64_t accum = 0;
+    uint64_t last_time = timer_us_gettime64();
+    uint32_t frames = 0;
+
     while(1) {
         if(check_start())
             break;
 
         DrawGLScene();
+
+        ++frames;
+        uint64_t now = timer_us_gettime64();
+        uint64_t diff = (now - last_time);
+        last_time = now;
+        accum += diff;
+
+        if(accum > 5000000) {
+            printf("Frame time: %f\n", (float)accum / frames / 1000000.0f);
+            printf("FPS: %f\n", ((float) frames) / 5.0f);
+            fflush(stdout);
+            frames = 0;
+            accum = 0;
+        }
     }
 
     return 0;