static void
compile_instruction(
struct gl_context *ctx,
struct st_translate *t,
const struct prog_instruction *inst,
boolean clamp_dst_color_output)
{
struct ureg_program *ureg = t->ureg;
GLuint i;
struct ureg_dst dst[1] = { { 0 } };
struct ureg_src src[4];
unsigned num_dst;
unsigned num_src;
num_dst = _mesa_num_inst_dst_regs( inst->Opcode );
num_src = _mesa_num_inst_src_regs( inst->Opcode );
if (num_dst)
dst[0] = translate_dst( t,
&inst->DstReg,
inst->Saturate,
clamp_dst_color_output);
for (i = 0; i < num_src; i++)
src[i] = translate_src( t, &inst->SrcReg[i] );
switch( inst->Opcode ) {
case OPCODE_SWZ:
emit_swz( t, dst[0], &inst->SrcReg[0] );
return;
case OPCODE_BGNLOOP:
case OPCODE_CAL:
case OPCODE_ELSE:
case OPCODE_ENDLOOP:
debug_assert(num_dst == 0);
ureg_label_insn( ureg,
translate_opcode( inst->Opcode ),
src, num_src,
get_label( t, inst->BranchTarget ));
return;
case OPCODE_IF:
debug_assert(num_dst == 0);
ureg_label_insn( ureg,
ctx->Const.NativeIntegers ? TGSI_OPCODE_UIF : TGSI_OPCODE_IF,
src, num_src,
get_label( t, inst->BranchTarget ));
return;
case OPCODE_TEX:
case OPCODE_TXB:
case OPCODE_TXD:
case OPCODE_TXL:
case OPCODE_TXP:
src[num_src++] = t->samplers[inst->TexSrcUnit];
ureg_tex_insn( ureg,
translate_opcode( inst->Opcode ),
dst, num_dst,
st_translate_texture_target( inst->TexSrcTarget,
inst->TexShadow ),
NULL, 0,
src, num_src );
return;
case OPCODE_SCS:
dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XY );
ureg_insn( ureg,
translate_opcode( inst->Opcode ),
dst, num_dst,
src, num_src );
break;
case OPCODE_XPD:
dst[0] = ureg_writemask(dst[0], TGSI_WRITEMASK_XYZ );
ureg_insn( ureg,
translate_opcode( inst->Opcode ),
dst, num_dst,
src, num_src );
break;
case OPCODE_NOISE1:
case OPCODE_NOISE2:
case OPCODE_NOISE3:
case OPCODE_NOISE4:
/* At some point, a motivated person could add a better
* implementation of noise. Currently not even the nvidia
* binary drivers do anything more than this. In any case, the
* place to do this is in the GL state tracker, not the poor
* driver.
*/
ureg_MOV( ureg, dst[0], ureg_imm1f(ureg, 0.5) );
break;
case OPCODE_DDY:
emit_ddy( t, dst[0], &inst->SrcReg[0] );
break;
case OPCODE_RSQ:
ureg_RSQ( ureg, dst[0], ureg_abs(src[0]) );
break;
default:
ureg_insn( ureg,
translate_opcode( inst->Opcode ),
dst, num_dst,
src, num_src );
break;
}
}
static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
{
#define MAX_IFSN 32
#define MAX_LOOP_DEPTH 32
struct brw_instruction *if_inst[MAX_IFSN], *loop_inst[MAX_LOOP_DEPTH];
struct brw_instruction *inst0, *inst1;
int i, if_insn = 0, loop_insn = 0;
struct brw_compile *p = &c->func;
struct brw_indirect stack_index = brw_indirect(0, 0);
c->reg_index = 0;
prealloc_reg(c);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack));
for (i = 0; i < c->nr_fp_insns; i++) {
struct prog_instruction *inst = &c->prog_instructions[i];
struct prog_instruction *orig_inst;
if ((orig_inst = inst->Data) != 0)
orig_inst->Data = current_insn(p);
if (inst->CondUpdate)
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
else
brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
switch (inst->Opcode) {
case WM_PIXELXY:
emit_pixel_xy(c, inst);
break;
case WM_DELTAXY:
emit_delta_xy(c, inst);
break;
case WM_PIXELW:
emit_pixel_w(c, inst);
break;
case WM_LINTERP:
emit_linterp(c, inst);
break;
case WM_PINTERP:
emit_pinterp(c, inst);
break;
case WM_CINTERP:
emit_cinterp(c, inst);
break;
case WM_WPOSXY:
emit_wpos_xy(c, inst);
break;
case WM_FB_WRITE:
emit_fb_write(c, inst);
break;
case OPCODE_ABS:
emit_abs(c, inst);
break;
case OPCODE_ADD:
emit_add(c, inst);
break;
case OPCODE_SUB:
emit_sub(c, inst);
break;
case OPCODE_FRC:
emit_frc(c, inst);
break;
case OPCODE_FLR:
emit_flr(c, inst);
break;
case OPCODE_LRP:
emit_lrp(c, inst);
break;
case OPCODE_INT:
emit_int(c, inst);
break;
case OPCODE_MOV:
emit_mov(c, inst);
break;
case OPCODE_DP3:
emit_dp3(c, inst);
break;
case OPCODE_DP4:
emit_dp4(c, inst);
break;
case OPCODE_XPD:
emit_xpd(c, inst);
break;
case OPCODE_DPH:
emit_dph(c, inst);
break;
case OPCODE_RCP:
emit_rcp(c, inst);
break;
case OPCODE_RSQ:
emit_rsq(c, inst);
break;
case OPCODE_SIN:
emit_sin(c, inst);
break;
case OPCODE_COS:
emit_cos(c, inst);
break;
case OPCODE_EX2:
emit_ex2(c, inst);
break;
case OPCODE_LG2:
emit_lg2(c, inst);
break;
case OPCODE_MAX:
emit_max(c, inst);
break;
case OPCODE_MIN:
emit_min(c, inst);
break;
case OPCODE_DDX:
emit_ddx(c, inst);
break;
case OPCODE_DDY:
emit_ddy(c, inst);
break;
case OPCODE_SLT:
emit_slt(c, inst);
break;
case OPCODE_SLE:
emit_sle(c, inst);
break;
case OPCODE_SGT:
emit_sgt(c, inst);
break;
case OPCODE_SGE:
emit_sge(c, inst);
break;
case OPCODE_SEQ:
emit_seq(c, inst);
break;
case OPCODE_SNE:
emit_sne(c, inst);
break;
case OPCODE_MUL:
emit_mul(c, inst);
break;
case OPCODE_POW:
emit_pow(c, inst);
break;
case OPCODE_MAD:
emit_mad(c, inst);
break;
case OPCODE_TEX:
emit_tex(c, inst);
break;
case OPCODE_TXB:
emit_txb(c, inst);
break;
case OPCODE_KIL_NV:
emit_kil(c);
break;
case OPCODE_IF:
assert(if_insn < MAX_IFSN);
if_inst[if_insn++] = brw_IF(p, BRW_EXECUTE_8);
break;
case OPCODE_ELSE:
if_inst[if_insn-1] = brw_ELSE(p, if_inst[if_insn-1]);
break;
case OPCODE_ENDIF:
assert(if_insn > 0);
brw_ENDIF(p, if_inst[--if_insn]);
break;
case OPCODE_BGNSUB:
case OPCODE_ENDSUB:
break;
case OPCODE_CAL:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_access_mode(p, BRW_ALIGN_1);
brw_ADD(p, deref_1ud(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(4));
orig_inst = inst->Data;
orig_inst->Data = &p->store[p->nr_insn];
brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
brw_pop_insn_state(p);
break;
case OPCODE_RET:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(-4));
brw_set_access_mode(p, BRW_ALIGN_1);
brw_MOV(p, brw_ip_reg(), deref_1ud(stack_index, 0));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_pop_insn_state(p);
break;
case OPCODE_BGNLOOP:
loop_inst[loop_insn++] = brw_DO(p, BRW_EXECUTE_8);
break;
case OPCODE_BRK:
brw_BREAK(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_CONT:
brw_CONT(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_ENDLOOP:
loop_insn--;
inst0 = inst1 = brw_WHILE(p, loop_inst[loop_insn]);
/* patch all the BREAK instructions from
last BEGINLOOP */
while (inst0 > loop_inst[loop_insn]) {
inst0--;
if (inst0->header.opcode == BRW_OPCODE_BREAK) {
inst0->bits3.if_else.jump_count = inst1 - inst0 + 1;
inst0->bits3.if_else.pop_count = 0;
} else if (inst0->header.opcode == BRW_OPCODE_CONTINUE) {
inst0->bits3.if_else.jump_count = inst1 - inst0;
inst0->bits3.if_else.pop_count = 0;
}
}
break;
default:
_mesa_printf("unsupported IR in fragment shader %d\n",
inst->Opcode);
}
if (inst->CondUpdate)
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
else
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
post_wm_emit(c);
for (i = 0; i < c->fp->program.Base.NumInstructions; i++)
c->fp->program.Base.Instructions[i].Data = NULL;
}
