/**

 * $Id$

 *

 * PVR2 Tile Accelerator implementation

 *

 * Copyright (c) 2005 Nathan Keynes.

 *

 * This program is free software; you can redistribute it and/or modify

 * it under the terms of the GNU General Public License as published by

 * the Free Software Foundation; either version 2 of the License, or

 * (at your option) any later version.

 *

 * This program is distributed in the hope that it will be useful,

 * but WITHOUT ANY WARRANTY; without even the implied warranty of

 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

 * GNU General Public License for more details.

 */

#include <string.h>

#include "lxdream.h"

#include "pvr2/pvr2.h"

#include "pvr2/pvr2mmio.h"

#include "asic.h"

#include "dream.h"

#define STATE_IDLE                 0

#define STATE_IN_LIST              1

#define STATE_IN_POLYGON           2

#define STATE_EXPECT_POLY_BLOCK2   3

#define STATE_EXPECT_VERTEX_BLOCK2 4

#define STATE_ERROR                5

#define STATE_EXPECT_END_VERTEX_BLOCK2 7

#define TA_CMD(i) ( (i) >> 29 )

#define TA_CMD_END_LIST 0

#define TA_CMD_CLIP 1

#define TA_CMD_POLYGON_CONTEXT 4

#define TA_CMD_SPRITE_CONTEXT 5

#define TA_CMD_VERTEX 7

#define TA_LIST_NONE -1

#define TA_LIST_OPAQUE 0

#define TA_LIST_OPAQUE_MOD 1

#define TA_LIST_TRANS 2

#define TA_LIST_TRANS_MOD 3

#define TA_LIST_PUNCH_OUT 4

#define TA_IS_MODIFIER_LIST(list) (list == TA_LIST_OPAQUE_MOD || list == TA_LIST_TRANS_MOD)

#define TA_GROW_UP 0

#define TA_GROW_DOWN 1

#define TA_VERTEX_NONE                        -1

#define TA_VERTEX_PACKED                      0x00

#define TA_VERTEX_TEX_PACKED                  0x08

#define TA_VERTEX_TEX_SPEC_PACKED             0x0C

#define TA_VERTEX_TEX_UV16_PACKED             0x09

#define TA_VERTEX_TEX_UV16_SPEC_PACKED        0x0D

#define TA_VERTEX_FLOAT                       0x10

#define TA_VERTEX_TEX_FLOAT                   0x18

#define TA_VERTEX_TEX_SPEC_FLOAT              0x1C

#define TA_VERTEX_TEX_UV16_FLOAT              0x19

#define TA_VERTEX_TEX_UV16_SPEC_FLOAT         0x1D

#define TA_VERTEX_INTENSITY                   0x20

#define TA_VERTEX_TEX_INTENSITY               0x28

#define TA_VERTEX_TEX_SPEC_INTENSITY          0x2C

#define TA_VERTEX_TEX_UV16_INTENSITY          0x29

#define TA_VERTEX_TEX_UV16_SPEC_INTENSITY     0x2D

#define TA_VERTEX_PACKED_MOD                  0x40

#define TA_VERTEX_TEX_PACKED_MOD              0x48

#define TA_VERTEX_TEX_SPEC_PACKED_MOD         0x4C

#define TA_VERTEX_TEX_UV16_PACKED_MOD         0x49

#define TA_VERTEX_TEX_UV16_SPEC_PACKED_MOD    0x4D

#define TA_VERTEX_INTENSITY_MOD               0x60

#define TA_VERTEX_TEX_INTENSITY_MOD           0x68

#define TA_VERTEX_TEX_SPEC_INTENSITY_MOD      0x6C

#define TA_VERTEX_TEX_UV16_INTENSITY_MOD      0x69

#define TA_VERTEX_TEX_UV16_SPEC_INTENSITY_MOD 0x6D

#define TA_VERTEX_SPRITE                      0x80

#define TA_VERTEX_TEX_SPRITE                  0x88

#define TA_VERTEX_MOD_VOLUME                  0x81

#define TA_VERTEX_LISTLESS                    0xFF

#define TA_IS_NORMAL_POLY() (ta_status.current_vertex_type < TA_VERTEX_SPRITE)

static int strip_lengths[4] = {3,4,6,8}; /* in vertexes */

#define TA_POLYCMD_LISTTYPE(i) ( ((i) >> 24) & 0x0F )

#define TA_POLYCMD_USELENGTH(i) ( i & 0x00800000 )

#define TA_POLYCMD_LENGTH(i)  strip_lengths[((i >> 18) & 0x03)]

#define TA_POLYCMD_CLIP(i)  ((i>>16)&0x03)

#define TA_POLYCMD_CLIP_NONE 0

#define TA_POLYCMD_CLIP_INSIDE 2

#define TA_POLYCMD_CLIP_OUTSIDE 3

#define TA_POLYCMD_COLOURFMT(i)  (i & 0x00000030)

#define TA_POLYCMD_COLOURFMT_ARGB32 0x00000000

#define TA_POLYCMD_COLOURFMT_FLOAT 0x00000010

#define TA_POLYCMD_COLOURFMT_INTENSITY 0x00000020

#define TA_POLYCMD_COLOURFMT_LASTINT 0x00000030

#define TA_POLYCMD_MODIFIED 0x00000080

#define TA_POLYCMD_FULLMOD  0x00000040

#define TA_POLYCMD_TEXTURED 0x00000008

#define TA_POLYCMD_SPECULAR 0x00000004

#define TA_POLYCMD_SHADED 0x00000002

#define TA_POLYCMD_UV16 0x00000001

#define TA_POLYCMD_IS_SPECULAR(i) ((i & 0x0000000C)==0x0000000C) /* Only applies to textured polys */

#define TA_POLYCMD_IS_FULLMOD(i) ((i & 0x000000C0)==0x000000C0)

#define TA_IS_END_VERTEX(i) (i & 0x10000000)

/** Note these are not the IEEE 754 definitions - the TA treats NANs

 * as if they were INFs of the appropriate sign.

 */

#define TA_IS_INF(f) (((*((uint32_t *)&f)) & 0xFF800000) == 0x7F800000)

#define TA_IS_NINF(f) (((*((uint32_t *)&f)) & 0xFF800000) == 0xFF800000)

#define MIN3( x1, x2, x3 ) ( (x1)<(x2)? ((x1)<(x3)?(x1):(x3)) : ((x2)<(x3)?(x2):(x3)) )

#define MAX3( x1, x2, x3 ) ( (x1)>(x2)? ((x1)>(x3)?(x1):(x3)) : ((x2)>(x3)?(x2):(x3)) )

#define TILESLOT( x, y ) (ta_status.current_tile_matrix + (ta_status.current_tile_size * (y * ta_status.width+ x) << 2))

extern char *video_base;

#define PVRRAM(addr) (*(uint32_t *)(video_base + ((addr)&PVR2_RAM_MASK)))

struct pvr2_ta_vertex {

    float x,y,z;

    uint32_t detail[8]; /* 0-8 detail words */

};

struct tile_bounds {

    int32_t x1, y1, x2, y2;

};

struct pvr2_ta_status {

    int32_t state;

    int32_t width, height; /* Tile resolution, ie 20x15 */

    int32_t tilelist_dir; /* Growth direction of the tilelist, 0 = up, 1 = down */

    uint32_t tilelist_size; /* Size of the tilelist segments */

    uint32_t tilelist_start; /* Initial address of the tilelist */

    uint32_t polybuf_start; /* Initial bank address of the polygon buffer (ie &0x00F00000) */

    int32_t current_vertex_type;

    uint32_t accept_vertexes; /* 0 = NO, 1 = YES */

    int32_t vertex_count; /* index of last start-vertex seen, or -1 if no vertexes 

			 * are present

			 */

    uint32_t max_vertex;     /* Maximum number of vertexes in the current polygon (3/4/6/8) */

    uint32_t current_list_type;

    uint32_t current_tile_matrix; /* Memory location of the first tile for the current list. */

    uint32_t current_tile_size; /* Size of the tile matrix space  in 32-bit words (0/8/16/32)*/

    uint32_t intensity1, intensity2;

    struct tile_bounds clip;

    int32_t clip_mode;

    /**

     * Current working object

     */

    int32_t poly_context_size;

    int32_t poly_vertex_size;

    int32_t poly_parity;

    uint32_t poly_context[5];

    uint32_t poly_pointer;

    struct tile_bounds last_triangle_bounds;

    struct pvr2_ta_vertex poly_vertex[8];

    uint32_t debug_output;

};

static struct pvr2_ta_status ta_status;

static int tilematrix_sizes[4] = {0,8,16,32};

/**

 * Convenience union - ta data is either 32-bit integer or 32-bit float.

 */

union ta_data {

    unsigned int i;

    float f;

};

void pvr2_ta_reset() {

    ta_status.state = STATE_ERROR; /* State not valid until initialized */

    ta_status.debug_output = 0;

}

void pvr2_ta_save_state( FILE *f )

{

    fwrite( &ta_status, sizeof(ta_status), 1, f );

}

int pvr2_ta_load_state( FILE *f )

{

    if( fread( &ta_status, sizeof(ta_status), 1, f ) != 1 )

	return 1;

    return 0;

}

void pvr2_ta_init() {

    ta_status.state = STATE_IDLE;

    ta_status.current_list_type = -1;

    ta_status.current_vertex_type = -1;

    ta_status.poly_parity = 0;

    ta_status.vertex_count = 0;

    ta_status.max_vertex = 3;

    ta_status.current_vertex_type = TA_VERTEX_LISTLESS;

    ta_status.poly_vertex_size = 0;

    memset(&ta_status.poly_context[1], 0, 4);

    ta_status.last_triangle_bounds.x1 = -1;

    ta_status.accept_vertexes = TRUE;

    ta_status.clip.x1 = 0;

    ta_status.clip.y1 = 0;

    ta_status.clip_mode = TA_POLYCMD_CLIP_NONE;

    uint32_t size = MMIO_READ( PVR2, TA_TILESIZE );

    ta_status.width = (size & 0xFFFF) + 1;

    ta_status.height = (size >> 16) + 1;

    ta_status.clip.x2 = ta_status.width-1;

    ta_status.clip.y2 = ta_status.height-1;

    uint32_t control = MMIO_READ( PVR2, TA_TILECFG );

    ta_status.tilelist_dir = (control >> 20) & 0x01;

    ta_status.tilelist_size = tilematrix_sizes[ (control & 0x03) ];

    MMIO_WRITE( PVR2, TA_POLYPOS, MMIO_READ( PVR2, TA_POLYBASE ) );

    uint32_t plistpos = MMIO_READ( PVR2, TA_LISTBASE ) >> 2;

    if( ta_status.tilelist_dir == TA_GROW_DOWN ) {

	plistpos -= ta_status.tilelist_size;

    }

    MMIO_WRITE( PVR2, TA_LISTPOS, plistpos );

    ta_status.tilelist_start = plistpos;

    ta_status.polybuf_start = MMIO_READ( PVR2, TA_POLYBASE ) & 0x00F00000;

}

static uint32_t parse_float_colour( float a, float r, float g, float b ) {

    int ai,ri,gi,bi;

    if( TA_IS_INF(a) ) {

	ai = 255;

    } else {

	ai = 256 * CLAMP(a,0.0,1.0) - 1;

	if( ai < 0 ) ai = 0;

    }

    if( TA_IS_INF(r) ) {

	ri = 255;

    } else {

	ri = 256 * CLAMP(r,0.0,1.0) - 1;

	if( ri < 0 ) ri = 0;

    }

    if( TA_IS_INF(g) ) {

	gi = 255;

    } else {

	gi = 256 * CLAMP(g,0.0,1.0) - 1;

	if( gi < 0 ) gi = 0;

    }

    if( TA_IS_INF(b) ) {

	bi = 255;

    } else {

	bi = 256 * CLAMP(b,0.0,1.0) - 1;

	if( bi < 0 ) bi = 0;

    }

    return (ai << 24) | (ri << 16) | (gi << 8) | bi;

}

static uint32_t parse_intensity_colour( uint32_t base, float intensity )

{

    unsigned int i = (unsigned int)(256 * CLAMP(intensity, 0.0,1.0));

    return

	(((((base & 0xFF) * i) & 0xFF00) |

	  (((base & 0xFF00) * i) & 0xFF0000) |

	  (((base & 0xFF0000) * i) & 0xFF000000)) >> 8) |

	(base & 0xFF000000);

}

/**

 * Initialize the specified TA list.

 */

static void ta_init_list( unsigned int listtype ) {

    int config = MMIO_READ( PVR2, TA_TILECFG );

    int tile_matrix = MMIO_READ( PVR2, TA_TILEBASE );

    int list_end = MMIO_READ( PVR2, TA_LISTEND );

    ta_status.current_tile_matrix = tile_matrix;

    /* If the list grows down, the end must be < tile matrix start. 

     * If it grows up, the end must be > tile matrix start.

     * Don't ask me why, it just does...

     */

    if( ((ta_status.tilelist_dir == TA_GROW_DOWN && list_end <= tile_matrix) ||

	 (ta_status.tilelist_dir == TA_GROW_UP && list_end >= tile_matrix )) &&

	listtype <= TA_LIST_PUNCH_OUT ) {

	int i;

	uint32_t *p;

	for( i=0; i < listtype; i++ ) {

	    int size = tilematrix_sizes[(config & 0x03)] << 2;

	    ta_status.current_tile_matrix += ta_status.width * ta_status.height * size;

	    config >>= 4;

	}

	ta_status.current_tile_size = tilematrix_sizes[(config & 0x03)];

	/* Initialize each tile to 0xF0000000 */

	if( ta_status.current_tile_size != 0 ) {

	    p = (uint32_t *)(video_base + ta_status.current_tile_matrix);

	    for( i=0; i< ta_status.width * ta_status.height; i++ ) {

		*p = 0xF0000000;

		p += ta_status.current_tile_size;

	    }

	}

    } else {

	ta_status.current_tile_size = 0;

    }

    if( tile_matrix == list_end ) {

	ta_status.current_tile_size = 0;

    }

    ta_status.state = STATE_IN_LIST;

    ta_status.current_list_type = listtype;

    ta_status.last_triangle_bounds.x1 = -1;

}

static int list_events[5] = {EVENT_PVR_OPAQUE_DONE, EVENT_PVR_OPAQUEMOD_DONE, 

			     EVENT_PVR_TRANS_DONE, EVENT_PVR_TRANSMOD_DONE,

			     EVENT_PVR_PUNCHOUT_DONE };

static void ta_end_list() {

    if( ta_status.current_list_type != TA_LIST_NONE ) {

	asic_event( list_events[ta_status.current_list_type] );

    }

    ta_status.current_list_type = TA_LIST_NONE;

    ta_status.current_vertex_type = TA_VERTEX_LISTLESS;

    ta_status.poly_vertex_size = 0;

    memset(&ta_status.poly_context[1], 0, 4);

    ta_status.state = STATE_IDLE;

}

static void ta_bad_input_error() {

    asic_event( EVENT_PVR_BAD_INPUT );

}

/**

 * Write data out to the polygon buffer.

 * If the end-of-buffer is reached, asserts EVENT_PVR_PRIM_ALLOC_FAIL

 * @param data to be written

 * @param length Number of 32-bit words to write.

 * @return number of words actually written

 */

static int ta_write_polygon_buffer( uint32_t *data, int length )

{

    int rv;

    int posn = MMIO_READ( PVR2, TA_POLYPOS );

    int end = MMIO_READ( PVR2, TA_POLYEND );

    uint32_t *target = (uint32_t *)(video_base + posn);

    for( rv=0; rv < length; rv++ ) {

	if( posn == end ) {

	    asic_event( EVENT_PVR_PRIM_ALLOC_FAIL );

	    //	    ta_status.state = STATE_ERROR;

	    break;

	}

	if( posn < PVR2_RAM_SIZE ) {

	    *target++ = *data++;

	}

	posn += 4;

    }

    MMIO_WRITE( PVR2, TA_POLYPOS, posn );

    return rv;

}

#define TA_NO_ALLOC 0xFFFFFFFF

/**

 * Allocate a new tile list block from the grow space and update the

 * word at reference to be a link to the new block.

 */

static uint32_t ta_alloc_tilelist( uint32_t reference ) {

    uint32_t posn = MMIO_READ( PVR2, TA_LISTPOS );

    uint32_t limit = MMIO_READ( PVR2, TA_LISTEND ) >> 2;

    uint32_t newposn;

    if( ta_status.tilelist_dir == TA_GROW_DOWN ) {

	newposn = posn - ta_status.tilelist_size;

	if( posn == limit ) {

	    PVRRAM(posn<<2) = 0xF0000000;

	    PVRRAM(reference) = 0xE0000000 | (posn<<2);

	    return TA_NO_ALLOC;

	} else if( posn < limit ) {

	    PVRRAM(reference) = 0xE0000000 | (posn<<2);

	    return TA_NO_ALLOC;

	} else if( newposn <= limit ) {

	} else if( newposn <= (limit + ta_status.tilelist_size) ) {

	    asic_event( EVENT_PVR_MATRIX_ALLOC_FAIL );

	    MMIO_WRITE( PVR2, TA_LISTPOS, newposn );

	} else {

	    MMIO_WRITE( PVR2, TA_LISTPOS, newposn );

	}

	PVRRAM(reference) = 0xE0000000 | (posn<<2);

	return posn << 2;

    } else {

	newposn = posn + ta_status.tilelist_size;

	if( posn == limit ) {

	    PVRRAM(posn<<2) = 0xF0000000;

	    PVRRAM(reference) = 0xE0000000 | (posn<<2);

	    return TA_NO_ALLOC;

	} else if ( posn > limit ) {

	    PVRRAM(reference) = 0xE0000000 | (posn<<2);

	    return TA_NO_ALLOC;

	} else if( newposn >= limit ) {

	} else if( newposn >= (limit - ta_status.tilelist_size) ) {

	    asic_event( EVENT_PVR_MATRIX_ALLOC_FAIL );

	    MMIO_WRITE( PVR2, TA_LISTPOS, newposn );

	} else {

	    MMIO_WRITE( PVR2, TA_LISTPOS, newposn );

	}	    

	PVRRAM(reference) = 0xE0000000 | (posn<<2);

	return posn << 2;

    }

}

/**

 * Write a tile entry out to the matrix.

 */

static void ta_write_tile_entry( int x, int y, uint32_t tile_entry ) {

    uint32_t tile = TILESLOT(x,y);

    uint32_t tilestart = tile;

    uint32_t value;

    uint32_t lasttri = 0;

    int i;

    if( ta_status.clip_mode == TA_POLYCMD_CLIP_OUTSIDE &&

	x >= ta_status.clip.x1 && x <= ta_status.clip.x2 &&

	y >= ta_status.clip.y1 && y <= ta_status.clip.y2 ) {

	/* Tile clipped out */

	return;

    }

    if( (tile_entry & 0x80000000) && 

	ta_status.last_triangle_bounds.x1 != -1 &&

	ta_status.last_triangle_bounds.x1 <= x &&

	ta_status.last_triangle_bounds.x2 >= x &&

	ta_status.last_triangle_bounds.y1 <= y &&

	ta_status.last_triangle_bounds.y2 >= y ) {

	/* potential for triangle stacking */

	lasttri = tile_entry & 0xE1E00000;

    }

    if( PVRRAM(tile) == 0xF0000000 ) {

	PVRRAM(tile) = tile_entry;

	PVRRAM(tile+4) = 0xF0000000;

	return;

    }

    while(1) {

	value = PVRRAM(tile);

	for( i=1; i<ta_status.current_tile_size; i++ ) {

	    tile += 4;

	    uint32_t nextval = PVRRAM(tile);

	    if( nextval == 0xF0000000 ) {

		if( lasttri != 0 && lasttri == (value&0xE1E00000) ) {

		    int count = (value & 0x1E000000) + 0x02000000;

		    if( count < 0x20000000 ) {

			PVRRAM(tile-4) = (value & 0xE1FFFFFF) | count;

			return;

		    }

		}

		if( i < ta_status.current_tile_size-1 ) {

		    PVRRAM(tile) = tile_entry;

		    PVRRAM(tile+4) = 0xF0000000;

		    return;

		}

	    }

	    value = nextval;

	}

	if( value == 0xF0000000 ) {

	    tile = ta_alloc_tilelist(tile);

	    if( tile != TA_NO_ALLOC ) {

		PVRRAM(tile) = tile_entry;

		PVRRAM(tile+4) = 0xF0000000;

	    }

	    return;

	} else if( (value & 0xFF000000) == 0xE0000000 ) {

	    value &= 0x00FFFFFF;

	    if( value == tilestart )

		return; /* Loop */

	    tilestart = tile = value;

	} else {

	    /* This should never happen */

	    return;

	}

    }

}

/**

 * Write a completed polygon out to the memory buffers 

 * OPTIMIZEME: This is not terribly efficient at the moment.

 */

static void ta_commit_polygon( ) {

    int i, x, y;

    int tx[ta_status.vertex_count], ty[ta_status.vertex_count];

    struct tile_bounds triangle_bound[ta_status.vertex_count - 2];

    struct tile_bounds polygon_bound;

    uint32_t poly_context[5];

    memcpy( poly_context, ta_status.poly_context, ta_status.poly_context_size * 4 );

    /* Compute the tile coordinates for each vertex (need to be careful with

     * clamping here)

     */

    for( i=0; i<ta_status.vertex_count; i++ ) {

	if( ta_status.poly_vertex[i].x < 0.0 || TA_IS_NINF(ta_status.poly_vertex[i].x) ) {

	    tx[i] = -1;

	} else if( ta_status.poly_vertex[i].x > (float)INT_MAX || TA_IS_INF(ta_status.poly_vertex[i].x) ) {

	    tx[i] = INT_MAX/32;

	} else {

	    tx[i] = (int)(ta_status.poly_vertex[i].x / 32.0);

	}

	if( ta_status.poly_vertex[i].y < 0.0 || TA_IS_NINF(ta_status.poly_vertex[i].y)) {

	    ty[i] = -1;

	} else if( ta_status.poly_vertex[i].y > (float)INT_MAX || TA_IS_INF(ta_status.poly_vertex[i].y) ) {

	    ty[i] = INT_MAX/32;

	} else {

	    ty[i] = (int)(ta_status.poly_vertex[i].y / 32.0);

	}

    }

    /* Compute bounding box for each triangle individually, as well

     * as the overall polygon.

     */

    for( i=0; i<ta_status.vertex_count-2; i++ ) {

	triangle_bound[i].x1 = MIN3(tx[i],tx[i+1],tx[i+2]);

	triangle_bound[i].x2 = MAX3(tx[i],tx[i+1],tx[i+2]);

	triangle_bound[i].y1 = MIN3(ty[i],ty[i+1],ty[i+2]);

	triangle_bound[i].y2 = MAX3(ty[i],ty[i+1],ty[i+2]);

	if( i == 0 ) {

	    polygon_bound.x1 = triangle_bound[0].x1;

	    polygon_bound.y1 = triangle_bound[0].y1;

	    polygon_bound.x2 = triangle_bound[0].x2;

	    polygon_bound.y2 = triangle_bound[0].y2;

	} else {

	    polygon_bound.x1 = MIN(polygon_bound.x1, triangle_bound[i].x1);

	    polygon_bound.x2 = MAX(polygon_bound.x2, triangle_bound[i].x2);

	    polygon_bound.y1 = MIN(polygon_bound.y1, triangle_bound[i].y1);

	    polygon_bound.y2 = MAX(polygon_bound.y2, triangle_bound[i].y2);

	}

    }

    /* Clamp the polygon bounds to the frustum */

    if( polygon_bound.x1 < 0 ) polygon_bound.x1 = 0;

    if( polygon_bound.x2 >= ta_status.width ) polygon_bound.x2 = ta_status.width-1;

    if( polygon_bound.y1 < 0 ) polygon_bound.y1 = 0;

    if( polygon_bound.y2 >= ta_status.width ) polygon_bound.y2 = ta_status.height-1;

    /* Set the "single tile" flag if it's entirely contained in 1 tile */

    if( polygon_bound.x1 == polygon_bound.x2 &&

	polygon_bound.y1 == polygon_bound.y2 ) {

	poly_context[0] |= 0x00200000;

    }

    /* If the polygon is entirely clipped, don't even write the polygon data */

    switch( ta_status.clip_mode ) {

    case TA_POLYCMD_CLIP_NONE:

	if( polygon_bound.x2 < 0 || polygon_bound.x1 >= ta_status.width ||

	    polygon_bound.y2 < 0 || polygon_bound.y1 >= ta_status.height ) {

	    return;

	}

	break;

    case TA_POLYCMD_CLIP_INSIDE:

	if( polygon_bound.x2 < ta_status.clip.x1 || polygon_bound.x1 > ta_status.clip.x2 ||

	    polygon_bound.y2 < ta_status.clip.y1 || polygon_bound.y1 > ta_status.clip.y2 ) {

	    return;

	} else {

	    /* Clamp to clip bounds */

	    if( polygon_bound.x1 < ta_status.clip.x1 ) polygon_bound.x1 = ta_status.clip.x1;

	    if( polygon_bound.x2 > ta_status.clip.x2 ) polygon_bound.x2 = ta_status.clip.x2;

	    if( polygon_bound.y1 < ta_status.clip.y1 ) polygon_bound.y1 = ta_status.clip.y1;

	    if( polygon_bound.y2 > ta_status.clip.y2 ) polygon_bound.y2 = ta_status.clip.y2;

	}

	break;

    case TA_POLYCMD_CLIP_OUTSIDE:

	if( polygon_bound.x1 >= ta_status.clip.x1 && polygon_bound.x2 <= ta_status.clip.x2 &&

	    polygon_bound.y1 >= ta_status.clip.y1 && polygon_bound.y2 <= ta_status.clip.y2 ) {

	    return;

	}

	break;

    }

    /* Ok, we're good to go - write out the polygon first */

    uint32_t tile_entry = (MMIO_READ( PVR2, TA_POLYPOS ) - ta_status.polybuf_start) >> 2 | 

	ta_status.poly_pointer;

    int status = ta_write_polygon_buffer( poly_context, ta_status.poly_context_size );

    if( status == 0 ) {

	/* No memory available - abort */

	return;

    } else {

	for( i=0; i<ta_status.vertex_count && status != 0; i++ ) {

	    status = ta_write_polygon_buffer( (uint32_t *)(&ta_status.poly_vertex[i]), 3 + ta_status.poly_vertex_size );

	}

    }

    if( ta_status.current_tile_size == 0 ) {

	/* No memory for tile entry, so don't write anything */

	return;

    }

    /* And now the tile entries. Triangles are different from everything else */

    if( ta_status.vertex_count == 3 ) {

	tile_entry |= 0x80000000;

	for( y=polygon_bound.y1; y<=polygon_bound.y2; y++ ) {

	    for( x=polygon_bound.x1; x<=polygon_bound.x2; x++ ) {

		ta_write_tile_entry( x,y,tile_entry );

	    }

	}

	ta_status.last_triangle_bounds.x1 = polygon_bound.x1;

	ta_status.last_triangle_bounds.y1 = polygon_bound.y1;

	ta_status.last_triangle_bounds.x2 = polygon_bound.x2;

	ta_status.last_triangle_bounds.y2 = polygon_bound.y2;

    } else if( ta_status.current_vertex_type == TA_VERTEX_SPRITE ||

	       ta_status.current_vertex_type == TA_VERTEX_TEX_SPRITE ) {

	tile_entry |= 0xA0000000;

	for( y=polygon_bound.y1; y<=polygon_bound.y2; y++ ) {

	    for( x=polygon_bound.x1; x<=polygon_bound.x2; x++ ) {

		ta_write_tile_entry( x,y,tile_entry );

	    }

	}

	ta_status.last_triangle_bounds.x1 = polygon_bound.x1;

	ta_status.last_triangle_bounds.y1 = polygon_bound.y1;

	ta_status.last_triangle_bounds.x2 = polygon_bound.x2;

	ta_status.last_triangle_bounds.y2 = polygon_bound.y2;

    } else {

	for( y=polygon_bound.y1; y<=polygon_bound.y2; y++ ) {

	    for( x=polygon_bound.x1; x<=polygon_bound.x2; x++ ) {

		uint32_t entry = tile_entry;

		for( i=0; i<ta_status.vertex_count-2; i++ ) {

		    if( triangle_bound[i].x1 <= x && triangle_bound[i].x2 >= x &&

			triangle_bound[i].y1 <= y && triangle_bound[i].y2 >= y ) {

			entry |= (0x40000000>>i);

		    }

		}

		ta_write_tile_entry( x, y, entry );

	    }

	}

	ta_status.last_triangle_bounds.x1 = -1;

    }

}

/**

 * Variant of ta_split_polygon called when vertex_count == max_vertex, but 

 * the client hasn't sent the LAST VERTEX flag. Commit the poly as normal

 * first, then start a new poly with the first 2 vertexes taken from the 

 * current one.

 */

static void ta_split_polygon() {

    ta_commit_polygon();

    if( TA_IS_NORMAL_POLY() ) { 

	/* This only applies to ordinary polys - Sprites + modifier lists are

	 * handled differently

	 */

	if( ta_status.vertex_count == 3 ) {

	    /* Triangles use an odd/even scheme */

	    if( ta_status.poly_parity == 0 ) {

		memcpy( &ta_status.poly_vertex[0], &ta_status.poly_vertex[2], 

			sizeof(struct pvr2_ta_vertex) );

		ta_status.poly_parity = 1;

	    } else {

		memcpy( &ta_status.poly_vertex[1], &ta_status.poly_vertex[2],

			sizeof(struct pvr2_ta_vertex) );

		ta_status.poly_parity = 0;

	    }

	} else {

	    /* Everything else just uses the last 2 vertexes in order */

	    memcpy( &ta_status.poly_vertex[0], &ta_status.poly_vertex[ta_status.vertex_count-2], 

		    sizeof(struct pvr2_ta_vertex)*2 );

	    ta_status.poly_parity = 0;

	}

	ta_status.vertex_count = 2;

    } else {

	ta_status.vertex_count = 0;

    }

}

/**

 * Parse the polygon context block and setup the internal state to receive

 * vertexes.

 * @param data 32 bytes of parameter data.

 */

static void ta_parse_polygon_context( union ta_data *data ) {

    int colourfmt = TA_POLYCMD_COLOURFMT(data[0].i);

    if( TA_POLYCMD_USELENGTH(data[0].i) ) {

	ta_status.max_vertex = TA_POLYCMD_LENGTH(data[0].i);

    }

    ta_status.clip_mode = TA_POLYCMD_CLIP(data[0].i);

    if( ta_status.clip_mode == 1 ) { /* Reserved - treat as CLIP_INSIDE */

	ta_status.clip_mode = TA_POLYCMD_CLIP_INSIDE;

    }

    ta_status.vertex_count = 0;

    ta_status.poly_context[0] = 

	(data[1].i & 0xFC1FFFFF) | ((data[0].i & 0x0B) << 22);

    ta_status.poly_context[1] = data[2].i;

    ta_status.poly_context[3] = data[4].i;

    ta_status.poly_parity = 0;

    if( data[0].i & TA_POLYCMD_TEXTURED ) {

	ta_status.current_vertex_type = data[0].i & 0x0D;

	ta_status.poly_context[2] = data[3].i;

	ta_status.poly_context[4] = data[5].i;

	if( data[0].i & TA_POLYCMD_SPECULAR ) {

	    ta_status.poly_context[0] |= 0x01000000;

	    ta_status.poly_vertex_size = 4;

	} else {

	    ta_status.poly_vertex_size = 3;

	}

	if( data[0].i & TA_POLYCMD_UV16 ) {

	    ta_status.poly_vertex_size--;

	}

    } else {

	ta_status.current_vertex_type = 0;

	ta_status.poly_vertex_size = 1;

	ta_status.poly_context[2] = 0;

	ta_status.poly_context[4] = 0;

    }

    ta_status.poly_pointer = (ta_status.poly_vertex_size << 21);

    ta_status.poly_context_size = 3;

    if( data[0].i & TA_POLYCMD_MODIFIED ) {

	ta_status.poly_pointer |= 0x01000000;

	if( data[0].i & TA_POLYCMD_FULLMOD ) {

	    ta_status.poly_context_size = 5;

	    ta_status.poly_vertex_size <<= 1;

	    ta_status.current_vertex_type |= 0x40;

	    /* Modified/float not supported - behaves as per last intensity */

	    if( colourfmt == TA_POLYCMD_COLOURFMT_FLOAT ) {

		colourfmt = TA_POLYCMD_COLOURFMT_LASTINT;

	    }

	}

    }

    if( colourfmt == TA_POLYCMD_COLOURFMT_INTENSITY ) {

	if( TA_POLYCMD_IS_FULLMOD(data[0].i) ||

	    TA_POLYCMD_IS_SPECULAR(data[0].i) ) {

	    ta_status.state = STATE_EXPECT_POLY_BLOCK2;

	} else {

	    ta_status.intensity1 = 

		parse_float_colour( data[4].f, data[5].f, data[6].f, data[7].f );

	}

    } else if( colourfmt == TA_POLYCMD_COLOURFMT_LASTINT ) {

	colourfmt = TA_POLYCMD_COLOURFMT_INTENSITY;

    }

    ta_status.current_vertex_type |= colourfmt;

}

/**

 * Parse the modifier volume context block and setup the internal state to 

 * receive modifier vertexes.

 * @param data 32 bytes of parameter data.

 */

static void ta_parse_modifier_context( union ta_data *data ) {

    ta_status.current_vertex_type = TA_VERTEX_MOD_VOLUME;

    ta_status.poly_vertex_size = 0;

    ta_status.clip_mode = TA_POLYCMD_CLIP(data[0].i);

    if( ta_status.clip_mode == 1 ) { /* Reserved - treat as CLIP_INSIDE */

	ta_status.clip_mode = TA_POLYCMD_CLIP_INSIDE;

    }

    ta_status.poly_context_size = 3;

    ta_status.poly_context[0] = (data[1].i & 0xFC1FFFFF) |

	((data[0].i & 0x0B)<<22);

    if( TA_POLYCMD_IS_SPECULAR(data[0].i) ) {

	ta_status.poly_context[0] |= 0x01000000;

    }

    ta_status.poly_context[1] = 0;

    ta_status.poly_context[2] = 0;

    ta_status.vertex_count = 0;

    ta_status.max_vertex = 3;

    ta_status.poly_pointer = 0;

}

/**

 * Parse the sprite context block and setup the internal state to receive

 * vertexes.

 * @param data 32 bytes of parameter data.

 */

static void ta_parse_sprite_context( union ta_data *data ) {

    ta_status.poly_context_size = 3;

    ta_status.poly_context[0] = (data[1].i & 0xFC1FFFFF) |

	((data[0].i & 0x0B)<<22) | 0x00400000;

    ta_status.clip_mode = TA_POLYCMD_CLIP(data[0].i);

    if( ta_status.clip_mode == 1 ) { /* Reserved - treat as CLIP_INSIDE */

	ta_status.clip_mode = TA_POLYCMD_CLIP_INSIDE;

    }

    if( TA_POLYCMD_IS_SPECULAR(data[0].i) ) {

	ta_status.poly_context[0] |= 0x01000000;

    }

    ta_status.poly_context[1] = data[2].i;

    ta_status.poly_context[2] = data[3].i;

    if( data[0].i & TA_POLYCMD_TEXTURED ) {

	ta_status.poly_vertex_size = 2;

	ta_status.poly_vertex[2].detail[1] = data[4].i;

	ta_status.current_vertex_type = TA_VERTEX_TEX_SPRITE;

    } else {

	ta_status.poly_vertex_size = 1;

	ta_status.poly_vertex[2].detail[0] = data[4].i;

	ta_status.current_vertex_type = TA_VERTEX_SPRITE;

    }

    ta_status.vertex_count = 0;

    ta_status.max_vertex = 4;

    ta_status.poly_pointer = (ta_status.poly_vertex_size << 21);

}

/**

 * Copy the last read vertex into all vertexes up to max_vertex. Used for

 * Aborted polygons under some circumstances.

 */

static void ta_fill_vertexes( ) {

    int i;

    for( i=ta_status.vertex_count; i<ta_status.max_vertex; i++ ) {

	memcpy( &ta_status.poly_vertex[i], &ta_status.poly_vertex[ta_status.vertex_count-1],

		sizeof( struct pvr2_ta_vertex ) );

    }

}

static void ta_parse_vertex( union ta_data *data ) {

    struct pvr2_ta_vertex *vertex = &ta_status.poly_vertex[ta_status.vertex_count];

    vertex->x = data[1].f;

    vertex->y = data[2].f;

    vertex->z = data[3].f;

    switch( ta_status.current_vertex_type ) {

    case TA_VERTEX_PACKED:

	vertex->detail[0] = data[6].i;

	break;

    case TA_VERTEX_FLOAT:

	vertex->detail[0] = parse_float_colour( data[4].f, data[5].f, data[6].f, data[7].f );

	break;

    case TA_VERTEX_INTENSITY:

	vertex->detail[0] = parse_intensity_colour( ta_status.intensity1, data[6].f );

	break;

    case TA_VERTEX_TEX_SPEC_PACKED:

	vertex->detail[3] = data[7].i; /* ARGB */

	/* Fallthrough */

    case TA_VERTEX_TEX_PACKED:

	vertex->detail[0] = data[4].i; /* U */

	vertex->detail[1] = data[5].i; /* V */

	vertex->detail[2] = data[6].i; /* ARGB */

	break;

    case TA_VERTEX_TEX_UV16_SPEC_PACKED:

	vertex->detail[2] = data[7].i; /* ARGB */

	/* Fallthrough */

    case TA_VERTEX_TEX_UV16_PACKED:

	vertex->detail[0] = data[4].i; /* UV */

	vertex->detail[1] = data[6].i; /* ARGB */

	break;

    case TA_VERTEX_TEX_FLOAT:

    case TA_VERTEX_TEX_SPEC_FLOAT:

	vertex->detail[0] = data[4].i; /* U */

	vertex->detail[1] = data[5].i; /* UV */

	ta_status.state = STATE_EXPECT_VERTEX_BLOCK2;

	break;

    case TA_VERTEX_TEX_UV16_FLOAT:

    case TA_VERTEX_TEX_UV16_SPEC_FLOAT:

	vertex->detail[0] = data[4].i; /* UV */

	ta_status.state = STATE_EXPECT_VERTEX_BLOCK2;

	break;

    case TA_VERTEX_TEX_SPEC_INTENSITY:

	vertex->detail[3] = parse_intensity_colour( ta_status.intensity2, data[7].f );

	/* Fallthrough */

    case TA_VERTEX_TEX_INTENSITY:

	vertex->detail[0] = data[4].i; /* U */

	vertex->detail[1] = data[5].i; /* V */

	vertex->detail[2] = parse_intensity_colour( ta_status.intensity1, data[6].f );

	break;

    case TA_VERTEX_TEX_UV16_SPEC_INTENSITY:

	vertex->detail[2] = parse_intensity_colour( ta_status.intensity2, data[7].f );

	/* Fallthrough */

    case TA_VERTEX_TEX_UV16_INTENSITY:

	vertex->detail[0] = data[4].i; /* UV */

	vertex->detail[1] = parse_intensity_colour( ta_status.intensity1, data[6].f );

	break;

    case TA_VERTEX_PACKED_MOD:

	vertex->detail[0] = data[4].i; /* ARGB */

	vertex->detail[1] = data[5].i; /* ARGB */

	break;

    case TA_VERTEX_INTENSITY_MOD:

	vertex->detail[0] = parse_intensity_colour( ta_status.intensity1, data[4].f );

	vertex->detail[1] = parse_intensity_colour( ta_status.intensity2, data[5].f );

	break;

    case TA_VERTEX_TEX_SPEC_PACKED_MOD:

	vertex->detail[3] = data[7].i; /* ARGB0 */

	/* Fallthrough */

    case TA_VERTEX_TEX_PACKED_MOD:

	vertex->detail[0] = data[4].i; /* U0 */

	vertex->detail[1] = data[5].i; /* V0 */

	vertex->detail[2] = data[6].i; /* ARGB0 */

	ta_status.state = STATE_EXPECT_VERTEX_BLOCK2;

	break;

    case TA_VERTEX_TEX_UV16_SPEC_PACKED_MOD:

	vertex->detail[2] = data[7].i; /* ARGB0 */

	/* Fallthrough */

    case TA_VERTEX_TEX_UV16_PACKED_MOD:

	vertex->detail[0] = data[4].i; /* UV0 */

	vertex->detail[1] = data[6].i; /* ARGB0 */

	ta_status.state = STATE_EXPECT_VERTEX_BLOCK2;

	break;

    case TA_VERTEX_TEX_SPEC_INTENSITY_MOD:

	vertex->detail[3] = parse_intensity_colour( ta_status.intensity1, data[7].f );

	/* Fallthrough */

    case TA_VERTEX_TEX_INTENSITY_MOD:

	vertex->detail[0] = data[4].i; /* U0 */

	vertex->detail[1] = data[5].i; /* V0 */

	vertex->detail[2] = parse_intensity_colour( ta_status.intensity1, data[6].f );

	ta_status.state = STATE_EXPECT_VERTEX_BLOCK2;

	break;

    case TA_VERTEX_TEX_UV16_SPEC_INTENSITY_MOD:

	vertex->detail[2] = parse_intensity_colour( ta_status.intensity1, data[7].f );

	/* Fallthrough */

    case TA_VERTEX_TEX_UV16_INTENSITY_MOD:

	vertex->detail[0] = data[4].i; /* UV0 */

	vertex->detail[1] = parse_intensity_colour( ta_status.intensity1, data[6].f );

	ta_status.state = STATE_EXPECT_VERTEX_BLOCK2;

	break;

    case TA_VERTEX_SPRITE:

    case TA_VERTEX_TEX_SPRITE:

    case TA_VERTEX_MOD_VOLUME:

    case TA_VERTEX_LISTLESS:

	vertex++;

	vertex->x = data[4].f;

	vertex->y = data[5].f;

	vertex->z = data[6].f;

	vertex++;

	vertex->x = data[7].f;

	ta_status.vertex_count += 2;

	ta_status.state = STATE_EXPECT_VERTEX_BLOCK2;

	break;

    }

    ta_status.vertex_count++;

}

static void ta_parse_vertex_block2( union ta_data *data ) {

    struct pvr2_ta_vertex *vertex = &ta_status.poly_vertex[ta_status.vertex_count-1];

    switch( ta_status.current_vertex_type ) {

    case TA_VERTEX_TEX_SPEC_FLOAT:

	vertex->detail[3] = parse_float_colour( data[4].f, data[5].f, data[6].f, data[7].f );

	/* Fallthrough */

    case TA_VERTEX_TEX_FLOAT:

	vertex->detail[2] = parse_float_colour( data[0].f, data[1].f, data[2].f, data[3].f );

	break;

    case TA_VERTEX_TEX_UV16_SPEC_FLOAT:

	vertex->detail[2] = parse_float_colour( data[4].f, data[5].f, data[6].f, data[7].f );

	/* Fallthrough */

    case TA_VERTEX_TEX_UV16_FLOAT:

	vertex->detail[1] = parse_float_colour( data[0].f, data[1].f, data[2].f, data[3].f );

	break;

    case TA_VERTEX_TEX_PACKED_MOD:

	vertex->detail[3] = data[0].i; /* U1 */

	vertex->detail[4] = data[1].i; /* V1 */

	vertex->detail[5] = data[2].i; /* ARGB1 */

	break;

    case TA_VERTEX_TEX_SPEC_PACKED_MOD:

	vertex->detail[4] = data[0].i; /* U1 */

	vertex->detail[5] = data[1].i; /* V1 */

	vertex->detail[6] = data[2].i; /* ARGB1 */

	vertex->detail[7] = data[3].i; /* ARGB1 */

	break;

    case TA_VERTEX_TEX_UV16_PACKED_MOD:

	vertex->detail[2] = data[0].i; /* UV1 */

	vertex->detail[3] = data[2].i; /* ARGB1 */

	break;

    case TA_VERTEX_TEX_UV16_SPEC_PACKED_MOD:

	vertex->detail[3] = data[0].i; /* UV1 */

	vertex->detail[4] = data[2].i; /* ARGB1 */

	vertex->detail[5] = data[3].i; /* ARGB1 */

	break;

    case TA_VERTEX_TEX_INTENSITY_MOD:

	vertex->detail[3] = data[0].i; /* U1 */

	vertex->detail[4] = data[1].i; /* V1 */

	vertex->detail[5] = parse_intensity_colour( ta_status.intensity2, data[2].f ); /* ARGB1 */

	break;

    case TA_VERTEX_TEX_SPEC_INTENSITY_MOD:

	vertex->detail[4] = data[0].i; /* U1 */