static void emit_fb_write( struct brw_wm_compile *c )
{
struct brw_fp_src payload_r0_depth = src_reg(BRW_FILE_PAYLOAD, PAYLOAD_DEPTH);
struct brw_fp_src outdepth = find_output_by_semantic(c, TGSI_SEMANTIC_POSITION, 0);
GLuint i;
outdepth = src_scalar(outdepth, Z);
for (i = 0 ; i < c->key.nr_cbufs; i++) {
struct brw_fp_src outcolor;
outcolor = find_output_by_semantic(c, TGSI_SEMANTIC_COLOR, i);
/* Use emit_tex_op so that we can specify the inst->target
* field, which is abused to contain the FB write target and the
* EOT marker
*/
emit_tex_op(c, WM_FB_WRITE,
dst_undef(),
(i == c->key.nr_cbufs - 1), /* EOT */
i,
0, /* no sampler */
outcolor,
payload_r0_depth,
outdepth);
}
}
static void emit_render_target_writes( struct brw_wm_compile *c )
{
struct prog_src_register payload_r0_depth = src_reg(PROGRAM_PAYLOAD, PAYLOAD_DEPTH);
struct prog_src_register outdepth = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DEPTH);
struct prog_src_register outcolor;
GLuint i;
struct prog_instruction *inst, *last_inst;
/* The inst->Aux field is used for FB write target and the EOT marker */
if (c->key.nr_color_regions > 1) {
for (i = 0 ; i < c->key.nr_color_regions; i++) {
outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DATA0 + i);
last_inst = inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(), 0),
0, outcolor, payload_r0_depth, outdepth);
inst->Aux = INST_AUX_TARGET(i);
if (c->fp_fragcolor_emitted) {
outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR);
last_inst = inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(), 0),
0, outcolor, payload_r0_depth, outdepth);
inst->Aux = INST_AUX_TARGET(i);
}
}
last_inst->Aux |= INST_AUX_EOT;
}
else {
/* if gl_FragData[0] is written, use it, else use gl_FragColor */
if (c->fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DATA0))
outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DATA0);
else
outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR);
inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(),0),
0, outcolor, payload_r0_depth, outdepth);
inst->Aux = INST_AUX_EOT | INST_AUX_TARGET(0);
}
}
static void emit_fb_write( struct brw_wm_compile *c )
{
struct prog_src_register outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLR);
struct prog_src_register payload_r0_depth = src_reg(PROGRAM_PAYLOAD, PAYLOAD_DEPTH);
struct prog_src_register outdepth = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DEPR);
emit_op(c,
WM_FB_WRITE,
dst_mask(dst_undef(),0),
0, 0, 0,
outcolor,
payload_r0_depth,
outdepth);
}
/**
* 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);
}
static void emit_insn( struct brw_wm_compile *c,
const struct tgsi_full_instruction *inst )
{
unsigned opcode = inst->Instruction.Opcode;
struct brw_fp_dst dst;
struct brw_fp_src src[3];
int i;
dst = translate_dst( c, &inst->Dst[0],
inst->Instruction.Saturate );
for (i = 0; i < inst->Instruction.NumSrcRegs; i++)
src[i] = translate_src( c, &inst->Src[i] );
switch (opcode) {
case TGSI_OPCODE_ABS:
emit_op1(c, TGSI_OPCODE_MOV,
dst,
src_abs(src[0]));
break;
case TGSI_OPCODE_SUB:
emit_op2(c, TGSI_OPCODE_ADD,
dst,
src[0],
src_negate(src[1]));
break;
case TGSI_OPCODE_SCS:
emit_op1(c, TGSI_OPCODE_SCS,
dst_mask(dst, BRW_WRITEMASK_XY),
src[0]);
break;
case TGSI_OPCODE_DST:
precalc_dst(c, dst, src[0], src[1]);
break;
case TGSI_OPCODE_LIT:
precalc_lit(c, dst, src[0]);
break;
case TGSI_OPCODE_TEX:
precalc_tex(c, dst,
inst->Texture.Texture,
src[1].index, /* use sampler unit for tex idx */
src[0], /* coord */
src[1]); /* sampler */
break;
case TGSI_OPCODE_TXP:
precalc_txp(c, dst,
inst->Texture.Texture,
src[1].index, /* use sampler unit for tex idx */
src[0], /* coord */
src[1]); /* sampler */
break;
case TGSI_OPCODE_TXB:
/* XXX: TXB not done
*/
precalc_tex(c, dst,
inst->Texture.Texture,
src[1].index, /* use sampler unit for tex idx*/
src[0],
src[1]);
break;
case TGSI_OPCODE_XPD:
emit_op2(c, TGSI_OPCODE_XPD,
dst_mask(dst, BRW_WRITEMASK_XYZ),
src[0],
src[1]);
break;
case TGSI_OPCODE_KIL:
emit_op1(c, TGSI_OPCODE_KIL,
dst_mask(dst_undef(), 0),
src[0]);
break;
case TGSI_OPCODE_END:
emit_fb_write(c);
break;
default:
if (!c->key.has_flow_control &&
brw_wm_is_scalar_result(opcode))
emit_scalar_insn(c, opcode, dst, src[0], src[1], src[2]);
else
emit_op3(c, opcode, dst, src[0], src[1], src[2]);
break;
}
}
