}
}
-
-/**
- * Same as \sa get_src_reg() but if the register is a literal, emit
- * a brw_reg encoding the literal.
- * Note that a brw instruction only allows one src operand to be a literal.
- * For instructions with more than one operand, only the second can be a
- * literal. This means that we treat some literals as constants/uniforms
- * (which why PROGRAM_CONSTANT is checked in fetch_constants()).
- *
- */
-static struct brw_reg get_src_reg_imm(struct brw_wm_compile *c,
- const struct prog_instruction *inst,
- GLuint srcRegIndex, GLuint channel)
-{
- const struct prog_src_register *src = &inst->SrcReg[srcRegIndex];
- if (src->File == PROGRAM_CONSTANT) {
- /* a literal */
- const int component = GET_SWZ(src->Swizzle, channel);
- const GLfloat *param =
- c->fp->program.Base.Parameters->ParameterValues[src->Index];
- GLfloat value = param[component];
- if (src->Negate & (1 << channel))
- value = -value;
- if (src->Abs)
- value = FABSF(value);
-#if 0
- printf(" form immed value %f for chan %d\n", value, channel);
-#endif
- return brw_imm_f(value);
- }
- else {
- return get_src_reg(c, inst, srcRegIndex, channel);
- }
-}
-
-
/**
* Subroutines are minimal support for resusable instruction sequences.
* They are implemented as simply as possible to minimise overhead: there
release_tmps(c, mark);
}
-static void emit_pixel_xy(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_reg r1 = brw_vec1_grf(1, 0);
- struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW);
-
- struct brw_reg dst0, dst1;
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
-
- dst0 = get_dst_reg(c, inst, 0);
- dst1 = get_dst_reg(c, inst, 1);
- /* Calculate pixel centers by adding 1 or 0 to each of the
- * micro-tile coordinates passed in r1.
- */
- if (mask & WRITEMASK_X) {
- brw_ADD(p,
- vec8(retype(dst0, BRW_REGISTER_TYPE_UW)),
- stride(suboffset(r1_uw, 4), 2, 4, 0),
- brw_imm_v(0x10101010));
- }
-
- if (mask & WRITEMASK_Y) {
- brw_ADD(p,
- vec8(retype(dst1, BRW_REGISTER_TYPE_UW)),
- stride(suboffset(r1_uw, 5), 2, 4, 0),
- brw_imm_v(0x11001100));
- }
-}
-
-static void emit_delta_xy(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_reg r1 = brw_vec1_grf(1, 0);
- struct brw_reg dst0, dst1, src0, src1;
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
-
- dst0 = get_dst_reg(c, inst, 0);
- dst1 = get_dst_reg(c, inst, 1);
- src0 = get_src_reg(c, inst, 0, 0);
- src1 = get_src_reg(c, inst, 0, 1);
- /* Calc delta X,Y by subtracting origin in r1 from the pixel
- * centers.
- */
- if (mask & WRITEMASK_X) {
- brw_ADD(p,
- dst0,
- retype(src0, BRW_REGISTER_TYPE_UW),
- negate(r1));
- }
-
- if (mask & WRITEMASK_Y) {
- brw_ADD(p,
- dst1,
- retype(src1, BRW_REGISTER_TYPE_UW),
- negate(suboffset(r1,1)));
-
- }
-}
-
-static void fire_fb_write( struct brw_wm_compile *c,
- GLuint base_reg,
- GLuint nr,
- GLuint target,
- GLuint eot)
-{
- struct brw_compile *p = &c->func;
- /* Pass through control information:
- */
- /* mov (8) m1.0<1>:ud r1.0<8;8,1>:ud { Align1 NoMask } */
- {
- brw_push_insn_state(p);
- brw_set_mask_control(p, BRW_MASK_DISABLE); /* ? */
- brw_MOV(p,
- brw_message_reg(base_reg + 1),
- brw_vec8_grf(1, 0));
- brw_pop_insn_state(p);
- }
- /* Send framebuffer write message: */
- brw_fb_WRITE(p,
- retype(vec8(brw_null_reg()), BRW_REGISTER_TYPE_UW),
- base_reg,
- retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW),
- target,
- nr,
- 0,
- eot);
-}
-
-static void emit_fb_write(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- int nr = 2;
- int channel;
- GLuint target, eot;
- struct brw_reg src0;
-
- /* Reserve a space for AA - may not be needed:
- */
- if (c->key.aa_dest_stencil_reg)
- nr += 1;
-
- brw_push_insn_state(p);
- for (channel = 0; channel < 4; channel++) {
- src0 = get_src_reg(c, inst, 0, channel);
- /* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */
- /* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */
- brw_MOV(p, brw_message_reg(nr + channel), src0);
- }
- /* skip over the regs populated above: */
- nr += 8;
- brw_pop_insn_state(p);
-
- if (c->key.source_depth_to_render_target) {
- if (c->key.computes_depth) {
- src0 = get_src_reg(c, inst, 2, 2);
- brw_MOV(p, brw_message_reg(nr), src0);
- }
- else {
- src0 = get_src_reg(c, inst, 1, 1);
- brw_MOV(p, brw_message_reg(nr), src0);
- }
-
- nr += 2;
- }
-
- if (c->key.dest_depth_reg) {
- const GLuint comp = c->key.dest_depth_reg / 2;
- const GLuint off = c->key.dest_depth_reg % 2;
-
- if (off != 0) {
- /* XXX this code needs review/testing */
- struct brw_reg arg1_0 = get_src_reg(c, inst, 1, comp);
- struct brw_reg arg1_1 = get_src_reg(c, inst, 1, comp+1);
-
- brw_push_insn_state(p);
- brw_set_compression_control(p, BRW_COMPRESSION_NONE);
-
- brw_MOV(p, brw_message_reg(nr), offset(arg1_0, 1));
- /* 2nd half? */
- brw_MOV(p, brw_message_reg(nr+1), arg1_1);
- brw_pop_insn_state(p);
- }
- else
- {
- struct brw_reg src = get_src_reg(c, inst, 1, 1);
- brw_MOV(p, brw_message_reg(nr), src);
- }
- nr += 2;
- }
-
- target = INST_AUX_GET_TARGET(inst->Aux);
- eot = inst->Aux & INST_AUX_EOT;
- fire_fb_write(c, 0, nr, target, eot);
-}
-
-static void emit_pixel_w( struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
- if (mask & WRITEMASK_W) {
- struct brw_reg dst, src0, delta0, delta1;
- struct brw_reg interp3;
-
- dst = get_dst_reg(c, inst, 3);
- src0 = get_src_reg(c, inst, 0, 0);
- delta0 = get_src_reg(c, inst, 1, 0);
- delta1 = get_src_reg(c, inst, 1, 1);
-
- interp3 = brw_vec1_grf(src0.nr+1, 4);
- /* Calc 1/w - just linterp wpos[3] optimized by putting the
- * result straight into a message reg.
- */
- brw_LINE(p, brw_null_reg(), interp3, delta0);
- brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), delta1);
-
- /* Calc w */
- brw_math_16( p, dst,
- BRW_MATH_FUNCTION_INV,
- BRW_MATH_SATURATE_NONE,
- 2, brw_null_reg(),
- BRW_MATH_PRECISION_FULL);
- }
-}
-
-static void emit_linterp(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
- struct brw_reg interp[4];
- struct brw_reg dst, delta0, delta1;
- struct brw_reg src0;
- GLuint nr, i;
-
- src0 = get_src_reg(c, inst, 0, 0);
- delta0 = get_src_reg(c, inst, 1, 0);
- delta1 = get_src_reg(c, inst, 1, 1);
- nr = src0.nr;
-
- interp[0] = brw_vec1_grf(nr, 0);
- interp[1] = brw_vec1_grf(nr, 4);
- interp[2] = brw_vec1_grf(nr+1, 0);
- interp[3] = brw_vec1_grf(nr+1, 4);
-
- for(i = 0; i < 4; i++ ) {
- if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i);
- brw_LINE(p, brw_null_reg(), interp[i], delta0);
- brw_MAC(p, dst, suboffset(interp[i],1), delta1);
- }
- }
-}
-
-static void emit_cinterp(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
-
- struct brw_reg interp[4];
- struct brw_reg dst, src0;
- GLuint nr, i;
-
- src0 = get_src_reg(c, inst, 0, 0);
- nr = src0.nr;
-
- interp[0] = brw_vec1_grf(nr, 0);
- interp[1] = brw_vec1_grf(nr, 4);
- interp[2] = brw_vec1_grf(nr+1, 0);
- interp[3] = brw_vec1_grf(nr+1, 4);
-
- for(i = 0; i < 4; i++ ) {
- if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i);
- brw_MOV(p, dst, suboffset(interp[i],3));
- }
- }
-}
-
-static void emit_pinterp(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
-
- struct brw_reg interp[4];
- struct brw_reg dst, delta0, delta1;
- struct brw_reg src0, w;
- GLuint nr, i;
-
- src0 = get_src_reg(c, inst, 0, 0);
- delta0 = get_src_reg(c, inst, 1, 0);
- delta1 = get_src_reg(c, inst, 1, 1);
- w = get_src_reg(c, inst, 2, 3);
- nr = src0.nr;
-
- interp[0] = brw_vec1_grf(nr, 0);
- interp[1] = brw_vec1_grf(nr, 4);
- interp[2] = brw_vec1_grf(nr+1, 0);
- interp[3] = brw_vec1_grf(nr+1, 4);
-
- for(i = 0; i < 4; i++ ) {
- if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i);
- brw_LINE(p, brw_null_reg(), interp[i], delta0);
- brw_MAC(p, dst, suboffset(interp[i],1),
- delta1);
- brw_MUL(p, dst, dst, w);
- }
- }
-}
-
-/* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */
-static void emit_frontfacing(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD);
- struct brw_reg dst;
- GLuint mask = inst->DstReg.WriteMask;
- int i;
-
- for (i = 0; i < 4; i++) {
- if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i);
- brw_MOV(p, dst, brw_imm_f(0.0));
- }
- }
-
- /* bit 31 is "primitive is back face", so checking < (1 << 31) gives
- * us front face
- */
- brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, r1_6ud, brw_imm_ud(1 << 31));
- for (i = 0; i < 4; i++) {
- if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i);
- brw_MOV(p, dst, brw_imm_f(1.0));
- }
- }
- brw_set_predicate_control_flag_value(p, 0xff);
-}
-
-static void emit_xpd(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- int i;
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
- for (i = 0; i < 4; i++) {
- GLuint i2 = (i+2)%3;
- GLuint i1 = (i+1)%3;
- if (mask & (1<<i)) {
- struct brw_reg src0, src1, dst;
- dst = get_dst_reg(c, inst, i);
- src0 = negate(get_src_reg(c, inst, 0, i2));
- src1 = get_src_reg_imm(c, inst, 1, i1);
- brw_MUL(p, brw_null_reg(), src0, src1);
- src0 = get_src_reg(c, inst, 0, i1);
- src1 = get_src_reg_imm(c, inst, 1, i2);
- brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF);
- brw_MAC(p, dst, src0, src1);
- brw_set_saturate(p, 0);
- }
- }
- brw_set_saturate(p, 0);
-}
-
-/**
- * Emit a scalar instruction, like RCP, RSQ, LOG, EXP.
- * Note that the result of the function is smeared across the dest
- * register's X, Y, Z and W channels (subject to writemasking of course).
- */
-static void emit_math1(struct brw_wm_compile *c,
- const struct prog_instruction *inst, GLuint func)
-{
- struct brw_compile *p = &c->func;
- struct brw_reg src0, dst;
- GLuint mask = inst->DstReg.WriteMask;
- int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1;
-
- if (!(mask & WRITEMASK_XYZW))
- return;
-
- assert(is_power_of_two(mask & WRITEMASK_XYZW));
-
- /* Get first component of source register */
- dst = get_dst_reg(c, inst, dst_chan);
- src0 = get_src_reg(c, inst, 0, 0);
-
- brw_MOV(p, brw_message_reg(2), src0);
- brw_math(p,
- dst,
- func,
- (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
- 2,
- brw_null_reg(),
- BRW_MATH_DATA_VECTOR,
- BRW_MATH_PRECISION_FULL);
-}
-
-static void emit_rcp(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_INV);
-}
-
-static void emit_rsq(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_RSQ);
-}
-
-static void emit_sin(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_SIN);
-}
-
-static void emit_cos(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_COS);
-}
-
-static void emit_ex2(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_EXP);
-}
-
-static void emit_lg2(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_LOG);
-}
-
static void emit_arl(struct brw_wm_compile *c,
const struct prog_instruction *inst)
{
release_tmps(c, mark);
}
-static void emit_pow(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- struct brw_reg dst, src0, src1;
- GLuint mask = inst->DstReg.WriteMask;
- int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1;
-
- if (!(mask & WRITEMASK_XYZW))
- return;
-
- assert(is_power_of_two(mask & WRITEMASK_XYZW));
-
- dst = get_dst_reg(c, inst, dst_chan);
- src0 = get_src_reg_imm(c, inst, 0, 0);
- src1 = get_src_reg_imm(c, inst, 1, 0);
-
- brw_MOV(p, brw_message_reg(2), src0);
- brw_MOV(p, brw_message_reg(3), src1);
-
- brw_math(p,
- dst,
- BRW_MATH_FUNCTION_POW,
- (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
- 2,
- brw_null_reg(),
- BRW_MATH_DATA_VECTOR,
- BRW_MATH_PRECISION_FULL);
-}
-
/**
* For GLSL shaders, this KIL will be unconditional.
* It may be contained inside an IF/ENDIF structure of course.
release_tmps( c, mark );
}
-
-static void emit_wpos_xy(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
- struct brw_reg src0[2], dst[2];
-
- dst[0] = get_dst_reg(c, inst, 0);
- dst[1] = get_dst_reg(c, inst, 1);
- src0[0] = get_src_reg(c, inst, 0, 0);
- src0[1] = get_src_reg(c, inst, 0, 1);
-
- /* Calculate the pixel offset from window bottom left into destination
- * X and Y channels.
- */
- if (mask & WRITEMASK_X) {
- /* X' = X - origin_x */
- brw_ADD(p,
- dst[0],
- retype(src0[0], BRW_REGISTER_TYPE_W),
- brw_imm_d(0 - c->key.origin_x));
- }
-
- if (mask & WRITEMASK_Y) {
- /* Y' = height - (Y - origin_y) = height + origin_y - Y */
- brw_ADD(p,
- dst[1],
- negate(retype(src0[1], BRW_REGISTER_TYPE_W)),
- brw_imm_d(c->key.origin_y + c->key.drawable_height - 1));
- }
-}
/* TODO
BIAS on SIMD8 not working yet...
switch (inst->Opcode) {
case WM_PIXELXY:
- emit_pixel_xy(c, inst);
+ emit_pixel_xy(c, dst, dst_flags);
break;
case WM_DELTAXY:
- emit_delta_xy(c, inst);
+ emit_delta_xy(p, dst, dst_flags, args[0]);
break;
case WM_PIXELW:
- emit_pixel_w(c, inst);
+ emit_pixel_w(c, dst, dst_flags, args[0], args[1]);
break;
case WM_LINTERP:
- emit_linterp(c, inst);
+ emit_linterp(p, dst, dst_flags, args[0], args[1]);
break;
case WM_PINTERP:
- emit_pinterp(c, inst);
+ emit_pinterp(p, dst, dst_flags, args[0], args[1], args[2]);
break;
case WM_CINTERP:
- emit_cinterp(c, inst);
+ emit_cinterp(p, dst, dst_flags, args[0]);
break;
case WM_WPOSXY:
- emit_wpos_xy(c, inst);
+ emit_wpos_xy(c, dst, dst_flags, args[0]);
break;
case WM_FB_WRITE:
- emit_fb_write(c, inst);
+ emit_fb_write(c, args[0], args[1], args[2],
+ INST_AUX_GET_TARGET(inst->Aux),
+ inst->Aux & INST_AUX_EOT);
break;
case WM_FRONTFACING:
- emit_frontfacing(c, inst);
+ emit_frontfacing(p, dst, dst_flags);
break;
case OPCODE_ADD:
emit_alu2(p, brw_ADD, dst, dst_flags, args[0], args[1]);
emit_dp4(p, dst, dst_flags, args[0], args[1]);
break;
case OPCODE_XPD:
- emit_xpd(c, inst);
+ emit_xpd(p, dst, dst_flags, args[0], args[1]);
break;
case OPCODE_DPH:
emit_dph(p, dst, dst_flags, args[0], args[1]);
break;
case OPCODE_RCP:
- emit_rcp(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]);
break;
case OPCODE_RSQ:
- emit_rsq(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]);
break;
case OPCODE_SIN:
- emit_sin(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]);
break;
case OPCODE_COS:
- emit_cos(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]);
break;
case OPCODE_EX2:
- emit_ex2(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]);
break;
case OPCODE_LG2:
- emit_lg2(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]);
break;
case OPCODE_MIN:
case OPCODE_MAX:
emit_alu2(p, brw_MUL, dst, dst_flags, args[0], args[1]);
break;
case OPCODE_POW:
- emit_pow(c, inst);
+ emit_math2(c, BRW_MATH_FUNCTION_POW,
+ dst, dst_flags, args[0], args[1]);
break;
case OPCODE_MAD:
emit_mad(p, dst, dst_flags, args[0], args[1], args[2]);
projects / mesa.git / blobdiff