struct dynfn *tnl_makeX86Attr4ubv( TNLcontext *tnl, int key )
{
struct dynfn *dfn = MALLOC_STRUCT( dynfn );
insert_at_head( &tnl->dfn_cache.Color4ubv, dfn );
dfn->key = key;
if (TNL_DEBUG & DEBUG_CODEGEN)
_mesa_debug(NULL, "%s 0x%08x\n", __FUNCTION__, key );
if (key & TNL_CP_VC_FRMT_PKCOLOR) {
static char temp[] = {
0x8b, 0x44, 0x24, 0x04, /* mov 0x4(%esp,1),%eax */
0xba, 0x78, 0x56, 0x34, 0x12, /* mov $DEST,%edx */
0x8b, 0x00, /* mov (%eax),%eax */
0x89, 0x02, /* mov %eax,(%edx) */
0xc3, /* ret */
};
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 5, 0x12345678, (int)tnl->ubytecolorptr);
return dfn;
}
else {
static char temp[] = {
0x53, /* push %ebx */
0xba, 0x00, 0x00, 0x00, 0x00, /* mov $0x0,%edx */
0x31, 0xc0, /* xor %eax,%eax */
0x31, 0xc9, /* xor %ecx,%ecx */
0x8b, 0x5c, 0x24, 0x08, /* mov 0x8(%esp,1), %ebx */
0x8b, 0x1b, /* mov (%ebx), %ebx */
0x88, 0xd8, /* mov %bl, %al */
0x88, 0xf9, /* mov %bh, %cl */
0x8b, 0x04, 0x82, /* mov (%edx,%eax,4),%eax */
0x8b, 0x0c, 0x8a, /* mov (%edx,%ecx,4),%ecx */
0xa3, 0xaf, 0xbe, 0xad, 0xde, /* mov %eax,0xdeadbeaf */
0x89, 0x0d, 0xaf, 0xbe, 0xad, 0xde, /* mov %ecx,0xdeadbeaf */
0x31, 0xc0, /* xor %eax,%eax */
0x31, 0xc9, /* xor %ecx,%ecx */
0xc1, 0xeb, 0x10, /* shr $0x10, %ebx */
0x88, 0xd8, /* mov %bl, %al */
0x88, 0xf9, /* mov %bh, %cl */
0x8b, 0x04, 0x82, /* mov (%edx,%eax,4),%eax */
0x8b, 0x0c, 0x8a, /* mov (%edx,%ecx,4),%ecx */
0xa3, 0xaf, 0xbe, 0xad, 0xde, /* mov %eax,0xdeadbeaf */
0x89, 0x0d, 0xaf, 0xbe, 0xad, 0xde, /* mov %ecx,0xdeadbeaf */
0x5b, /* pop %ebx */
0xc3, /* ret */
};
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 2, 0x00000000, (int)_mesa_ubyte_to_float_color_tab);
FIXUP(dfn->code, 27, 0xdeadbeaf, (int)tnl->floatcolorptr);
FIXUP(dfn->code, 33, 0xdeadbeaf, (int)tnl->floatcolorptr+4);
FIXUP(dfn->code, 55, 0xdeadbeaf, (int)tnl->floatcolorptr+8);
FIXUP(dfn->code, 61, 0xdeadbeaf, (int)tnl->floatcolorptr+12);
return dfn;
}
}
struct dynfn *tnl_makeX86Attr3f( TNLcontext *tnl, int key )
{
static char temp[] = {
0xba, 0x78, 0x56, 0x34, 0x12, /* mov $DEST,%edx */
0x8b, 0x44, 0x24, 0x04, /* mov 0x4(%esp,1),%eax */
0x89, 0x02, /* mov %eax,(%edx) */
0x8b, 0x44, 0x24, 0x08, /* mov 0x8(%esp,1),%eax */
0x89, 0x42, 0x04, /* mov %eax,0x4(%edx) */
0x8b, 0x44, 0x24, 0x0c, /* mov 0xc(%esp,1),%eax */
0x89, 0x42, 0x08, /* mov %eax,0x8(%edx) */
0xc3, /* ret */
};
struct dynfn *dfn = MALLOC_STRUCT( dynfn );
if (TNL_DEBUG & DEBUG_CODEGEN)
_mesa_debug(NULL, "%s 0x%08x\n", __FUNCTION__, key );
insert_at_head( &tnl->dfn_cache.Normal3f, dfn );
dfn->key = key;
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 1, 0x12345678, (int)tnl->normalptr);
return dfn;
}
static void do_import( struct vertex_buffer *VB,
struct gl_client_array *to,
struct gl_client_array *from )
{
GLuint count = VB->Count;
if (!to->Ptr) {
to->Ptr = ALIGN_MALLOC( VB->Size * 4 * sizeof(GLubyte), 32 );
to->Type = GL_UNSIGNED_BYTE;
}
/* No need to transform the same value 3000 times.
*/
if (!from->StrideB) {
to->StrideB = 0;
count = 1;
}
else
to->StrideB = 4 * sizeof(GLubyte);
_math_trans_4ub( (GLubyte (*)[4]) to->Ptr,
from->Ptr,
from->StrideB,
from->Type,
from->Size,
0,
count);
}
static struct dynfn *makeSSENormal3fv( GLcontext *ctx, const int *key )
{
/* Requires P4 (sse2?)
*/
static unsigned char temp[] = {
0x8b, 0x44, 0x24, 0x04, /* mov 0x4(%esp,1),%eax */
0xba, 0x78, 0x56, 0x34, 0x12, /* mov $0x12345678,%edx */
0xf3, 0x0f, 0x7e, 0x00, /* movq (%eax),%xmm0 */
0x66, 0x0f, 0x6e, 0x48, 0x08, /* movd 0x8(%eax),%xmm1 */
0x66, 0x0f, 0xd6, 0x42, 0x0c, /* movq %xmm0,0xc(%edx) */
0x66, 0x0f, 0x7e, 0x4a, 0x14, /* movd %xmm1,0x14(%edx) */
0xc3, /* ret */
};
struct dynfn *dfn = MALLOC_STRUCT( dynfn );
r200ContextPtr rmesa = R200_CONTEXT(ctx);
insert_at_head( &rmesa->vb.dfn_cache.Normal3fv, dfn );
dfn->key[0] = key[0];
dfn->key[1] = key[1];
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 5, 0x0, (int)vb.normalptr);
return dfn;
}
/**
* Called the first time stage->run is called. In effect, don't
* allocate data until the first time the stage is run.
*/
static GLboolean init_vp( GLcontext *ctx,
struct tnl_pipeline_stage *stage )
{
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vertex_buffer *VB = &(tnl->vb);
struct vp_stage_data *store;
const GLuint size = VB->Size;
GLuint i;
stage->privatePtr = MALLOC(sizeof(*store));
store = VP_STAGE_DATA(stage);
if (!store)
return GL_FALSE;
/* Allocate arrays of vertex output values */
for (i = 0; i < 15; i++) {
_mesa_vector4f_alloc( &store->attribs[i], 0, size, 32 );
store->attribs[i].size = 4;
}
/* a few other misc allocations */
_mesa_vector4f_alloc( &store->ndcCoords, 0, size, 32 );
store->clipmask = (GLubyte *) ALIGN_MALLOC(sizeof(GLubyte)*size, 32 );
return GL_TRUE;
}
struct dynfn *tnl_makeX86Vertex2f( TNLcontext *tnl, int key )
{
struct dynfn *dfn = MALLOC_STRUCT( dynfn );
if (RADEON_DEBUG & DEBUG_CODEGEN)
_mesa_debug(NULL, "%s 0x%08x\n", __FUNCTION__, key );
switch (tnl->vertex_size) {
default: {
/* Repz convenient as it's possible to emit code for any size
* vertex with little tweaking. Might as well read vertsize
* though, and have only one of these.
*/
static char temp[] = {
0x57, /* push %edi */
0x56, /* push %esi */
0xbe, 0, 0, 0, 0, /* mov $VERTEX+2,%esi */
0x8b, 0x3d, 0, 0, 0, 0, /* mov DMAPTR,%edi */
0x8b, 0x44, 0x24, 0x0c, /* mov 0x0c(%esp,1),%eax */
0x8b, 0x54, 0x24, 0x10, /* mov 0x10(%esp,1),%edx */
0x89, 0x07, /* mov %eax,(%edi) */
0x89, 0x57, 0x04, /* mov %edx,0x4(%edi) */
0x83, 0xc7, 0x08, /* add $0x8,%edi */
0xb9, 0, 0, 0, 0, /* mov $VERTSIZE-2,%ecx */
0xf3, 0xa5, /* repz movsl %ds:(%esi),%es:(%edi)*/
0xa1, 0, 0, 0, 0, /* mov COUNTER,%eax */
0x89, 0x3d, 0, 0, 0, 0, /* mov %edi,DMAPTR */
0x48, /* dec %eax */
0xa3, 0, 0, 0, 0, /* mov %eax,COUNTER */
0x5e, /* pop %esi */
0x5f, /* pop %edi */
0x74, 0x01, /* je +1 */
0xc3, /* ret */
0xff, 0x25, 0, 0, 0, 0 /* jmp NOTIFY */
};
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 3, 0x0, (int)&tnl->vertex[2]);
FIXUP(dfn->code, 9, 0x0, (int)&tnl->dmaptr);
FIXUP(dfn->code, 37, 0x0, tnl->vertex_size-2);
FIXUP(dfn->code, 44, 0x0, (int)&tnl->counter);
FIXUP(dfn->code, 50, 0x0, (int)&tnl->dmaptr);
FIXUP(dfn->code, 56, 0x0, (int)&tnl->counter);
FIXUP(dfn->code, 67, 0x0, (int)&tnl->notify);
break;
}
}
insert_at_head( &tnl->dfn_cache.Vertex3f, dfn );
dfn->key = key;
return dfn;
}
void fxAllocVB( GLcontext *ctx )
{
fxMesaContext fxMesa = FX_CONTEXT(ctx);
GLuint size = TNL_CONTEXT(ctx)->vb.Size;
static int firsttime = 1;
if (firsttime) {
init_setup_tab();
firsttime = 0;
}
fxMesa->verts = (GrVertex *)ALIGN_MALLOC(size * sizeof(GrVertex), 32);
fxMesa->SetupIndex = SETUP_XYZW|SETUP_RGBA;
}
void tdfxInitVB( GLcontext *ctx )
{
tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
GLuint size = TNL_CONTEXT(ctx)->vb.Size;
static int firsttime = 1;
if (firsttime) {
init_setup_tab();
firsttime = 0;
}
fxMesa->verts = ALIGN_MALLOC(size * sizeof(tdfxVertex), 32);
fxMesa->vertexFormat = TDFX_LAYOUT_TINY;
fxMesa->SetupIndex = TDFX_XYZ_BIT|TDFX_RGBA_BIT;
}
void ffbInitVB( GLcontext *ctx )
{
ffbContextPtr fmesa = FFB_CONTEXT(ctx);
GLuint size = TNL_CONTEXT(ctx)->vb.Size;
fmesa->verts = (ffb_vertex *)ALIGN_MALLOC(size * sizeof(ffb_vertex), 32);
{
static int firsttime = 1;
if (firsttime) {
init_setup_tab();
firsttime = 0;
}
}
}
void mach64InitVB( GLcontext *ctx )
{
mach64ContextPtr mmesa = MACH64_CONTEXT(ctx);
GLuint size = TNL_CONTEXT(ctx)->vb.Size;
mmesa->verts = (GLubyte *)ALIGN_MALLOC(size * 4 * 16, 32);
{
static int firsttime = 1;
if (firsttime) {
init_setup_tab();
firsttime = 0;
}
}
}
void i830InitVB( GLcontext *ctx )
{
i830ContextPtr imesa = I830_CONTEXT(ctx);
GLuint size = TNL_CONTEXT(ctx)->vb.Size;
imesa->verts = (char *)ALIGN_MALLOC(size * 4 * 16, 32);
{
static int firsttime = 1;
if (firsttime) {
init_setup_tab();
firsttime = 0;
}
}
}
void gammaInitVB( GLcontext *ctx )
{
gammaContextPtr gmesa = GAMMA_CONTEXT(ctx);
GLuint size = TNL_CONTEXT(ctx)->vb.Size;
gmesa->verts = (GLubyte *)ALIGN_MALLOC(size * 4 * 16, 32);
{
static int firsttime = 1;
if (firsttime) {
init_setup_tab();
firsttime = 0;
gmesa->vertex_size = 16; /* FIXME - only one vertex setup */
}
}
}
void s3vInitVB( GLcontext *ctx )
{
s3vContextPtr vmesa = S3V_CONTEXT(ctx);
GLuint size = TNL_CONTEXT(ctx)->vb.Size;
vmesa->verts = (char *)ALIGN_MALLOC(size * 4 * 16, 32);
{
static int firsttime = 1;
if (firsttime) {
init_setup_tab();
firsttime = 0;
vmesa->vertex_stride_shift = 6 /* 4 */; /* FIXME - only one vertex setup */
}
}
}
static struct immediate *real_alloc_immediate( GLcontext *ctx )
{
struct immediate *IM = ALIGN_MALLOC_STRUCT( immediate, 32 );
GLuint j;
if (!IM)
return 0;
/* memset(IM, 0, sizeof(*IM)); */
IM->id = id++;
IM->ref_count = 0;
IM->backref = ctx;
IM->FlushElt = 0;
IM->LastPrimitive = IMM_MAX_COPIED_VERTS;
IM->Count = IMM_MAX_COPIED_VERTS;
IM->Start = IMM_MAX_COPIED_VERTS;
IM->Material = 0;
IM->MaterialMask = 0;
IM->MaxTextureUnits = ctx->Const.MaxTextureUnits;
IM->TexSize = 0;
IM->NormalLengthPtr = 0;
IM->CopyTexSize = 0;
IM->CopyStart = IM->Start;
/* TexCoord0 is special.
*/
IM->TexCoord[0] = IM->TexCoord0;
for (j = 1; j < ctx->Const.MaxTextureUnits; j++) {
IM->TexCoord[j] = (GLfloat (*)[4])
ALIGN_MALLOC( IMM_SIZE * sizeof(GLfloat) * 4, 32 );
}
/* KW: Removed initialization of normals as these are now treated
* identically to all other data types.
*/
MEMSET(IM->Flag, 0, sizeof(IM->Flag));
MEMSET(IM->Normal, 0.0 , sizeof(IM->Normal));
return IM;
}
void vbo_exec_vtx_init( struct vbo_exec_context *exec )
{
GLcontext *ctx = exec->ctx;
struct vbo_context *vbo = vbo_context(ctx);
GLuint i;
/* Allocate a buffer object. Will just reuse this object
* continuously, unless vbo_use_buffer_objects() is called to enable
* use of real VBOs.
*/
_mesa_reference_buffer_object(ctx,
&exec->vtx.bufferobj,
ctx->Shared->NullBufferObj);
ASSERT(!exec->vtx.buffer_map);
exec->vtx.buffer_map = (GLfloat *)ALIGN_MALLOC(VBO_VERT_BUFFER_SIZE, 64);
exec->vtx.buffer_ptr = exec->vtx.buffer_map;
vbo_exec_vtxfmt_init( exec );
/* Hook our functions into the dispatch table.
*/
_mesa_install_exec_vtxfmt( exec->ctx, &exec->vtxfmt );
for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) {
ASSERT(i < Elements(exec->vtx.attrsz));
exec->vtx.attrsz[i] = 0;
ASSERT(i < Elements(exec->vtx.active_sz));
exec->vtx.active_sz[i] = 0;
}
for (i = 0 ; i < VERT_ATTRIB_MAX; i++) {
ASSERT(i < Elements(exec->vtx.inputs));
ASSERT(i < Elements(exec->vtx.arrays));
exec->vtx.inputs[i] = &exec->vtx.arrays[i];
}
{
struct gl_client_array *arrays = exec->vtx.arrays;
memcpy(arrays, vbo->legacy_currval, 16 * sizeof(arrays[0]));
memcpy(arrays + 16, vbo->generic_currval, 16 * sizeof(arrays[0]));
}
exec->vtx.vertex_size = 0;
}
/* Install the codegen'ed choosers.
* We should keep a list and free them in the end...
*/
void _tnl_x86choosers( tnl_attrfv_func (*choose)[4],
tnl_attrfv_func (*do_choose)( GLuint attr,
GLuint sz ))
{
int attr, size;
for (attr = 0; attr < _TNL_MAX_ATTR_CODEGEN; attr++) {
for (size = 0; size < 4; size++) {
char *code;
char *start = (char *)&_tnl_x86_choose_fv;
char *end = (char *)&_tnl_x86_choose_fv_end;
int offset = 0;
code = ALIGN_MALLOC( end - start, 16 );
memcpy (code, start, end - start);
FIXUP(code, 0, 0, attr);
FIXUP(code, 0, 1, size + 1);
FIXUPREL(code, 0, 2, do_choose);
choose[attr][size] = (tnl_attrfv_func)code;
}
}
}
void mgaInitVB( GLcontext *ctx )
{
mgaContextPtr mmesa = MGA_CONTEXT(ctx);
GLuint size = TNL_CONTEXT(ctx)->vb.Size;
mmesa->verts = (char *)ALIGN_MALLOC(size * sizeof(mgaVertex), 32);
{
static int firsttime = 1;
if (firsttime) {
init_setup_tab();
firsttime = 0;
}
}
mmesa->new_state |= MGA_NEW_WARP;
mmesa->dirty |= MGA_UPLOAD_PIPE;
mmesa->vertex_format = setup_tab[0].vertex_format;
mmesa->vertex_size = setup_tab[0].vertex_size;
mmesa->vertex_stride_shift = setup_tab[0].vertex_stride_shift;
}
struct dynfn *tnl_makeX86Attr4ub( TNLcontext *tnl, int key )
{
if (TNL_DEBUG & DEBUG_CODEGEN)
_mesa_debug(NULL, "%s 0x%08x\n", __FUNCTION__, key );
if (key & TNL_CP_VC_FRMT_PKCOLOR) {
/* XXX push/pop */
static char temp[] = {
0x53, /* push %ebx */
0x8b, 0x44, 0x24, 0x08, /* mov 0x8(%esp,1),%eax */
0x8b, 0x54, 0x24, 0x0c, /* mov 0xc(%esp,1),%edx */
0x8b, 0x4c, 0x24, 0x10, /* mov 0x10(%esp,1),%ecx */
0x8b, 0x5c, 0x24, 0x14, /* mov 0x14(%esp,1),%ebx */
0xa2, 0, 0, 0, 0, /* mov %al,DEST */
0x88, 0x15, 0, 0, 0, 0, /* mov %dl,DEST+1 */
0x88, 0x0d, 0, 0, 0, 0, /* mov %cl,DEST+2 */
0x88, 0x1d, 0, 0, 0, 0, /* mov %bl,DEST+3 */
0x5b, /* pop %ebx */
0xc3, /* ret */
};
struct dynfn *dfn = MALLOC_STRUCT( dynfn );
insert_at_head( &tnl->dfn_cache.Color4ub, dfn );
dfn->key = key;
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 18, 0x0, (int)tnl->ubytecolorptr);
FIXUP(dfn->code, 24, 0x0, (int)tnl->ubytecolorptr+1);
FIXUP(dfn->code, 30, 0x0, (int)tnl->ubytecolorptr+2);
FIXUP(dfn->code, 36, 0x0, (int)tnl->ubytecolorptr+3);
return dfn;
}
else
return 0;
}
struct dynfn *tnl_makeX86Vertex3fv( TNLcontext *tnl, int key )
{
struct dynfn *dfn = MALLOC_STRUCT( dynfn );
if (TNL_DEBUG & DEBUG_CODEGEN)
_mesa_debug(NULL, "%s 0x%08x\n", __FUNCTION__, key );
switch (tnl->vertex_size) {
case 6: {
static char temp[] = {
0xa1, 0x00, 0x00, 0, 0, /* mov 0x0,%eax */
0x8b, 0x4c, 0x24, 0x04, /* mov 0x4(%esp,1),%ecx */
0x8b, 0x11, /* mov (%ecx),%edx */
0x89, 0x10, /* mov %edx,(%eax) */
0x8b, 0x51, 0x04, /* mov 0x4(%ecx),%edx */
0x8b, 0x49, 0x08, /* mov 0x8(%ecx),%ecx */
0x89, 0x50, 0x04, /* mov %edx,0x4(%eax) */
0x89, 0x48, 0x08, /* mov %ecx,0x8(%eax) */
0x8b, 0x15, 0x1c, 0, 0, 0, /* mov 0x1c,%edx */
0x8b, 0x0d, 0x20, 0, 0, 0, /* mov 0x20,%ecx */
0x89, 0x50, 0x0c, /* mov %edx,0xc(%eax) */
0x89, 0x48, 0x10, /* mov %ecx,0x10(%eax) */
0x8b, 0x15, 0x24, 0, 0, 0, /* mov 0x24,%edx */
0x89, 0x50, 0x14, /* mov %edx,0x14(%eax) */
0x83, 0xc0, 0x18, /* add $0x18,%eax */
0xa3, 0x00, 0x00, 0, 0, /* mov %eax,0x0 */
0xa1, 0x04, 0x00, 0, 0, /* mov 0x4,%eax */
0x48, /* dec %eax */
0xa3, 0x04, 0x00, 0, 0, /* mov %eax,0x4 */
0x74, 0x01, /* je 2a4 <.f11> */
0xc3, /* ret */
0xff, 0x25, 0x08, 0, 0, 0, /* jmp *0x8 */
};
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 1, 0x00000000, (int)&tnl->dmaptr);
FIXUP(dfn->code, 27, 0x0000001c, (int)&tnl->vertex[3]);
FIXUP(dfn->code, 33, 0x00000020, (int)&tnl->vertex[4]);
FIXUP(dfn->code, 45, 0x00000024, (int)&tnl->vertex[5]);
FIXUP(dfn->code, 56, 0x00000000, (int)&tnl->dmaptr);
FIXUP(dfn->code, 61, 0x00000004, (int)&tnl->counter);
FIXUP(dfn->code, 67, 0x00000004, (int)&tnl->counter);
FIXUP(dfn->code, 76, 0x00000008, (int)&tnl->notify);
break;
}
case 8: {
static char temp[] = {
0xa1, 0x00, 0x00, 0, 0, /* mov 0x0,%eax */
0x8b, 0x4c, 0x24, 0x04, /* mov 0x4(%esp,1),%ecx */
0x8b, 0x11, /* mov (%ecx),%edx */
0x89, 0x10, /* mov %edx,(%eax) */
0x8b, 0x51, 0x04, /* mov 0x4(%ecx),%edx */
0x8b, 0x49, 0x08, /* mov 0x8(%ecx),%ecx */
0x89, 0x50, 0x04, /* mov %edx,0x4(%eax) */
0x89, 0x48, 0x08, /* mov %ecx,0x8(%eax) */
0x8b, 0x15, 0x1c, 0, 0, 0, /* mov 0x1c,%edx */
0x8b, 0x0d, 0x20, 0, 0, 0, /* mov 0x20,%ecx */
0x89, 0x50, 0x0c, /* mov %edx,0xc(%eax) */
0x89, 0x48, 0x10, /* mov %ecx,0x10(%eax) */
0x8b, 0x15, 0x1c, 0, 0, 0, /* mov 0x1c,%edx */
0x8b, 0x0d, 0x20, 0, 0, 0, /* mov 0x20,%ecx */
0x89, 0x50, 0x14, /* mov %edx,0x14(%eax) */
0x89, 0x48, 0x18, /* mov %ecx,0x18(%eax) */
0x8b, 0x15, 0x24, 0, 0, 0, /* mov 0x24,%edx */
0x89, 0x50, 0x1c, /* mov %edx,0x1c(%eax) */
0x83, 0xc0, 0x20, /* add $0x20,%eax */
0xa3, 0x00, 0x00, 0, 0, /* mov %eax,0x0 */
0xa1, 0x04, 0x00, 0, 0, /* mov 0x4,%eax */
0x48, /* dec %eax */
0xa3, 0x04, 0x00, 0, 0, /* mov %eax,0x4 */
0x74, 0x01, /* je 2a4 <.f11> */
0xc3, /* ret */
0xff, 0x25, 0x08, 0, 0, 0, /* jmp *0x8 */
};
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 1, 0x00000000, (int)&tnl->dmaptr);
FIXUP(dfn->code, 27, 0x0000001c, (int)&tnl->vertex[3]);
FIXUP(dfn->code, 33, 0x00000020, (int)&tnl->vertex[4]);
FIXUP(dfn->code, 45, 0x0000001c, (int)&tnl->vertex[5]);
FIXUP(dfn->code, 51, 0x00000020, (int)&tnl->vertex[6]);
FIXUP(dfn->code, 63, 0x00000024, (int)&tnl->vertex[7]);
FIXUP(dfn->code, 74, 0x00000000, (int)&tnl->dmaptr);
FIXUP(dfn->code, 79, 0x00000004, (int)&tnl->counter);
FIXUP(dfn->code, 85, 0x00000004, (int)&tnl->counter);
FIXUP(dfn->code, 94, 0x00000008, (int)&tnl->notify);
break;
}
default: {
/* Repz convenient as it's possible to emit code for any size
* vertex with little tweaking. Might as well read vertsize
* though, and have only one of these.
*/
static char temp[] = {
0x8b, 0x54, 0x24, 0x04, /* mov 0x4(%esp,1),%edx */
0x57, /* push %edi */
0x56, /* push %esi */
0x8b, 0x3d, 1,1,1,1, /* mov DMAPTR,%edi */
0x8b, 0x02, /* mov (%edx),%eax */
0x8b, 0x4a, 0x04, /* mov 0x4(%edx),%ecx */
0x8b, 0x72, 0x08, /* mov 0x8(%edx),%esi */
0x89, 0x07, /* mov %eax,(%edi) */
0x89, 0x4f, 0x04, /* mov %ecx,0x4(%edi) */
0x89, 0x77, 0x08, /* mov %esi,0x8(%edi) */
0x83, 0xc7, 0x0c, /* add $0xc,%edi */
0xb9, 0x06, 0x00, 0x00, 0x00, /* mov $VERTSIZE-3,%ecx */
0xbe, 0x58, 0x00, 0x00, 0x00, /* mov $VERTEX[3],%esi */
0xf3, 0xa5, /* repz movsl %ds:(%esi),%es:(%edi)*/
0x89, 0x3d, 1, 1, 1, 1, /* mov %edi,DMAPTR */
0xa1, 2, 2, 2, 2, /* mov COUNTER,%eax */
0x5e, /* pop %esi */
0x5f, /* pop %edi */
0x48, /* dec %eax */
0xa3, 2, 2, 2, 2, /* mov %eax,COUNTER */
0x74, 0x01, /* je +1 */
0xc3, /* ret */
0xff, 0x25, 0, 0, 0, 0 /* jmp NOTIFY */
};
dfn->code = ALIGN_MALLOC( sizeof(temp), 16 );
memcpy (dfn->code, temp, sizeof(temp));
FIXUP(dfn->code, 8, 0x01010101, (int)&tnl->dmaptr);
FIXUP(dfn->code, 32, 0x00000006, tnl->vertex_size-3);
FIXUP(dfn->code, 37, 0x00000058, (int)&tnl->vertex[3]);
FIXUP(dfn->code, 45, 0x01010101, (int)&tnl->dmaptr);
FIXUP(dfn->code, 50, 0x02020202, (int)&tnl->counter);
FIXUP(dfn->code, 58, 0x02020202, (int)&tnl->counter);
FIXUP(dfn->code, 67, 0x0, (int)&tnl->notify);
break;
}
}
insert_at_head( &tnl->dfn_cache.Vertex3fv, dfn );
dfn->key = key;
return dfn;
}
static int test_norm_function( normal_func func, int mtype,
int masked, long *cycles )
{
GLvector3f source[1], dest[1], dest2[1], ref[1], ref2[1];
GLmatrix mat[1];
GLfloat s[TEST_COUNT][5], d[TEST_COUNT][3], r[TEST_COUNT][3];
GLfloat d2[TEST_COUNT][3], r2[TEST_COUNT][3], length[TEST_COUNT];
GLfloat scale;
GLfloat *m;
GLubyte mask[TEST_COUNT];
int i, j;
#ifdef RUN_XFORM_BENCHMARK
int cycle_i; /* the counter for the benchmarks we run */
#endif
(void) cycles;
mat->m = (GLfloat *) ALIGN_MALLOC( 16 * sizeof(GLfloat), 16 );
mat->inv = m = mat->m;
m[0] = 63.0; m[4] = 43.0; m[ 8] = 29.0; m[12] = 43.0;
m[1] = 55.0; m[5] = 17.0; m[ 9] = 31.0; m[13] = 7.0;
m[2] = 44.0; m[6] = 9.0; m[10] = 7.0; m[14] = 3.0;
m[3] = 11.0; m[7] = 23.0; m[11] = 91.0; m[15] = 9.0;
scale = 1.0F + rnd () * norm_scale_types[mtype];
for ( i = 0 ; i < 4 ; i++ ) {
for ( j = 0 ; j < 4 ; j++ ) {
switch ( norm_templates[mtype][i * 4 + j] ) {
case NIL:
m[j * 4 + i] = 0.0;
break;
case ONE:
m[j * 4 + i] = 1.0;
break;
case NEG:
m[j * 4 + i] = -1.0;
break;
case VAR:
break;
default:
abort();
}
}
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
mask[i] = i % 2; /* mask every 2nd element */
d[i][0] = s[i][0] = d2[i][0] = 0.0;
d[i][1] = s[i][1] = d2[i][1] = 0.0;
d[i][2] = s[i][2] = d2[i][2] = 0.0;
for ( j = 0 ; j < 3 ; j++ )
s[i][j] = rnd();
length[i] = 1 / sqrt( s[i][0]*s[i][0] +
s[i][1]*s[i][1] +
s[i][2]*s[i][2] );
}
source->data = (GLfloat(*)[3])s;
source->start = (GLfloat *)s;
source->count = TEST_COUNT;
source->stride = sizeof(s[0]);
source->flags = 0;
dest->data = (GLfloat(*)[3])d;
dest->start = (GLfloat *)d;
dest->count = TEST_COUNT;
dest->stride = sizeof(float[3]);
dest->flags = 0;
dest2->data = (GLfloat(*)[3])d2;
dest2->start = (GLfloat *)d2;
dest2->count = TEST_COUNT;
dest2->stride = sizeof(float[3]);
dest2->flags = 0;
ref->data = (GLfloat(*)[3])r;
ref->start = (GLfloat *)r;
ref->count = TEST_COUNT;
ref->stride = sizeof(float[3]);
ref->flags = 0;
ref2->data = (GLfloat(*)[3])r2;
ref2->start = (GLfloat *)r2;
ref2->count = TEST_COUNT;
ref2->stride = sizeof(float[3]);
ref2->flags = 0;
if ( norm_normalize_types[mtype] == 0 ) {
ref_norm_transform_rescale( mat, scale, source, NULL, NULL, ref );
} else {
ref_norm_transform_normalize( mat, scale, source, NULL, NULL, ref );
ref_norm_transform_normalize( mat, scale, source, length, NULL, ref2 );
}
if ( mesa_profile ) {
if ( masked ) {
BEGIN_RACE( *cycles );
func( mat, scale, source, NULL, mask, dest );
END_RACE( *cycles );
func( mat, scale, source, length, mask, dest2 );
} else {
BEGIN_RACE( *cycles );
func( mat, scale, source, NULL, NULL, dest );
END_RACE( *cycles );
func( mat, scale, source, length, NULL, dest2 );
}
} else {
if ( masked ) {
func( mat, scale, source, NULL, mask, dest );
func( mat, scale, source, length, mask, dest2 );
} else {
func( mat, scale, source, NULL, NULL, dest );
func( mat, scale, source, length, NULL, dest2 );
}
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
if ( masked && !(mask[i] & 1) )
continue;
for ( j = 0 ; j < 3 ; j++ ) {
if ( significand_match( d[i][j], r[i][j] ) < REQUIRED_PRECISION ) {
printf( "-----------------------------\n" );
printf( "(i = %i, j = %i)\n", i, j );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][0], r[i][0], r[i][0]/d[i][0],
MAX_PRECISION - significand_match( d[i][0], r[i][0] ) );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][1], r[i][1], r[i][1]/d[i][1],
MAX_PRECISION - significand_match( d[i][1], r[i][1] ) );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][2], r[i][2], r[i][2]/d[i][2],
MAX_PRECISION - significand_match( d[i][2], r[i][2] ) );
return 0;
}
if ( norm_normalize_types[mtype] != 0 ) {
if ( significand_match( d2[i][j], r2[i][j] ) < REQUIRED_PRECISION ) {
printf( "------------------- precalculated length case ------\n" );
printf( "(i = %i, j = %i)\n", i, j );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][0], r2[i][0], r2[i][0]/d2[i][0],
MAX_PRECISION - significand_match( d2[i][0], r2[i][0] ) );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][1], r2[i][1], r2[i][1]/d2[i][1],
MAX_PRECISION - significand_match( d2[i][1], r2[i][1] ) );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][2], r2[i][2], r2[i][2]/d2[i][2],
MAX_PRECISION - significand_match( d2[i][2], r2[i][2] ) );
return 0;
}
}
}
}
ALIGN_FREE( mat->m );
return 1;
}
static int test_transform_function( transform_func func, int psize, int mtype,
int masked, long *cycles )
{
GLvector4f source[1], dest[1], ref[1];
GLmatrix mat[1];
GLfloat *m;
GLubyte mask[TEST_COUNT];
int i, j;
#ifdef RUN_XFORM_BENCHMARK
int cycle_i; /* the counter for the benchmarks we run */
#endif
(void) cycles;
if ( psize > 4 ) {
gl_problem( NULL, "test_transform_function called with psize > 4\n" );
return 0;
}
mat->m = (GLfloat *) ALIGN_MALLOC( 16 * sizeof(GLfloat), 16 );
mat->type = mtypes[mtype];
m = mat->m;
m[0] = 63.0; m[4] = 43.0; m[ 8] = 29.0; m[12] = 43.0;
m[1] = 55.0; m[5] = 17.0; m[ 9] = 31.0; m[13] = 7.0;
m[2] = 44.0; m[6] = 9.0; m[10] = 7.0; m[14] = 3.0;
m[3] = 11.0; m[7] = 23.0; m[11] = 91.0; m[15] = 9.0;
for ( i = 0 ; i < 4 ; i++ ) {
for ( j = 0 ; j < 4 ; j++ ) {
switch ( templates[mtype][i * 4 + j] ) {
case NIL:
m[j * 4 + i] = 0.0;
break;
case ONE:
m[j * 4 + i] = 1.0;
break;
case NEG:
m[j * 4 + i] = -1.0;
break;
case VAR:
break;
default:
abort();
}
}
}
for ( i = 0 ; i < TEST_COUNT ; i++) {
mask[i] = i % 2; /* mask every 2nd element */
d[i][0] = s[i][0] = 0.0;
d[i][1] = s[i][1] = 0.0;
d[i][2] = s[i][2] = 0.0;
d[i][3] = s[i][3] = 1.0;
for ( j = 0 ; j < psize ; j++ )
s[i][j] = rnd();
}
source->data = (GLfloat(*)[4])s;
source->start = (GLfloat *)s;
source->count = TEST_COUNT;
source->stride = sizeof(s[0]);
source->size = 4;
source->flags = 0;
dest->data = (GLfloat(*)[4])d;
dest->start = (GLfloat *)d;
dest->count = TEST_COUNT;
dest->stride = sizeof(float[4]);
dest->size = 0;
dest->flags = 0;
ref->data = (GLfloat(*)[4])r;
ref->start = (GLfloat *)r;
ref->count = TEST_COUNT;
ref->stride = sizeof(float[4]);
ref->size = 0;
ref->flags = 0;
ref_transform( ref, mat, source, NULL, 0 );
if ( mesa_profile ) {
if ( masked ) {
BEGIN_RACE( *cycles );
func( dest, mat->m, source, mask, 1 );
END_RACE( *cycles );
} else {
BEGIN_RACE( *cycles );
func( dest, mat->m, source, NULL, 0 );
END_RACE( *cycles );
}
}
else {
if ( masked ) {
func( dest, mat->m, source, mask, 1 );
} else {
func( dest, mat->m, source, NULL, 0 );
}
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
if ( masked && (mask[i] & 1) )
continue;
for ( j = 0 ; j < 4 ; j++ ) {
if ( significand_match( d[i][j], r[i][j] ) < REQUIRED_PRECISION ) {
printf( "-----------------------------\n" );
printf( "(i = %i, j = %i)\n", i, j );
printf( "%f \t %f \t [diff = %e - %i bit missed]\n",
d[i][0], r[i][0], r[i][0]-d[i][0],
MAX_PRECISION - significand_match( d[i][0], r[i][0] ) );
printf( "%f \t %f \t [diff = %e - %i bit missed]\n",
d[i][1], r[i][1], r[i][1]-d[i][1],
MAX_PRECISION - significand_match( d[i][1], r[i][1] ) );
printf( "%f \t %f \t [diff = %e - %i bit missed]\n",
d[i][2], r[i][2], r[i][2]-d[i][2],
MAX_PRECISION - significand_match( d[i][2], r[i][2] ) );
printf( "%f \t %f \t [diff = %e - %i bit missed]\n",
d[i][3], r[i][3], r[i][3]-d[i][3],
MAX_PRECISION - significand_match( d[i][3], r[i][3] ) );
return 0;
}
}
}
ALIGN_FREE( mat->m );
return 1;
}
static int test_transform_function( transform_func func, int psize,
int mtype, unsigned long *cycles )
{
GLvector4f source[1], dest[1], ref[1];
GLmatrix mat[1];
GLfloat *m;
int i, j;
#ifdef RUN_DEBUG_BENCHMARK
int cycle_i; /* the counter for the benchmarks we run */
#endif
(void) cycles;
if ( psize > 4 ) {
_mesa_problem( NULL, "test_transform_function called with psize > 4\n" );
return 0;
}
mat->m = (GLfloat *) ALIGN_MALLOC( 16 * sizeof(GLfloat), 16 );
mat->type = mtypes[mtype];
m = mat->m;
ASSERT( ((long)m & 15) == 0 );
init_matrix( m );
for ( i = 0 ; i < 4 ; i++ ) {
for ( j = 0 ; j < 4 ; j++ ) {
switch ( templates[mtype][i * 4 + j] ) {
case NIL:
m[j * 4 + i] = 0.0;
break;
case ONE:
m[j * 4 + i] = 1.0;
break;
case NEG:
m[j * 4 + i] = -1.0;
break;
case VAR:
break;
default:
ASSERT(0);
return 0;
}
}
}
for ( i = 0 ; i < TEST_COUNT ; i++) {
ASSIGN_4V( d[i], 0.0, 0.0, 0.0, 1.0 );
ASSIGN_4V( s[i], 0.0, 0.0, 0.0, 1.0 );
for ( j = 0 ; j < psize ; j++ )
s[i][j] = rnd();
}
source->data = (GLfloat(*)[4])s;
source->start = (GLfloat *)s;
source->count = TEST_COUNT;
source->stride = sizeof(s[0]);
source->size = 4;
source->flags = 0;
dest->data = (GLfloat(*)[4])d;
dest->start = (GLfloat *)d;
dest->count = TEST_COUNT;
dest->stride = sizeof(float[4]);
dest->size = 0;
dest->flags = 0;
ref->data = (GLfloat(*)[4])r;
ref->start = (GLfloat *)r;
ref->count = TEST_COUNT;
ref->stride = sizeof(float[4]);
ref->size = 0;
ref->flags = 0;
ref_transform( ref, mat, source );
if ( mesa_profile ) {
BEGIN_RACE( *cycles );
func( dest, mat->m, source );
END_RACE( *cycles );
}
else {
func( dest, mat->m, source );
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
for ( j = 0 ; j < 4 ; j++ ) {
if ( significand_match( d[i][j], r[i][j] ) < REQUIRED_PRECISION ) {
_mesa_printf("-----------------------------\n" );
_mesa_printf("(i = %i, j = %i)\n", i, j );
_mesa_printf("%f \t %f \t [diff = %e - %i bit missed]\n",
d[i][0], r[i][0], r[i][0]-d[i][0],
MAX_PRECISION - significand_match( d[i][0], r[i][0] ) );
_mesa_printf("%f \t %f \t [diff = %e - %i bit missed]\n",
d[i][1], r[i][1], r[i][1]-d[i][1],
MAX_PRECISION - significand_match( d[i][1], r[i][1] ) );
_mesa_printf("%f \t %f \t [diff = %e - %i bit missed]\n",
d[i][2], r[i][2], r[i][2]-d[i][2],
MAX_PRECISION - significand_match( d[i][2], r[i][2] ) );
_mesa_printf("%f \t %f \t [diff = %e - %i bit missed]\n",
d[i][3], r[i][3], r[i][3]-d[i][3],
MAX_PRECISION - significand_match( d[i][3], r[i][3] ) );
return 0;
}
}
}
ALIGN_FREE( mat->m );
return 1;
}
/* Create the device specific context.
*/
GLboolean mach64CreateContext( const __GLcontextModes *glVisual,
__DRIcontextPrivate *driContextPriv,
void *sharedContextPrivate )
{
GLcontext *ctx, *shareCtx;
__DRIscreenPrivate *driScreen = driContextPriv->driScreenPriv;
struct dd_function_table functions;
mach64ContextPtr mmesa;
mach64ScreenPtr mach64Screen;
int i, heap;
GLuint *c_textureSwapsPtr = NULL;
#if DO_DEBUG
MACH64_DEBUG = driParseDebugString(getenv("MACH64_DEBUG"), debug_control);
#endif
/* Allocate the mach64 context */
mmesa = (mach64ContextPtr) CALLOC( sizeof(*mmesa) );
if ( !mmesa )
return GL_FALSE;
/* Init default driver functions then plug in our Mach64-specific functions
* (the texture functions are especially important)
*/
_mesa_init_driver_functions( &functions );
mach64InitDriverFuncs( &functions );
mach64InitIoctlFuncs( &functions );
mach64InitTextureFuncs( &functions );
/* Allocate the Mesa context */
if (sharedContextPrivate)
shareCtx = ((mach64ContextPtr) sharedContextPrivate)->glCtx;
else
shareCtx = NULL;
mmesa->glCtx = _mesa_create_context(glVisual, shareCtx,
&functions, (void *)mmesa);
if (!mmesa->glCtx) {
FREE(mmesa);
return GL_FALSE;
}
driContextPriv->driverPrivate = mmesa;
ctx = mmesa->glCtx;
mmesa->driContext = driContextPriv;
mmesa->driScreen = driScreen;
mmesa->driDrawable = NULL;
mmesa->hHWContext = driContextPriv->hHWContext;
mmesa->driHwLock = &driScreen->pSAREA->lock;
mmesa->driFd = driScreen->fd;
mach64Screen = mmesa->mach64Screen = (mach64ScreenPtr)driScreen->private;
/* Parse configuration files */
driParseConfigFiles (&mmesa->optionCache, &mach64Screen->optionCache,
mach64Screen->driScreen->myNum, "mach64");
mmesa->sarea = (drm_mach64_sarea_t *)((char *)driScreen->pSAREA +
sizeof(drm_sarea_t));
mmesa->CurrentTexObj[0] = NULL;
mmesa->CurrentTexObj[1] = NULL;
(void) memset( mmesa->texture_heaps, 0, sizeof( mmesa->texture_heaps ) );
make_empty_list( &mmesa->swapped );
mmesa->firstTexHeap = mach64Screen->firstTexHeap;
mmesa->lastTexHeap = mach64Screen->firstTexHeap + mach64Screen->numTexHeaps;
for ( i = mmesa->firstTexHeap ; i < mmesa->lastTexHeap ; i++ ) {
mmesa->texture_heaps[i] = driCreateTextureHeap( i, mmesa,
mach64Screen->texSize[i],
6, /* align to 64-byte boundary, use 12 for page-size boundary */
MACH64_NR_TEX_REGIONS,
(drmTextureRegionPtr)mmesa->sarea->tex_list[i],
&mmesa->sarea->tex_age[i],
&mmesa->swapped,
sizeof( mach64TexObj ),
(destroy_texture_object_t *) mach64DestroyTexObj );
#if ENABLE_PERF_BOXES
c_textureSwapsPtr = & mmesa->c_textureSwaps;
#endif
driSetTextureSwapCounterLocation( mmesa->texture_heaps[i],
c_textureSwapsPtr );
}
mmesa->RenderIndex = -1; /* Impossible value */
mmesa->vert_buf = NULL;
mmesa->num_verts = 0;
mmesa->new_state = MACH64_NEW_ALL;
mmesa->dirty = MACH64_UPLOAD_ALL;
/* Set the maximum texture size small enough that we can
* guarentee that both texture units can bind a maximal texture
* and have them both in memory (on-card or AGP) at once.
* Test for 2 textures * bytes/texel * size * size. There's no
* need to account for mipmaps since we only upload one level.
*/
ctx->Const.MaxTextureUnits = 2;
ctx->Const.MaxTextureImageUnits = 2;
ctx->Const.MaxTextureCoordUnits = 2;
heap = mach64Screen->IsPCI ? MACH64_CARD_HEAP : MACH64_AGP_HEAP;
driCalculateMaxTextureLevels( & mmesa->texture_heaps[heap],
1,
& ctx->Const,
mach64Screen->cpp,
10, /* max 2D texture size is 1024x1024 */
0, /* 3D textures unsupported. */
0, /* cube textures unsupported. */
0, /* texture rectangles unsupported. */
1, /* mipmapping unsupported. */
GL_TRUE, /* need to have both textures in
either local or AGP memory */
0 );
#if ENABLE_PERF_BOXES
mmesa->boxes = ( getenv( "LIBGL_PERFORMANCE_BOXES" ) != NULL );
#endif
/* Allocate the vertex buffer
*/
mmesa->vert_buf = ALIGN_MALLOC(MACH64_BUFFER_SIZE, 32);
if ( !mmesa->vert_buf )
return GL_FALSE;
mmesa->vert_used = 0;
mmesa->vert_total = MACH64_BUFFER_SIZE;
/* Initialize the software rasterizer and helper modules.
*/
_swrast_CreateContext( ctx );
_vbo_CreateContext( ctx );
_tnl_CreateContext( ctx );
_swsetup_CreateContext( ctx );
/* Install the customized pipeline:
*/
/* _tnl_destroy_pipeline( ctx ); */
/* _tnl_install_pipeline( ctx, mach64_pipeline ); */
/* Configure swrast and T&L to match hardware characteristics:
*/
_swrast_allow_pixel_fog( ctx, GL_FALSE );
_swrast_allow_vertex_fog( ctx, GL_TRUE );
_tnl_allow_pixel_fog( ctx, GL_FALSE );
_tnl_allow_vertex_fog( ctx, GL_TRUE );
driInitExtensions( ctx, card_extensions, GL_TRUE );
mach64InitVB( ctx );
mach64InitTriFuncs( ctx );
mach64DDInitStateFuncs( ctx );
mach64DDInitSpanFuncs( ctx );
mach64DDInitState( mmesa );
mmesa->do_irqs = (mmesa->mach64Screen->irq && !getenv("MACH64_NO_IRQS"));
mmesa->vblank_flags = (mmesa->do_irqs)
? driGetDefaultVBlankFlags(&mmesa->optionCache) : VBLANK_FLAG_NO_IRQ;
driContextPriv->driverPrivate = (void *)mmesa;
if (driQueryOptionb(&mmesa->optionCache, "no_rast")) {
fprintf(stderr, "disabling 3D acceleration\n");
FALLBACK(mmesa, MACH64_FALLBACK_DISABLE, 1);
}
return GL_TRUE;
}
static int test_norm_function( normal_func func, int mtype, long *cycles )
{
GLvector4f source[1], dest[1], dest2[1], ref[1], ref2[1];
GLmatrix mat[1];
GLfloat s[TEST_COUNT][5], d[TEST_COUNT][4], r[TEST_COUNT][4];
GLfloat d2[TEST_COUNT][4], r2[TEST_COUNT][4], length[TEST_COUNT];
GLfloat scale;
GLfloat *m;
int i, j;
#ifdef RUN_DEBUG_BENCHMARK
int cycle_i; /* the counter for the benchmarks we run */
#endif
(void) cycles;
mat->m = (GLfloat *) ALIGN_MALLOC( 16 * sizeof(GLfloat), 16 );
mat->inv = m = mat->m;
init_matrix( m );
scale = 1.0F + rnd () * norm_scale_types[mtype];
for ( i = 0 ; i < 4 ; i++ ) {
for ( j = 0 ; j < 4 ; j++ ) {
switch ( norm_templates[mtype][i * 4 + j] ) {
case NIL:
m[j * 4 + i] = 0.0;
break;
case ONE:
m[j * 4 + i] = 1.0;
break;
case NEG:
m[j * 4 + i] = -1.0;
break;
case VAR:
break;
default:
_mesa_exit(1);
}
}
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
ASSIGN_3V( d[i], 0.0, 0.0, 0.0 );
ASSIGN_3V( s[i], 0.0, 0.0, 0.0 );
ASSIGN_3V( d2[i], 0.0, 0.0, 0.0 );
for ( j = 0 ; j < 3 ; j++ )
s[i][j] = rnd();
length[i] = 1 / SQRTF( LEN_SQUARED_3FV( s[i] ) );
}
source->data = (GLfloat(*)[4]) s;
source->start = (GLfloat *) s;
source->count = TEST_COUNT;
source->stride = sizeof(s[0]);
source->flags = 0;
dest->data = d;
dest->start = (GLfloat *) d;
dest->count = TEST_COUNT;
dest->stride = sizeof(float[4]);
dest->flags = 0;
dest2->data = d2;
dest2->start = (GLfloat *) d2;
dest2->count = TEST_COUNT;
dest2->stride = sizeof(float[4]);
dest2->flags = 0;
ref->data = r;
ref->start = (GLfloat *) r;
ref->count = TEST_COUNT;
ref->stride = sizeof(float[4]);
ref->flags = 0;
ref2->data = r2;
ref2->start = (GLfloat *) r2;
ref2->count = TEST_COUNT;
ref2->stride = sizeof(float[4]);
ref2->flags = 0;
if ( norm_normalize_types[mtype] == 0 ) {
ref_norm_transform_rescale( mat, scale, source, NULL, ref );
} else {
ref_norm_transform_normalize( mat, scale, source, NULL, ref );
ref_norm_transform_normalize( mat, scale, source, length, ref2 );
}
if ( mesa_profile ) {
BEGIN_RACE( *cycles );
func( mat, scale, source, NULL, dest );
END_RACE( *cycles );
func( mat, scale, source, length, dest2 );
} else {
func( mat, scale, source, NULL, dest );
func( mat, scale, source, length, dest2 );
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
for ( j = 0 ; j < 3 ; j++ ) {
if ( significand_match( d[i][j], r[i][j] ) < REQUIRED_PRECISION ) {
_mesa_printf( "-----------------------------\n" );
_mesa_printf( "(i = %i, j = %i)\n", i, j );
_mesa_printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][0], r[i][0], r[i][0]/d[i][0],
MAX_PRECISION - significand_match( d[i][0], r[i][0] ) );
_mesa_printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][1], r[i][1], r[i][1]/d[i][1],
MAX_PRECISION - significand_match( d[i][1], r[i][1] ) );
_mesa_printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][2], r[i][2], r[i][2]/d[i][2],
MAX_PRECISION - significand_match( d[i][2], r[i][2] ) );
return 0;
}
if ( norm_normalize_types[mtype] != 0 ) {
if ( significand_match( d2[i][j], r2[i][j] ) < REQUIRED_PRECISION ) {
_mesa_printf( "------------------- precalculated length case ------\n" );
_mesa_printf( "(i = %i, j = %i)\n", i, j );
_mesa_printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][0], r2[i][0], r2[i][0]/d2[i][0],
MAX_PRECISION - significand_match( d2[i][0], r2[i][0] ) );
_mesa_printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][1], r2[i][1], r2[i][1]/d2[i][1],
MAX_PRECISION - significand_match( d2[i][1], r2[i][1] ) );
_mesa_printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][2], r2[i][2], r2[i][2]/d2[i][2],
MAX_PRECISION - significand_match( d2[i][2], r2[i][2] ) );
return 0;
}
}
}
}
ALIGN_FREE( mat->m );
return 1;
}
