/***********************************************************************
* Expand various instructions here to simpler forms.
*/
static void precalc_dst( struct brw_wm_compile *c,
const struct prog_instruction *inst )
{
struct prog_src_register src0 = inst->SrcReg[0];
struct prog_src_register src1 = inst->SrcReg[1];
struct prog_dst_register dst = inst->DstReg;
if (dst.WriteMask & WRITEMASK_Y) {
/* dst.y = mul src0.y, src1.y
*/
emit_op(c,
OPCODE_MUL,
dst_mask(dst, WRITEMASK_Y),
inst->SaturateMode,
src0,
src1,
src_undef());
}
if (dst.WriteMask & WRITEMASK_XZ) {
struct prog_instruction *swz;
GLuint z = GET_SWZ(src0.Swizzle, Z);
/* dst.xz = swz src0.1zzz
*/
swz = emit_op(c,
OPCODE_SWZ,
dst_mask(dst, WRITEMASK_XZ),
inst->SaturateMode,
src_swizzle(src0, SWIZZLE_ONE, z, z, z),
src_undef(),
src_undef());
/* Avoid letting negation flag of src0 affect our 1 constant. */
swz->SrcReg[0].Negate &= ~NEGATE_X;
}
if (dst.WriteMask & WRITEMASK_W) {
/* dst.w = mov src1.w
*/
emit_op(c,
OPCODE_MOV,
dst_mask(dst, WRITEMASK_W),
inst->SaturateMode,
src1,
src_undef(),
src_undef());
}
}
/* Internally generated immediates: overkill...
*/
static struct brw_fp_src src_imm( struct brw_wm_compile *c,
const GLfloat *v,
unsigned nr)
{
unsigned i, j;
unsigned swizzle;
/* Could do a first pass where we examine all existing immediates
* without expanding.
*/
for (i = 0; i < c->nr_immediates; i++) {
if (match_or_expand_immediate( v,
nr,
c->immediate[i].v,
&c->immediate[i].nr,
&swizzle ))
goto out;
}
if (c->nr_immediates < Elements(c->immediate)) {
i = c->nr_immediates++;
if (match_or_expand_immediate( v,
nr,
c->immediate[i].v,
&c->immediate[i].nr,
&swizzle ))
goto out;
}
c->error = 1;
return src_undef();
out:
/* Make sure that all referenced elements are from this immediate.
* Has the effect of making size-one immediates into scalars.
*/
for (j = nr; j < 4; j++)
swizzle |= (swizzle & 0x3) << (j * 2);
return src_swizzle( src_reg( TGSI_FILE_IMMEDIATE, i ),
BRW_GET_SWZ(swizzle, X),
BRW_GET_SWZ(swizzle, Y),
BRW_GET_SWZ(swizzle, Z),
BRW_GET_SWZ(swizzle, W) );
}
static struct brw_fp_src src_scalar( struct brw_fp_src reg, int x )
{
return src_swizzle(reg, x, x, x, x);
}
/**
* Some TEX instructions require extra code, cube map coordinate
* normalization, or coordinate scaling for RECT textures, etc.
* This function emits those extra instructions and the TEX
* instruction itself.
*/
static void precalc_tex( struct brw_wm_compile *c,
struct brw_fp_dst dst,
unsigned target,
unsigned unit,
struct brw_fp_src src0,
struct brw_fp_src sampler )
{
struct brw_fp_src coord;
struct brw_fp_dst tmp = dst_undef();
assert(unit < BRW_MAX_TEX_UNIT);
/* Cubemap: find longest component of coord vector and normalize
* it.
*/
if (target == TGSI_TEXTURE_CUBE) {
struct brw_fp_src tmpsrc;
tmp = get_temp(c);
tmpsrc = src_reg_from_dst(tmp);
/* tmp = abs(src0) */
emit_op1(c,
TGSI_OPCODE_MOV,
tmp,
src_abs(src0));
/* tmp.X = MAX(tmp.X, tmp.Y) */
emit_op2(c, TGSI_OPCODE_MAX,
dst_mask(tmp, BRW_WRITEMASK_X),
src_scalar(tmpsrc, X),
src_scalar(tmpsrc, Y));
/* tmp.X = MAX(tmp.X, tmp.Z) */
emit_op2(c, TGSI_OPCODE_MAX,
dst_mask(tmp, BRW_WRITEMASK_X),
tmpsrc,
src_scalar(tmpsrc, Z));
/* tmp.X = 1 / tmp.X */
emit_op1(c, TGSI_OPCODE_RCP,
dst_mask(tmp, BRW_WRITEMASK_X),
tmpsrc);
/* tmp = src0 * tmp.xxxx */
emit_op2(c, TGSI_OPCODE_MUL,
tmp,
src0,
src_scalar(tmpsrc, X));
coord = tmpsrc;
}
else if (target == TGSI_TEXTURE_RECT ||
target == TGSI_TEXTURE_SHADOWRECT) {
/* XXX: need a mechanism for internally generated constants.
*/
coord = src0;
}
else {
coord = src0;
}
/* Need to emit YUV texture conversions by hand. Probably need to
* do this here - the alternative is in brw_wm_emit.c, but the
* conversion requires allocating a temporary variable which we
* don't have the facility to do that late in the compilation.
*/
if (c->key.yuvtex_mask & (1 << unit)) {
/* convert ycbcr to RGBA */
GLboolean swap_uv = c->key.yuvtex_swap_mask & (1<<unit);
struct brw_fp_dst tmp = get_temp(c);
struct brw_fp_src tmpsrc = src_reg_from_dst(tmp);
struct brw_fp_src C0 = src_imm4f( c, -.5, -.0625, -.5, 1.164 );
struct brw_fp_src C1 = src_imm4f( c, 1.596, -0.813, 2.018, -.391 );
/* tmp = TEX ...
*/
emit_tex_op(c,
TGSI_OPCODE_TEX,
dst_saturate(tmp, dst.saturate),
unit,
target,
sampler.index,
coord,
src_undef(),
src_undef());
/* tmp.xyz = ADD TMP, C0
*/
emit_op2(c, TGSI_OPCODE_ADD,
dst_mask(tmp, BRW_WRITEMASK_XYZ),
tmpsrc,
C0);
/* YUV.y = MUL YUV.y, C0.w
*/
emit_op2(c, TGSI_OPCODE_MUL,
dst_mask(tmp, BRW_WRITEMASK_Y),
tmpsrc,
src_scalar(C0, W));
/*
* if (UV swaped)
* RGB.xyz = MAD YUV.zzx, C1, YUV.y
* else
* RGB.xyz = MAD YUV.xxz, C1, YUV.y
*/
emit_op3(c, TGSI_OPCODE_MAD,
dst_mask(dst, BRW_WRITEMASK_XYZ),
( swap_uv ?
src_swizzle(tmpsrc, Z,Z,X,X) :
src_swizzle(tmpsrc, X,X,Z,Z)),
C1,
src_scalar(tmpsrc, Y));
/* RGB.y = MAD YUV.z, C1.w, RGB.y
*/
emit_op3(c,
TGSI_OPCODE_MAD,
dst_mask(dst, BRW_WRITEMASK_Y),
src_scalar(tmpsrc, Z),
src_scalar(C1, W),
src_scalar(src_reg_from_dst(dst), Y));
release_temp(c, tmp);
}
else {
/* ordinary RGBA tex instruction */
emit_tex_op(c,
TGSI_OPCODE_TEX,
dst,
unit,
target,
sampler.index,
coord,
src_undef(),
src_undef());
}
/* XXX: add GL_EXT_texture_swizzle support to gallium -- by
* generating shader variants in mesa state tracker.
*/
/* Release this temp if we ended up allocating it:
*/
if (!dst_is_undef(tmp))
release_temp(c, tmp);
}
/**
* Some TEX instructions require extra code, cube map coordinate
* normalization, or coordinate scaling for RECT textures, etc.
* This function emits those extra instructions and the TEX
* instruction itself.
*/
static void precalc_tex( struct brw_wm_compile *c,
const struct prog_instruction *inst )
{
struct prog_src_register coord;
struct prog_dst_register tmpcoord;
const GLuint unit = c->fp->program.Base.SamplerUnits[inst->TexSrcUnit];
assert(unit < BRW_MAX_TEX_UNIT);
if (inst->TexSrcTarget == TEXTURE_CUBE_INDEX) {
struct prog_instruction *out;
struct prog_dst_register tmp0 = get_temp(c);
struct prog_src_register tmp0src = src_reg_from_dst(tmp0);
struct prog_dst_register tmp1 = get_temp(c);
struct prog_src_register tmp1src = src_reg_from_dst(tmp1);
struct prog_src_register src0 = inst->SrcReg[0];
/* find longest component of coord vector and normalize it */
tmpcoord = get_temp(c);
coord = src_reg_from_dst(tmpcoord);
/* tmpcoord = src0 (i.e.: coord = src0) */
out = emit_op(c, OPCODE_MOV,
tmpcoord,
0,
src0,
src_undef(),
src_undef());
out->SrcReg[0].Negate = NEGATE_NONE;
out->SrcReg[0].Abs = 1;
/* tmp0 = MAX(coord.X, coord.Y) */
emit_op(c, OPCODE_MAX,
tmp0,
0,
src_swizzle1(coord, X),
src_swizzle1(coord, Y),
src_undef());
/* tmp1 = MAX(tmp0, coord.Z) */
emit_op(c, OPCODE_MAX,
tmp1,
0,
tmp0src,
src_swizzle1(coord, Z),
src_undef());
/* tmp0 = 1 / tmp1 */
emit_op(c, OPCODE_RCP,
dst_mask(tmp0, WRITEMASK_X),
0,
tmp1src,
src_undef(),
src_undef());
/* tmpCoord = src0 * tmp0 */
emit_op(c, OPCODE_MUL,
tmpcoord,
0,
src0,
src_swizzle1(tmp0src, SWIZZLE_X),
src_undef());
release_temp(c, tmp0);
release_temp(c, tmp1);
}
else if (inst->TexSrcTarget == TEXTURE_RECT_INDEX) {
struct prog_src_register scale =
search_or_add_param5( c,
STATE_INTERNAL,
STATE_TEXRECT_SCALE,
unit,
0,0 );
tmpcoord = get_temp(c);
/* coord.xy = MUL inst->SrcReg[0], { 1/width, 1/height }
*/
emit_op(c,
OPCODE_MUL,
tmpcoord,
0,
inst->SrcReg[0],
src_swizzle(scale,
SWIZZLE_X,
SWIZZLE_Y,
SWIZZLE_ONE,
SWIZZLE_ONE),
src_undef());
coord = src_reg_from_dst(tmpcoord);
}
else {
coord = inst->SrcReg[0];
}
/* Need to emit YUV texture conversions by hand. Probably need to
* do this here - the alternative is in brw_wm_emit.c, but the
* conversion requires allocating a temporary variable which we
* don't have the facility to do that late in the compilation.
*/
if (c->key.yuvtex_mask & (1 << unit)) {
/* convert ycbcr to RGBA */
GLboolean swap_uv = c->key.yuvtex_swap_mask & (1<<unit);
/*
CONST C0 = { -.5, -.0625, -.5, 1.164 }
CONST C1 = { 1.596, -0.813, 2.018, -.391 }
UYV = TEX ...
UYV.xyz = ADD UYV, C0
UYV.y = MUL UYV.y, C0.w
if (UV swaped)
RGB.xyz = MAD UYV.zzx, C1, UYV.y
else
RGB.xyz = MAD UYV.xxz, C1, UYV.y
RGB.y = MAD UYV.z, C1.w, RGB.y
*/
struct prog_dst_register dst = inst->DstReg;
struct prog_dst_register tmp = get_temp(c);
struct prog_src_register tmpsrc = src_reg_from_dst(tmp);
struct prog_src_register C0 = search_or_add_const4f( c, -.5, -.0625, -.5, 1.164 );
struct prog_src_register C1 = search_or_add_const4f( c, 1.596, -0.813, 2.018, -.391 );
/* tmp = TEX ...
*/
emit_tex_op(c,
OPCODE_TEX,
tmp,
inst->SaturateMode,
unit,
inst->TexSrcTarget,
inst->TexShadow,
coord,
src_undef(),
src_undef());
/* tmp.xyz = ADD TMP, C0
*/
emit_op(c,
OPCODE_ADD,
dst_mask(tmp, WRITEMASK_XYZ),
0,
tmpsrc,
C0,
src_undef());
/* YUV.y = MUL YUV.y, C0.w
*/
emit_op(c,
OPCODE_MUL,
dst_mask(tmp, WRITEMASK_Y),
0,
tmpsrc,
src_swizzle1(C0, W),
src_undef());
/*
* if (UV swaped)
* RGB.xyz = MAD YUV.zzx, C1, YUV.y
* else
* RGB.xyz = MAD YUV.xxz, C1, YUV.y
*/
emit_op(c,
OPCODE_MAD,
dst_mask(dst, WRITEMASK_XYZ),
0,
swap_uv?src_swizzle(tmpsrc, Z,Z,X,X):src_swizzle(tmpsrc, X,X,Z,Z),
C1,
src_swizzle1(tmpsrc, Y));
/* RGB.y = MAD YUV.z, C1.w, RGB.y
*/
emit_op(c,
OPCODE_MAD,
dst_mask(dst, WRITEMASK_Y),
0,
src_swizzle1(tmpsrc, Z),
src_swizzle1(C1, W),
src_swizzle1(src_reg_from_dst(dst), Y));
release_temp(c, tmp);
}
else {
/* ordinary RGBA tex instruction */
emit_tex_op(c,
OPCODE_TEX,
inst->DstReg,
inst->SaturateMode,
unit,
inst->TexSrcTarget,
inst->TexShadow,
coord,
src_undef(),
src_undef());
}
/* For GL_EXT_texture_swizzle: */
if (c->key.tex_swizzles[unit] != SWIZZLE_NOOP) {
/* swizzle the result of the TEX instruction */
struct prog_src_register tmpsrc = src_reg_from_dst(inst->DstReg);
emit_op(c, OPCODE_SWZ,
inst->DstReg,
SATURATE_OFF, /* saturate already done above */
src_swizzle4(tmpsrc, c->key.tex_swizzles[unit]),
src_undef(),
src_undef());
}
if ((inst->TexSrcTarget == TEXTURE_RECT_INDEX) ||
(inst->TexSrcTarget == TEXTURE_CUBE_INDEX))
release_temp(c, tmpcoord);
}
static void emit_interp( struct brw_wm_compile *c,
GLuint idx )
{
struct prog_dst_register dst = dst_reg(PROGRAM_INPUT, idx);
struct prog_src_register interp = src_reg(PROGRAM_PAYLOAD, idx);
struct prog_src_register deltas = get_delta_xy(c);
/* Need to use PINTERP on attributes which have been
* multiplied by 1/W in the SF program, and LINTERP on those
* which have not:
*/
switch (idx) {
case FRAG_ATTRIB_WPOS:
/* Have to treat wpos.xy specially:
*/
emit_op(c,
WM_WPOSXY,
dst_mask(dst, WRITEMASK_XY),
0,
get_pixel_xy(c),
src_undef(),
src_undef());
dst = dst_mask(dst, WRITEMASK_ZW);
/* PROGRAM_INPUT.attr.xyzw = INTERP payload.interp[attr].x, deltas.xyw
*/
emit_op(c,
WM_LINTERP,
dst,
0,
interp,
deltas,
src_undef());
break;
case FRAG_ATTRIB_COL0:
case FRAG_ATTRIB_COL1:
if (c->key.flat_shade) {
emit_op(c,
WM_CINTERP,
dst,
0,
interp,
src_undef(),
src_undef());
}
else {
if (c->key.linear_color) {
emit_op(c,
WM_LINTERP,
dst,
0,
interp,
deltas,
src_undef());
}
else {
/* perspective-corrected color interpolation */
emit_op(c,
WM_PINTERP,
dst,
0,
interp,
deltas,
get_pixel_w(c));
}
}
break;
case FRAG_ATTRIB_FOGC:
/* Interpolate the fog coordinate */
emit_op(c,
WM_PINTERP,
dst_mask(dst, WRITEMASK_X),
0,
interp,
deltas,
get_pixel_w(c));
emit_op(c,
OPCODE_MOV,
dst_mask(dst, WRITEMASK_YZW),
0,
src_swizzle(interp,
SWIZZLE_ZERO,
SWIZZLE_ZERO,
SWIZZLE_ZERO,
SWIZZLE_ONE),
src_undef(),
src_undef());
break;
case FRAG_ATTRIB_FACE:
emit_op(c,
WM_FRONTFACING,
dst_mask(dst, WRITEMASK_X),
0,
src_undef(),
src_undef(),
src_undef());
break;
case FRAG_ATTRIB_PNTC:
/* XXX review/test this case */
emit_op(c,
WM_PINTERP,
dst_mask(dst, WRITEMASK_XY),
0,
interp,
deltas,
get_pixel_w(c));
emit_op(c,
OPCODE_MOV,
dst_mask(dst, WRITEMASK_ZW),
0,
src_swizzle(interp,
SWIZZLE_ZERO,
SWIZZLE_ZERO,
SWIZZLE_ZERO,
SWIZZLE_ONE),
src_undef(),
src_undef());
break;
default:
emit_op(c,
WM_PINTERP,
dst,
0,
interp,
deltas,
get_pixel_w(c));
break;
}
c->fp_interp_emitted |= 1<<idx;
}
static struct prog_src_register src_swizzle1( struct prog_src_register reg, int x )
{
return src_swizzle(reg, x, x, x, x);
}
static void precalc_tex( struct brw_wm_compile *c,
const struct prog_instruction *inst )
{
struct prog_src_register coord;
struct prog_dst_register tmpcoord;
if (inst->TexSrcTarget == TEXTURE_RECT_INDEX) {
struct prog_src_register scale =
search_or_add_param5( c,
STATE_INTERNAL,
STATE_TEXRECT_SCALE,
inst->TexSrcUnit,
0,0 );
tmpcoord = get_temp(c);
/* coord.xy = MUL inst->SrcReg[0], { 1/width, 1/height }
*/
emit_op(c,
OPCODE_MUL,
tmpcoord,
0, 0, 0,
inst->SrcReg[0],
scale,
src_undef());
coord = src_reg_from_dst(tmpcoord);
}
else {
coord = inst->SrcReg[0];
}
/* Need to emit YUV texture conversions by hand. Probably need to
* do this here - the alternative is in brw_wm_emit.c, but the
* conversion requires allocating a temporary variable which we
* don't have the facility to do that late in the compilation.
*/
if (!(c->key.yuvtex_mask & (1<<inst->TexSrcUnit))) {
emit_op(c,
OPCODE_TEX,
inst->DstReg,
inst->SaturateMode,
inst->TexSrcUnit,
inst->TexSrcTarget,
coord,
src_undef(),
src_undef());
}
else {
/*
CONST C0 = { -.5, -.0625, -.5, 1.164 }
CONST C1 = { 1.596, -0.813, 2.018, -.391 }
UYV = TEX ...
UYV.xyz = ADD UYV, C0
UYV.y = MUL UYV.y, C0.w
RGB.xyz = MAD UYV.xxz, C1, UYV.y
RGB.y = MAD UYV.z, C1.w, RGB.y
*/
struct prog_dst_register dst = inst->DstReg;
struct prog_src_register src0 = inst->SrcReg[0];
struct prog_dst_register tmp = get_temp(c);
struct prog_src_register tmpsrc = src_reg_from_dst(tmp);
struct prog_src_register C0 = search_or_add_const4f( c, -.5, -.0625, -.5, 1.164 );
struct prog_src_register C1 = search_or_add_const4f( c, 1.596, -0.813, 2.018, -.391 );
/* tmp = TEX ...
*/
emit_op(c,
OPCODE_TEX,
tmp,
inst->SaturateMode,
inst->TexSrcUnit,
inst->TexSrcTarget,
src0,
src_undef(),
src_undef());
/* tmp.xyz = ADD TMP, C0
*/
emit_op(c,
OPCODE_ADD,
dst_mask(tmp, WRITEMASK_XYZ),
0, 0, 0,
tmpsrc,
C0,
src_undef());
/* YUV.y = MUL YUV.y, C0.w
*/
emit_op(c,
OPCODE_MUL,
dst_mask(tmp, WRITEMASK_Y),
0, 0, 0,
tmpsrc,
src_swizzle1(C0, W),
src_undef());
/* RGB.xyz = MAD YUV.xxz, C1, YUV.y
*/
emit_op(c,
OPCODE_MAD,
dst_mask(dst, WRITEMASK_XYZ),
0, 0, 0,
src_swizzle(tmpsrc, X,X,Z,Z),
C1,
src_swizzle1(tmpsrc, Y));
/* RGB.y = MAD YUV.z, C1.w, RGB.y
*/
emit_op(c,
OPCODE_MAD,
dst_mask(dst, WRITEMASK_Y),
0, 0, 0,
src_swizzle1(tmpsrc, Z),
src_swizzle1(C1, W),
src_swizzle1(src_reg_from_dst(dst), Y));
release_temp(c, tmp);
}
if (inst->TexSrcTarget == GL_TEXTURE_RECTANGLE_NV)
release_temp(c, tmpcoord);
}