template<typename VectorType> void lpNorm(const VectorType& v)
{
VectorType u = VectorType::Random(v.size());
VERIFY_IS_APPROX(u.template lpNorm<Infinity>(), u.cwise().abs().maxCoeff());
VERIFY_IS_APPROX(u.template lpNorm<1>(), u.cwise().abs().sum());
VERIFY_IS_APPROX(u.template lpNorm<2>(), ei_sqrt(u.cwise().abs().cwise().square().sum()));
VERIFY_IS_APPROX(ei_pow(u.template lpNorm<5>(), typename VectorType::RealScalar(5)), u.cwise().abs().cwise().pow(5).sum());
}
static void translate_vertex_program(struct r300_vertex_program_compiler * compiler)
{
struct rc_instruction *rci;
compiler->code->pos_end = 0; /* Not supported yet */
compiler->code->length = 0;
compiler->SetHwInputOutput(compiler);
for(rci = compiler->Base.Program.Instructions.Next; rci != &compiler->Base.Program.Instructions; rci = rci->Next) {
struct rc_sub_instruction *vpi = &rci->U.I;
unsigned int *inst = compiler->code->body.d + compiler->code->length;
/* Skip instructions writing to non-existing destination */
if (!valid_dst(compiler->code, &vpi->DstReg))
continue;
if (compiler->code->length >= VSF_MAX_FRAGMENT_LENGTH) {
rc_error(&compiler->Base, "Vertex program has too many instructions\n");
return;
}
switch (vpi->Opcode) {
case RC_OPCODE_ADD: ei_vector2(compiler->code, VE_ADD, vpi, inst); break;
case RC_OPCODE_ARL: ei_vector1(compiler->code, VE_FLT2FIX_DX, vpi, inst); break;
case RC_OPCODE_DP4: ei_vector2(compiler->code, VE_DOT_PRODUCT, vpi, inst); break;
case RC_OPCODE_DST: ei_vector2(compiler->code, VE_DISTANCE_VECTOR, vpi, inst); break;
case RC_OPCODE_EX2: ei_math1(compiler->code, ME_EXP_BASE2_FULL_DX, vpi, inst); break;
case RC_OPCODE_EXP: ei_math1(compiler->code, ME_EXP_BASE2_DX, vpi, inst); break;
case RC_OPCODE_FRC: ei_vector1(compiler->code, VE_FRACTION, vpi, inst); break;
case RC_OPCODE_LG2: ei_math1(compiler->code, ME_LOG_BASE2_FULL_DX, vpi, inst); break;
case RC_OPCODE_LIT: ei_lit(compiler->code, vpi, inst); break;
case RC_OPCODE_LOG: ei_math1(compiler->code, ME_LOG_BASE2_DX, vpi, inst); break;
case RC_OPCODE_MAD: ei_mad(compiler->code, vpi, inst); break;
case RC_OPCODE_MAX: ei_vector2(compiler->code, VE_MAXIMUM, vpi, inst); break;
case RC_OPCODE_MIN: ei_vector2(compiler->code, VE_MINIMUM, vpi, inst); break;
case RC_OPCODE_MOV: ei_vector1(compiler->code, VE_ADD, vpi, inst); break;
case RC_OPCODE_MUL: ei_vector2(compiler->code, VE_MULTIPLY, vpi, inst); break;
case RC_OPCODE_POW: ei_pow(compiler->code, vpi, inst); break;
case RC_OPCODE_RCP: ei_math1(compiler->code, ME_RECIP_DX, vpi, inst); break;
case RC_OPCODE_RSQ: ei_math1(compiler->code, ME_RECIP_SQRT_DX, vpi, inst); break;
case RC_OPCODE_SGE: ei_vector2(compiler->code, VE_SET_GREATER_THAN_EQUAL, vpi, inst); break;
case RC_OPCODE_SLT: ei_vector2(compiler->code, VE_SET_LESS_THAN, vpi, inst); break;
default:
rc_error(&compiler->Base, "Unknown opcode %i\n", vpi->Opcode);
return;
}
compiler->code->length += 4;
if (compiler->Base.Error)
return;
}
}
static void translate_vertex_program(struct radeon_compiler *c, void *user)
{
struct r300_vertex_program_compiler *compiler = (struct r300_vertex_program_compiler*)c;
struct rc_instruction *rci;
struct loop * loops = NULL;
int current_loop_depth = 0;
int loops_reserved = 0;
unsigned int branch_depth = 0;
compiler->code->pos_end = 0; /* Not supported yet */
compiler->code->length = 0;
compiler->code->num_temporaries = 0;
compiler->SetHwInputOutput(compiler);
for(rci = compiler->Base.Program.Instructions.Next; rci != &compiler->Base.Program.Instructions; rci = rci->Next) {
struct rc_sub_instruction *vpi = &rci->U.I;
unsigned int *inst = compiler->code->body.d + compiler->code->length;
const struct rc_opcode_info *info = rc_get_opcode_info(vpi->Opcode);
/* Skip instructions writing to non-existing destination */
if (!valid_dst(compiler->code, &vpi->DstReg))
continue;
if (info->HasDstReg) {
/* Neither is Saturate. */
if (vpi->SaturateMode != RC_SATURATE_NONE) {
rc_error(&compiler->Base, "Vertex program does not support the Saturate "
"modifier (yet).\n");
}
}
if (compiler->code->length >= c->max_alu_insts * 4) {
rc_error(&compiler->Base, "Vertex program has too many instructions\n");
return;
}
assert(compiler->Base.is_r500 ||
(vpi->Opcode != RC_OPCODE_SEQ &&
vpi->Opcode != RC_OPCODE_SNE));
switch (vpi->Opcode) {
case RC_OPCODE_ADD: ei_vector2(compiler->code, VE_ADD, vpi, inst); break;
case RC_OPCODE_ARL: ei_vector1(compiler->code, VE_FLT2FIX_DX, vpi, inst); break;
case RC_OPCODE_COS: ei_math1(compiler->code, ME_COS, vpi, inst); break;
case RC_OPCODE_DP4: ei_vector2(compiler->code, VE_DOT_PRODUCT, vpi, inst); break;
case RC_OPCODE_DST: ei_vector2(compiler->code, VE_DISTANCE_VECTOR, vpi, inst); break;
case RC_OPCODE_ELSE: ei_else(compiler, inst); break;
case RC_OPCODE_ENDIF: ei_endif(compiler, inst); branch_depth--; break;
case RC_OPCODE_EX2: ei_math1(compiler->code, ME_EXP_BASE2_FULL_DX, vpi, inst); break;
case RC_OPCODE_EXP: ei_math1(compiler->code, ME_EXP_BASE2_DX, vpi, inst); break;
case RC_OPCODE_FRC: ei_vector1(compiler->code, VE_FRACTION, vpi, inst); break;
case RC_OPCODE_IF: ei_if(compiler, rci, inst, branch_depth); branch_depth++; break;
case RC_OPCODE_LG2: ei_math1(compiler->code, ME_LOG_BASE2_FULL_DX, vpi, inst); break;
case RC_OPCODE_LIT: ei_lit(compiler->code, vpi, inst); break;
case RC_OPCODE_LOG: ei_math1(compiler->code, ME_LOG_BASE2_DX, vpi, inst); break;
case RC_OPCODE_MAD: ei_mad(compiler->code, vpi, inst); break;
case RC_OPCODE_MAX: ei_vector2(compiler->code, VE_MAXIMUM, vpi, inst); break;
case RC_OPCODE_MIN: ei_vector2(compiler->code, VE_MINIMUM, vpi, inst); break;
case RC_OPCODE_MOV: ei_vector1(compiler->code, VE_ADD, vpi, inst); break;
case RC_OPCODE_MUL: ei_vector2(compiler->code, VE_MULTIPLY, vpi, inst); break;
case RC_OPCODE_POW: ei_pow(compiler->code, vpi, inst); break;
case RC_OPCODE_RCP: ei_math1(compiler->code, ME_RECIP_DX, vpi, inst); break;
case RC_OPCODE_RSQ: ei_math1(compiler->code, ME_RECIP_SQRT_DX, vpi, inst); break;
case RC_OPCODE_SEQ: ei_vector2(compiler->code, VE_SET_EQUAL, vpi, inst); break;
case RC_OPCODE_SGE: ei_vector2(compiler->code, VE_SET_GREATER_THAN_EQUAL, vpi, inst); break;
case RC_OPCODE_SIN: ei_math1(compiler->code, ME_SIN, vpi, inst); break;
case RC_OPCODE_SLT: ei_vector2(compiler->code, VE_SET_LESS_THAN, vpi, inst); break;
case RC_OPCODE_SNE: ei_vector2(compiler->code, VE_SET_NOT_EQUAL, vpi, inst); break;
case RC_OPCODE_BGNLOOP:
{
struct loop * l;
if ((!compiler->Base.is_r500
&& loops_reserved >= R300_VS_MAX_LOOP_DEPTH)
|| loops_reserved >= R500_VS_MAX_FC_DEPTH) {
rc_error(&compiler->Base,
"Loops are nested too deep.");
return;
}
memory_pool_array_reserve(&compiler->Base.Pool,
struct loop, loops, current_loop_depth,
loops_reserved, 1);
l = &loops[current_loop_depth++];
memset(l , 0, sizeof(struct loop));
l->BgnLoop = (compiler->code->length / 4);
continue;
}
case RC_OPCODE_ENDLOOP:
{
struct loop * l;
unsigned int act_addr;
unsigned int last_addr;
unsigned int ret_addr;
assert(loops);
l = &loops[current_loop_depth - 1];
act_addr = l->BgnLoop - 1;
last_addr = (compiler->code->length / 4) - 1;
ret_addr = l->BgnLoop;
if (loops_reserved >= R300_VS_MAX_FC_OPS) {
rc_error(&compiler->Base,
"Too many flow control instructions.");
return;
}
if (compiler->Base.is_r500) {
compiler->code->fc_op_addrs.r500
[compiler->code->num_fc_ops].lw =
R500_PVS_FC_ACT_ADRS(act_addr)
| R500_PVS_FC_LOOP_CNT_JMP_INST(0xffff)
;
compiler->code->fc_op_addrs.r500
[compiler->code->num_fc_ops].uw =
R500_PVS_FC_LAST_INST(last_addr)
| R500_PVS_FC_RTN_INST(ret_addr)
;
} else {
compiler->code->fc_op_addrs.r300
[compiler->code->num_fc_ops] =
R300_PVS_FC_ACT_ADRS(act_addr)
| R300_PVS_FC_LOOP_CNT_JMP_INST(0xff)
| R300_PVS_FC_LAST_INST(last_addr)
| R300_PVS_FC_RTN_INST(ret_addr)
;
}
compiler->code->fc_loop_index[compiler->code->num_fc_ops] =
R300_PVS_FC_LOOP_INIT_VAL(0x0)
| R300_PVS_FC_LOOP_STEP_VAL(0x1)
;
compiler->code->fc_ops |= R300_VAP_PVS_FC_OPC_LOOP(
compiler->code->num_fc_ops);
compiler->code->num_fc_ops++;
current_loop_depth--;
continue;
}
default:
rc_error(&compiler->Base, "Unknown opcode %s\n", info->Name);
return;
}
/* Non-flow control instructions that are inside an if statement
* need to pay attention to the predicate bit. */
if (branch_depth
&& vpi->Opcode != RC_OPCODE_IF
&& vpi->Opcode != RC_OPCODE_ELSE
&& vpi->Opcode != RC_OPCODE_ENDIF) {
inst[0] |= (PVS_DST_PRED_ENABLE_MASK
<< PVS_DST_PRED_ENABLE_SHIFT);
inst[0] |= (PVS_DST_PRED_SENSE_MASK
<< PVS_DST_PRED_SENSE_SHIFT);
}
/* Update the number of temporaries. */
if (info->HasDstReg && vpi->DstReg.File == RC_FILE_TEMPORARY &&
vpi->DstReg.Index >= compiler->code->num_temporaries)
compiler->code->num_temporaries = vpi->DstReg.Index + 1;
for (unsigned i = 0; i < info->NumSrcRegs; i++)
if (vpi->SrcReg[i].File == RC_FILE_TEMPORARY &&
vpi->SrcReg[i].Index >= compiler->code->num_temporaries)
compiler->code->num_temporaries = vpi->SrcReg[i].Index + 1;
if (compiler->PredicateMask)
if (compiler->PredicateIndex >= compiler->code->num_temporaries)
compiler->code->num_temporaries = compiler->PredicateIndex + 1;
if (compiler->code->num_temporaries > compiler->Base.max_temp_regs) {
rc_error(&compiler->Base, "Too many temporaries.\n");
return;
}
compiler->code->length += 4;
if (compiler->Base.Error)
return;
}
}
static void translate_vertex_program(struct radeon_compiler *c, void *user)
{
struct r300_vertex_program_compiler *compiler = (struct r300_vertex_program_compiler*)c;
struct rc_instruction *rci;
unsigned loops[R500_PVS_MAX_LOOP_DEPTH];
unsigned loop_depth = 0;
compiler->code->pos_end = 0; /* Not supported yet */
compiler->code->length = 0;
compiler->code->num_temporaries = 0;
compiler->SetHwInputOutput(compiler);
for(rci = compiler->Base.Program.Instructions.Next; rci != &compiler->Base.Program.Instructions; rci = rci->Next) {
struct rc_sub_instruction *vpi = &rci->U.I;
unsigned int *inst = compiler->code->body.d + compiler->code->length;
const struct rc_opcode_info *info = rc_get_opcode_info(vpi->Opcode);
/* Skip instructions writing to non-existing destination */
if (!valid_dst(compiler->code, &vpi->DstReg))
continue;
if (info->HasDstReg) {
/* Neither is Saturate. */
if (vpi->SaturateMode != RC_SATURATE_NONE && !c->is_r500) {
rc_error(&compiler->Base, "Vertex program does not support the Saturate "
"modifier (yet).\n");
}
}
if (compiler->code->length >= c->max_alu_insts * 4) {
rc_error(&compiler->Base, "Vertex program has too many instructions\n");
return;
}
assert(compiler->Base.is_r500 ||
(vpi->Opcode != RC_OPCODE_SEQ &&
vpi->Opcode != RC_OPCODE_SNE));
switch (vpi->Opcode) {
case RC_OPCODE_ADD: ei_vector2(compiler->code, VE_ADD, vpi, inst); break;
case RC_OPCODE_ARL: ei_vector1(compiler->code, VE_FLT2FIX_DX, vpi, inst); break;
case RC_OPCODE_COS: ei_math1(compiler->code, ME_COS, vpi, inst); break;
case RC_OPCODE_DP4: ei_vector2(compiler->code, VE_DOT_PRODUCT, vpi, inst); break;
case RC_OPCODE_DST: ei_vector2(compiler->code, VE_DISTANCE_VECTOR, vpi, inst); break;
case RC_OPCODE_EX2: ei_math1(compiler->code, ME_EXP_BASE2_FULL_DX, vpi, inst); break;
case RC_OPCODE_EXP: ei_math1(compiler->code, ME_EXP_BASE2_DX, vpi, inst); break;
case RC_OPCODE_FRC: ei_vector1(compiler->code, VE_FRACTION, vpi, inst); break;
case RC_OPCODE_LG2: ei_math1(compiler->code, ME_LOG_BASE2_FULL_DX, vpi, inst); break;
case RC_OPCODE_LIT: ei_lit(compiler->code, vpi, inst); break;
case RC_OPCODE_LOG: ei_math1(compiler->code, ME_LOG_BASE2_DX, vpi, inst); break;
case RC_OPCODE_MAD: ei_mad(compiler->code, vpi, inst); break;
case RC_OPCODE_MAX: ei_vector2(compiler->code, VE_MAXIMUM, vpi, inst); break;
case RC_OPCODE_MIN: ei_vector2(compiler->code, VE_MINIMUM, vpi, inst); break;
case RC_OPCODE_MOV: ei_vector1(compiler->code, VE_ADD, vpi, inst); break;
case RC_OPCODE_MUL: ei_vector2(compiler->code, VE_MULTIPLY, vpi, inst); break;
case RC_OPCODE_POW: ei_pow(compiler->code, vpi, inst); break;
case RC_OPCODE_RCP: ei_math1(compiler->code, ME_RECIP_DX, vpi, inst); break;
case RC_OPCODE_RSQ: ei_math1(compiler->code, ME_RECIP_SQRT_DX, vpi, inst); break;
case RC_OPCODE_SEQ: ei_vector2(compiler->code, VE_SET_EQUAL, vpi, inst); break;
case RC_OPCODE_SGE: ei_vector2(compiler->code, VE_SET_GREATER_THAN_EQUAL, vpi, inst); break;
case RC_OPCODE_SIN: ei_math1(compiler->code, ME_SIN, vpi, inst); break;
case RC_OPCODE_SLT: ei_vector2(compiler->code, VE_SET_LESS_THAN, vpi, inst); break;
case RC_OPCODE_SNE: ei_vector2(compiler->code, VE_SET_NOT_EQUAL, vpi, inst); break;
case RC_OPCODE_BGNLOOP:
{
if ((!compiler->Base.is_r500
&& loop_depth >= R300_VS_MAX_LOOP_DEPTH)
|| loop_depth >= R500_PVS_MAX_LOOP_DEPTH) {
rc_error(&compiler->Base,
"Loops are nested too deep.");
return;
}
loops[loop_depth++] = ((compiler->code->length)/ 4) + 1;
break;
}
case RC_OPCODE_ENDLOOP:
{
unsigned int act_addr;
unsigned int last_addr;
unsigned int ret_addr;
ret_addr = loops[--loop_depth];
act_addr = ret_addr - 1;
last_addr = (compiler->code->length / 4) - 1;
if (loop_depth >= R300_VS_MAX_FC_OPS) {
rc_error(&compiler->Base,
"Too many flow control instructions.");
return;
}
if (compiler->Base.is_r500) {
compiler->code->fc_op_addrs.r500
[compiler->code->num_fc_ops].lw =
R500_PVS_FC_ACT_ADRS(act_addr)
| R500_PVS_FC_LOOP_CNT_JMP_INST(0x00ff)
;
compiler->code->fc_op_addrs.r500
[compiler->code->num_fc_ops].uw =
R500_PVS_FC_LAST_INST(last_addr)
| R500_PVS_FC_RTN_INST(ret_addr)
;
} else {
compiler->code->fc_op_addrs.r300
[compiler->code->num_fc_ops] =
R300_PVS_FC_ACT_ADRS(act_addr)
| R300_PVS_FC_LOOP_CNT_JMP_INST(0xff)
| R300_PVS_FC_LAST_INST(last_addr)
| R300_PVS_FC_RTN_INST(ret_addr)
;
}
compiler->code->fc_loop_index[compiler->code->num_fc_ops] =
R300_PVS_FC_LOOP_INIT_VAL(0x0)
| R300_PVS_FC_LOOP_STEP_VAL(0x1)
;
compiler->code->fc_ops |= R300_VAP_PVS_FC_OPC_LOOP(
compiler->code->num_fc_ops);
compiler->code->num_fc_ops++;
break;
}
case RC_ME_PRED_SET_CLR:
ei_math1(compiler->code, ME_PRED_SET_CLR, vpi, inst);
break;
case RC_ME_PRED_SET_INV:
ei_math1(compiler->code, ME_PRED_SET_INV, vpi, inst);
break;
case RC_ME_PRED_SET_POP:
ei_math1(compiler->code, ME_PRED_SET_POP, vpi, inst);
break;
case RC_ME_PRED_SET_RESTORE:
ei_math1(compiler->code, ME_PRED_SET_RESTORE, vpi, inst);
break;
case RC_ME_PRED_SEQ:
ei_math1(compiler->code, ME_PRED_SET_EQ, vpi, inst);
break;
case RC_ME_PRED_SNEQ:
ei_math1(compiler->code, ME_PRED_SET_NEQ, vpi, inst);
break;
case RC_VE_PRED_SNEQ_PUSH:
ei_vector2(compiler->code, VE_PRED_SET_NEQ_PUSH,
vpi, inst);
break;
default:
rc_error(&compiler->Base, "Unknown opcode %s\n", info->Name);
return;
}
if (vpi->DstReg.Pred != RC_PRED_DISABLED) {
inst[0] |= (PVS_DST_PRED_ENABLE_MASK
<< PVS_DST_PRED_ENABLE_SHIFT);
if (vpi->DstReg.Pred == RC_PRED_SET) {
inst[0] |= (PVS_DST_PRED_SENSE_MASK
<< PVS_DST_PRED_SENSE_SHIFT);
}
}
/* Update the number of temporaries. */
if (info->HasDstReg && vpi->DstReg.File == RC_FILE_TEMPORARY &&
vpi->DstReg.Index >= compiler->code->num_temporaries)
compiler->code->num_temporaries = vpi->DstReg.Index + 1;
for (unsigned i = 0; i < info->NumSrcRegs; i++)
if (vpi->SrcReg[i].File == RC_FILE_TEMPORARY &&
vpi->SrcReg[i].Index >= compiler->code->num_temporaries)
compiler->code->num_temporaries = vpi->SrcReg[i].Index + 1;
if (compiler->code->num_temporaries > compiler->Base.max_temp_regs) {
rc_error(&compiler->Base, "Too many temporaries.\n");
return;
}
compiler->code->length += 4;
if (compiler->Base.Error)
return;
}
}
