static void emit_load_R32G32( struct aos_compilation *cp,
struct x86_reg data,
struct x86_reg src_ptr )
{
sse_movups(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) );
sse_movlps(cp->func, data, src_ptr);
}
static void emit_load4f_2( struct x86_program *p,
struct x86_reg dest,
struct x86_reg arg0 )
{
/* Initialize from identity, then pull in low two words:
*/
sse_movups(&p->func, dest, get_identity(p));
sse_movlps(&p->func, dest, arg0);
}
static void emit_load4f_3( struct x86_program *p,
struct x86_reg dest,
struct x86_reg arg0 )
{
/* Have to jump through some hoops:
*
* c 0 0 0
* c 0 0 1
* 0 0 c 1
* a b c 1
*/
sse_movss(&p->func, dest, x86_make_disp(arg0, 8));
sse_shufps(&p->func, dest, get_identity(p), SHUF(X,Y,Z,W) );
sse_shufps(&p->func, dest, dest, SHUF(Y,Z,X,W) );
sse_movlps(&p->func, dest, arg0);
}
static void emit_store3f( struct x86_program *p,
struct x86_reg dest,
struct x86_reg arg0 )
{
if (p->outputs_safe) {
/* Emit the extra dword anyway. This may hurt writecombining,
* may cause other problems.
*/
sse_movups(&p->func, dest, arg0);
}
else {
/* Alternate strategy - emit two, shuffle, emit one.
*/
sse_movlps(&p->func, dest, arg0);
sse_shufps(&p->func, arg0, arg0, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */
sse_movss(&p->func, x86_make_disp(dest,8), arg0);
}
}
static void emit_load3f_3( struct x86_program *p,
struct x86_reg dest,
struct x86_reg arg0 )
{
/* Over-reads by 1 dword - potential SEGV if input is a vertex
* array.
*/
if (p->inputs_safe) {
sse_movups(&p->func, dest, arg0);
}
else {
/* c 0 0 0
* c c c c
* a b c c
*/
sse_movss(&p->func, dest, x86_make_disp(arg0, 8));
sse_shufps(&p->func, dest, dest, SHUF(X,X,X,X));
sse_movlps(&p->func, dest, arg0);
}
}
static void emit_load_R32G32B32( struct aos_compilation *cp,
struct x86_reg data,
struct x86_reg src_ptr )
{
#if 1
sse_movss(cp->func, data, x86_make_disp(src_ptr, 8));
/* data = z ? ? ? */
sse_shufps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) );
/* data = z ? 0 1 */
sse_shufps(cp->func, data, data, SHUF(Y,Z,X,W) );
/* data = ? 0 z 1 */
sse_movlps(cp->func, data, src_ptr);
/* data = x y z 1 */
#else
sse_movups(cp->func, data, src_ptr);
/* data = x y z ? */
sse2_pshufd(cp->func, data, data, SHUF(W,X,Y,Z) );
/* data = ? x y z */
sse_movss(cp->func, data, aos_get_internal_xmm( cp, IMM_ONES ) );
/* data = 1 x y z */
sse2_pshufd(cp->func, data, data, SHUF(Y,Z,W,X) );
/* data = x y z 1 */
#endif
}
static void emit_store2f( struct x86_program *p,
struct x86_reg dest,
struct x86_reg arg0 )
{
sse_movlps(&p->func, dest, arg0);
}
static void emit_load2f_2( struct x86_program *p,
struct x86_reg dest,
struct x86_reg arg0 )
{
sse_movlps(&p->func, dest, arg0);
}
