C++程序
|
3820行
|
109.11 KB
/*
* Copyright 2011 Christoph Bumiller
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#include "codegen/nv50_ir.h"
#include "codegen/nv50_ir_target.h"
#include "codegen/nv50_ir_build_util.h"
extern "C" {
#include "util/u_math.h"
}
namespace nv50_ir {
bool
Instruction::isNop() const
{
if (op == OP_PHI || op == OP_SPLIT || op == OP_MERGE || op == OP_CONSTRAINT)
return true;
if (terminator || join) // XXX: should terminator imply flow ?
return false;
if (op == OP_ATOM)
return false;
if (!fixed && op == OP_NOP)
return true;
if (defExists(0) && def(0).rep()->reg.data.id < 0) {
for (int d = 1; defExists(d); ++d)
if (def(d).rep()->reg.data.id >= 0)
WARN("part of vector result is unused !\n");
return true;
}
if (op == OP_MOV || op == OP_UNION) {
if (!getDef(0)->equals(getSrc(0)))
return false;
if (op == OP_UNION)
if (!def(0).rep()->equals(getSrc(1)))
return false;
return true;
}
return false;
}
bool Instruction::isDead() const
{
if (op == OP_STORE ||
op == OP_EXPORT ||
op == OP_ATOM ||
op == OP_SUSTB || op == OP_SUSTP || op == OP_SUREDP || op == OP_SUREDB ||
op == OP_WRSV)
return false;
for (int d = 0; defExists(d); ++d)
if (getDef(d)->refCount() || getDef(d)->reg.data.id >= 0)
return false;
if (terminator || asFlow())
return false;
if (fixed)
return false;
return true;
};
// =============================================================================
class CopyPropagation : public Pass
{
private:
virtual bool visit(BasicBlock *);
};
// Propagate all MOVs forward to make subsequent optimization easier, except if
// the sources stem from a phi, in which case we don't want to mess up potential
// swaps $rX <-> $rY, i.e. do not create live range overlaps of phi src and def.
bool
CopyPropagation::visit(BasicBlock *bb)
{
Instruction *mov, *si, *next;
for (mov = bb->getEntry(); mov; mov = next) {
next = mov->next;
if (mov->op != OP_MOV || mov->fixed || !mov->getSrc(0)->asLValue())
continue;
if (mov->getPredicate())
continue;
if (mov->def(0).getFile() != mov->src(0).getFile())
continue;
si = mov->getSrc(0)->getInsn();
if (mov->getDef(0)->reg.data.id < 0 && si && si->op != OP_PHI) {
// propagate
mov->def(0).replace(mov->getSrc(0), false);
delete_Instruction(prog, mov);
}
}
return true;
}
// =============================================================================
class MergeSplits : public Pass
{
private:
virtual bool visit(BasicBlock *);
};
// For SPLIT / MERGE pairs that operate on the same registers, replace the
// post-merge def with the SPLIT's source.
bool
MergeSplits::visit(BasicBlock *bb)
{
Instruction *i, *next, *si;
for (i = bb->getEntry(); i; i = next) {
next = i->next;
if (i->op != OP_MERGE || typeSizeof(i->dType) != 8)
continue;
si = i->getSrc(0)->getInsn();
if (si->op != OP_SPLIT || si != i->getSrc(1)->getInsn())
continue;
i->def(0).replace(si->getSrc(0), false);
delete_Instruction(prog, i);
}
return true;
}
// =============================================================================
class LoadPropagation : public Pass
{
private:
virtual bool visit(BasicBlock *);
void checkSwapSrc01(Instruction *);
bool isCSpaceLoad(Instruction *);
bool isImmdLoad(Instruction *);
bool isAttribOrSharedLoad(Instruction *);
};
bool
LoadPropagation::isCSpaceLoad(Instruction *ld)
{
return ld && ld->op == OP_LOAD && ld->src(0).getFile() == FILE_MEMORY_CONST;
}
bool
LoadPropagation::isImmdLoad(Instruction *ld)
{
if (!ld || (ld->op != OP_MOV) ||
((typeSizeof(ld->dType) != 4) && (typeSizeof(ld->dType) != 8)))
return false;
// A 0 can be replaced with a register, so it doesn't count as an immediate.
ImmediateValue val;
return ld->src(0).getImmediate(val) && !val.isInteger(0);
}
bool
LoadPropagation::isAttribOrSharedLoad(Instruction *ld)
{
return ld &&
(ld->op == OP_VFETCH ||
(ld->op == OP_LOAD &&
(ld->src(0).getFile() == FILE_SHADER_INPUT ||
ld->src(0).getFile() == FILE_MEMORY_SHARED)));
}
void
LoadPropagation::checkSwapSrc01(Instruction *insn)
{
const Target *targ = prog->getTarget();
if (!targ->getOpInfo(insn).commutative)
if (insn->op != OP_SET && insn->op != OP_SLCT && insn->op != OP_SUB)
return;
if (insn->src(1).getFile() != FILE_GPR)
return;
// This is the special OP_SET used for alphatesting, we can't reverse its
// arguments as that will confuse the fixup code.
if (insn->op == OP_SET && insn->subOp)
return;
Instruction *i0 = insn->getSrc(0)->getInsn();
Instruction *i1 = insn->getSrc(1)->getInsn();
// Swap sources to inline the less frequently used source. That way,
// optimistically, it will eventually be able to remove the instruction.
int i0refs = insn->getSrc(0)->refCount();
int i1refs = insn->getSrc(1)->refCount();
if ((isCSpaceLoad(i0) || isImmdLoad(i0)) && targ->insnCanLoad(insn, 1, i0)) {
if ((!isImmdLoad(i1) && !isCSpaceLoad(i1)) ||
!targ->insnCanLoad(insn, 1, i1) ||
i0refs < i1refs)
insn->swapSources(0, 1);
else
return;
} else
if (isAttribOrSharedLoad(i1)) {
if (!isAttribOrSharedLoad(i0))
insn->swapSources(0, 1);
else
return;
} else {
return;
}
if (insn->op == OP_SET || insn->op == OP_SET_AND ||
insn->op == OP_SET_OR || insn->op == OP_SET_XOR)
insn->asCmp()->setCond = reverseCondCode(insn->asCmp()->setCond);
else
if (insn->op == OP_SLCT)
insn->asCmp()->setCond = inverseCondCode(insn->asCmp()->setCond);
else
if (insn->op == OP_SUB) {
insn->src(0).mod = insn->src(0).mod ^ Modifier(NV50_IR_MOD_NEG);
insn->src(1).mod = insn->src(1).mod ^ Modifier(NV50_IR_MOD_NEG);
}
}
bool
LoadPropagation::visit(BasicBlock *bb)
{
const Target *targ = prog->getTarget();
Instruction *next;
for (Instruction *i = bb->getEntry(); i; i = next) {
next = i->next;
if (i->op == OP_CALL) // calls have args as sources, they must be in regs
continue;
if (i->op == OP_PFETCH) // pfetch expects arg1 to be a reg
continue;
if (i->srcExists(1))
checkSwapSrc01(i);
for (int s = 0; i->srcExists(s); ++s) {
Instruction *ld = i->getSrc(s)->getInsn();
if (!ld || ld->fixed || (ld->op != OP_LOAD && ld->op != OP_MOV))
continue;
if (!targ->insnCanLoad(i, s, ld))
continue;
// propagate !
i->setSrc(s, ld->getSrc(0));
if (ld->src(0).isIndirect(0))
i->setIndirect(s, 0, ld->getIndirect(0, 0));
if (ld->getDef(0)->refCount() == 0)
delete_Instruction(prog, ld);
}
}
return true;
}
// =============================================================================
class IndirectPropagation : public Pass
{
private:
virtual bool visit(BasicBlock *);
};
bool
IndirectPropagation::visit(BasicBlock *bb)
{
const Target *targ = prog->getTarget();
Instruction *next;
for (Instruction *i = bb->getEntry(); i; i = next) {
next = i->next;
for (int s = 0; i->srcExists(s); ++s) {
Instruction *insn;
ImmediateValue imm;
if (!i->src(s).isIndirect(0))
continue;
insn = i->getIndirect(s, 0)->getInsn();
if (!insn)
continue;
if (insn->op == OP_ADD && !isFloatType(insn->dType)) {
if (insn->src(0).getFile() != targ->nativeFile(FILE_ADDRESS) ||
!insn->src(1).getImmediate(imm) ||
!targ->insnCanLoadOffset(i, s, imm.reg.data.s32))
continue;
i->setIndirect(s, 0, insn->getSrc(0));
i->setSrc(s, cloneShallow(func, i->getSrc(s)));
i->src(s).get()->reg.data.offset += imm.reg.data.u32;
} else if (insn->op == OP_SUB && !isFloatType(insn->dType)) {
if (insn->src(0).getFile() != targ->nativeFile(FILE_ADDRESS) ||
!insn->src(1).getImmediate(imm) ||
!targ->insnCanLoadOffset(i, s, -imm.reg.data.s32))
continue;
i->setIndirect(s, 0, insn->getSrc(0));
i->setSrc(s, cloneShallow(func, i->getSrc(s)));
i->src(s).get()->reg.data.offset -= imm.reg.data.u32;
} else if (insn->op == OP_MOV) {
if (!insn->src(0).getImmediate(imm) ||
!targ->insnCanLoadOffset(i, s, imm.reg.data.s32))
continue;
i->setIndirect(s, 0, NULL);
i->setSrc(s, cloneShallow(func, i->getSrc(s)));
i->src(s).get()->reg.data.offset += imm.reg.data.u32;
}
}
}
return true;
}
// =============================================================================
// Evaluate constant expressions.
class ConstantFolding : public Pass
{
public:
bool foldAll(Program *);
private:
virtual bool visit(BasicBlock *);
void expr(Instruction *, ImmediateValue&, ImmediateValue&);
void expr(Instruction *, ImmediateValue&, ImmediateValue&, ImmediateValue&);
void opnd(Instruction *, ImmediateValue&, int s);
void opnd3(Instruction *, ImmediateValue&);
void unary(Instruction *, const ImmediateValue&);
void tryCollapseChainedMULs(Instruction *, const int s, ImmediateValue&);
CmpInstruction *findOriginForTestWithZero(Value *);
unsigned int foldCount;
BuildUtil bld;
};
// TODO: remember generated immediates and only revisit these
bool
ConstantFolding::foldAll(Program *prog)
{
unsigned int iterCount = 0;
do {
foldCount = 0;
if (!run(prog))
return false;
} while (foldCount && ++iterCount < 2);
return true;
}
bool
ConstantFolding::visit(BasicBlock *bb)
{
Instruction *i, *next;
for (i = bb->getEntry(); i; i = next) {
next = i->next;
if (i->op == OP_MOV || i->op == OP_CALL)
continue;
ImmediateValue src0, src1, src2;
if (i->srcExists(2) &&
i->src(0).getImmediate(src0) &&
i->src(1).getImmediate(src1) &&
i->src(2).getImmediate(src2))
expr(i, src0, src1, src2);
else
if (i->srcExists(1) &&
i->src(0).getImmediate(src0) && i->src(1).getImmediate(src1))
expr(i, src0, src1);
else
if (i->srcExists(0) && i->src(0).getImmediate(src0))
opnd(i, src0, 0);
else
if (i->srcExists(1) && i->src(1).getImmediate(src1))
opnd(i, src1, 1);
if (i->srcExists(2) && i->src(2).getImmediate(src2))
opnd3(i, src2);
}
return true;
}
CmpInstruction *
ConstantFolding::findOriginForTestWithZero(Value *value)
{
if (!value)
return NULL;
Instruction *insn = value->getInsn();
if (!insn)
return NULL;
if (insn->asCmp() && insn->op != OP_SLCT)
return insn->asCmp();
/* Sometimes mov's will sneak in as a result of other folding. This gets
* cleaned up later.
*/
if (insn->op == OP_MOV)
return findOriginForTestWithZero(insn->getSrc(0));
/* Deal with AND 1.0 here since nv50 can't fold into boolean float */
if (insn->op == OP_AND) {
int s = 0;
ImmediateValue imm;
if (!insn->src(s).getImmediate(imm)) {
s = 1;
if (!insn->src(s).getImmediate(imm))
return NULL;
}
if (imm.reg.data.f32 != 1.0f)
return NULL;
/* TODO: Come up with a way to handle the condition being inverted */
if (insn->src(!s).mod != Modifier(0))
return NULL;
return findOriginForTestWithZero(insn->getSrc(!s));
}
return NULL;
}
void
Modifier::applyTo(ImmediateValue& imm) const
{
if (!bits) // avoid failure if imm.reg.type is unhandled (e.g. b128)
return;
switch (imm.reg.type) {
case TYPE_F32:
if (bits & NV50_IR_MOD_ABS)
imm.reg.data.f32 = fabsf(imm.reg.data.f32);
if (bits & NV50_IR_MOD_NEG)
imm.reg.data.f32 = -imm.reg.data.f32;
if (bits & NV50_IR_MOD_SAT) {
if (imm.reg.data.f32 < 0.0f)
imm.reg.data.f32 = 0.0f;
else
if (imm.reg.data.f32 > 1.0f)
imm.reg.data.f32 = 1.0f;
}
assert(!(bits & NV50_IR_MOD_NOT));
break;
case TYPE_S8: // NOTE: will be extended
case TYPE_S16:
case TYPE_S32:
case TYPE_U8: // NOTE: treated as signed
case TYPE_U16:
case TYPE_U32:
if (bits & NV50_IR_MOD_ABS)
imm.reg.data.s32 = (imm.reg.data.s32 >= 0) ?
imm.reg.data.s32 : -imm.reg.data.s32;
if (bits & NV50_IR_MOD_NEG)
imm.reg.data.s32 = -imm.reg.data.s32;
if (bits & NV50_IR_MOD_NOT)
imm.reg.data.s32 = ~imm.reg.data.s32;
break;
case TYPE_F64:
if (bits & NV50_IR_MOD_ABS)
imm.reg.data.f64 = fabs(imm.reg.data.f64);
if (bits & NV50_IR_MOD_NEG)
imm.reg.data.f64 = -imm.reg.data.f64;
if (bits & NV50_IR_MOD_SAT) {
if (imm.reg.data.f64 < 0.0)
imm.reg.data.f64 = 0.0;
else
if (imm.reg.data.f64 > 1.0)
imm.reg.data.f64 = 1.0;
}
assert(!(bits & NV50_IR_MOD_NOT));
break;
default:
assert(!"invalid/unhandled type");
imm.reg.data.u64 = 0;
break;
}
}
operation
Modifier::getOp() const
{
switch (bits) {
case NV50_IR_MOD_ABS: return OP_ABS;
case NV50_IR_MOD_NEG: return OP_NEG;
case NV50_IR_MOD_SAT: return OP_SAT;
case NV50_IR_MOD_NOT: return OP_NOT;
case 0:
return OP_MOV;
default:
return OP_CVT;
}
}
void
ConstantFolding::expr(Instruction *i,
ImmediateValue &imm0, ImmediateValue &imm1)
{
struct Storage *const a = &imm0.reg, *const b = &imm1.reg;
struct Storage res;
DataType type = i->dType;
memset(&res.data, 0, sizeof(res.data));
switch (i->op) {
case OP_MAD:
case OP_FMA:
case OP_MUL:
if (i->dnz && i->dType == TYPE_F32) {
if (!isfinite(a->data.f32))
a->data.f32 = 0.0f;
if (!isfinite(b->data.f32))
b->data.f32 = 0.0f;
}
switch (i->dType) {
case TYPE_F32:
res.data.f32 = a->data.f32 * b->data.f32 * exp2f(i->postFactor);
break;
case TYPE_F64: res.data.f64 = a->data.f64 * b->data.f64; break;
case TYPE_S32:
if (i->subOp == NV50_IR_SUBOP_MUL_HIGH) {
res.data.s32 = ((int64_t)a->data.s32 * b->data.s32) >> 32;
break;
}
/* fallthrough */
case TYPE_U32:
if (i->subOp == NV50_IR_SUBOP_MUL_HIGH) {
res.data.u32 = ((uint64_t)a->data.u32 * b->data.u32) >> 32;
break;
}
res.data.u32 = a->data.u32 * b->data.u32; break;
default:
return;
}
break;
case OP_DIV:
if (b->data.u32 == 0)
break;
switch (i->dType) {
case TYPE_F32: res.data.f32 = a->data.f32 / b->data.f32; break;
case TYPE_F64: res.data.f64 = a->data.f64 / b->data.f64; break;
case TYPE_S32: res.data.s32 = a->data.s32 / b->data.s32; break;
case TYPE_U32: res.data.u32 = a->data.u32 / b->data.u32; break;
default:
return;
}
break;
case OP_ADD:
switch (i->dType) {
case TYPE_F32: res.data.f32 = a->data.f32 + b->data.f32; break;
case TYPE_F64: res.data.f64 = a->data.f64 + b->data.f64; break;
case TYPE_S32:
case TYPE_U32: res.data.u32 = a->data.u32 + b->data.u32; break;
default:
return;
}
break;
case OP_SUB:
switch (i->dType) {
case TYPE_F32: res.data.f32 = a->data.f32 - b->data.f32; break;
case TYPE_F64: res.data.f64 = a->data.f64 - b->data.f64; break;
case TYPE_S32:
case TYPE_U32: res.data.u32 = a->data.u32 - b->data.u32; break;
default:
return;
}
break;
case OP_POW:
switch (i->dType) {
case TYPE_F32: res.data.f32 = pow(a->data.f32, b->data.f32); break;
case TYPE_F64: res.data.f64 = pow(a->data.f64, b->data.f64); break;
default:
return;
}
break;
case OP_MAX:
switch (i->dType) {
case TYPE_F32: res.data.f32 = MAX2(a->data.f32, b->data.f32); break;
case TYPE_F64: res.data.f64 = MAX2(a->data.f64, b->data.f64); break;
case TYPE_S32: res.data.s32 = MAX2(a->data.s32, b->data.s32); break;
case TYPE_U32: res.data.u32 = MAX2(a->data.u32, b->data.u32); break;
default:
return;
}
break;
case OP_MIN:
switch (i->dType) {
case TYPE_F32: res.data.f32 = MIN2(a->data.f32, b->data.f32); break;
case TYPE_F64: res.data.f64 = MIN2(a->data.f64, b->data.f64); break;
case TYPE_S32: res.data.s32 = MIN2(a->data.s32, b->data.s32); break;
case TYPE_U32: res.data.u32 = MIN2(a->data.u32, b->data.u32); break;
default:
return;
}
break;
case OP_AND:
res.data.u64 = a->data.u64 & b->data.u64;
break;
case OP_OR:
res.data.u64 = a->data.u64 | b->data.u64;
break;
case OP_XOR:
res.data.u64 = a->data.u64 ^ b->data.u64;
break;
case OP_SHL:
res.data.u32 = a->data.u32 << b->data.u32;
break;
case OP_SHR:
switch (i->dType) {
case TYPE_S32: res.data.s32 = a->data.s32 >> b->data.u32; break;
case TYPE_U32: res.data.u32 = a->data.u32 >> b->data.u32; break;
default:
return;
}
break;
case OP_SLCT:
if (a->data.u32 != b->data.u32)
return;
res.data.u32 = a->data.u32;
break;
case OP_EXTBF: {
int offset = b->data.u32 & 0xff;
int width = (b->data.u32 >> 8) & 0xff;
int rshift = offset;
int lshift = 0;
if (width == 0) {
res.data.u32 = 0;
break;
}
if (width + offset < 32) {
rshift = 32 - width;
lshift = 32 - width - offset;
}
if (i->subOp == NV50_IR_SUBOP_EXTBF_REV)
res.data.u32 = util_bitreverse(a->data.u32);
else
res.data.u32 = a->data.u32;
switch (i->dType) {
case TYPE_S32: res.data.s32 = (res.data.s32 << lshift) >> rshift; break;
case TYPE_U32: res.data.u32 = (res.data.u32 << lshift) >> rshift; break;
default:
return;
}
break;
}
case OP_POPCNT:
res.data.u32 = util_bitcount(a->data.u32 & b->data.u32);
break;
case OP_PFETCH:
// The two arguments to pfetch are logically added together. Normally
// the second argument will not be constant, but that can happen.
res.data.u32 = a->data.u32 + b->data.u32;
type = TYPE_U32;
break;
case OP_MERGE:
switch (i->dType) {
case TYPE_U64:
case TYPE_S64:
case TYPE_F64:
res.data.u64 = (((uint64_t)b->data.u32) << 32) | a->data.u32;
break;
default:
return;
}
break;
default:
return;
}
++foldCount;
i->src(0).mod = Modifier(0);
i->src(1).mod = Modifier(0);
i->postFactor = 0;
i->setSrc(0, new_ImmediateValue(i->bb->getProgram(), res.data.u32));
i->setSrc(1, NULL);
i->getSrc(0)->reg.data = res.data;
i->getSrc(0)->reg.type = type;
i->getSrc(0)->reg.size = typeSizeof(type);
switch (i->op) {
case OP_MAD:
case OP_FMA: {
ImmediateValue src0, src1 = *i->getSrc(0)->asImm();
// Move the immediate into position 1, where we know it might be
// emittable. However it might not be anyways, as there may be other
// restrictions, so move it into a separate LValue.
bld.setPosition(i, false);
i->op = OP_ADD;
i->setSrc(1, bld.mkMov(bld.getSSA(type), i->getSrc(0), type)->getDef(0));
i->setSrc(0, i->getSrc(2));
i->src(0).mod = i->src(2).mod;
i->setSrc(2, NULL);
if (i->src(0).getImmediate(src0))
expr(i, src0, src1);
else
opnd(i, src1, 1);
break;
}
case OP_PFETCH:
// Leave PFETCH alone... we just folded its 2 args into 1.
break;
default:
i->op = i->saturate ? OP_SAT : OP_MOV;
if (i->saturate)
unary(i, *i->getSrc(0)->asImm());
break;
}
i->subOp = 0;
}
void
ConstantFolding::expr(Instruction *i,
ImmediateValue &imm0,
ImmediateValue &imm1,
ImmediateValue &imm2)
{
struct Storage *const a = &imm0.reg, *const b = &imm1.reg, *const c = &imm2.reg;
struct Storage res;
memset(&res.data, 0, sizeof(res.data));
switch (i->op) {
case OP_INSBF: {
int offset = b->data.u32 & 0xff;
int width = (b->data.u32 >> 8) & 0xff;
unsigned bitmask = ((1 << width) - 1) << offset;
res.data.u32 = ((a->data.u32 << offset) & bitmask) | (c->data.u32 & ~bitmask);
break;
}
case OP_MAD:
case OP_FMA: {
switch (i->dType) {
case TYPE_F32:
res.data.f32 = a->data.f32 * b->data.f32 * exp2f(i->postFactor) +
c->data.f32;
break;
case TYPE_F64:
res.data.f64 = a->data.f64 * b->data.f64 + c->data.f64;
break;
case TYPE_S32:
if (i->subOp == NV50_IR_SUBOP_MUL_HIGH) {
res.data.s32 = ((int64_t)a->data.s32 * b->data.s32 >> 32) + c->data.s32;
break;
}
/* fallthrough */
case TYPE_U32:
if (i->subOp == NV50_IR_SUBOP_MUL_HIGH) {
res.data.u32 = ((uint64_t)a->data.u32 * b->data.u32 >> 32) + c->data.u32;
break;
}
res.data.u32 = a->data.u32 * b->data.u32 + c->data.u32;
break;
default:
return;
}
break;
}
case OP_SHLADD:
res.data.u32 = (a->data.u32 << b->data.u32) + c->data.u32;
break;
default:
return;
}
++foldCount;
i->src(0).mod = Modifier(0);
i->src(1).mod = Modifier(0);
i->src(2).mod = Modifier(0);
i->setSrc(0, new_ImmediateValue(i->bb->getProgram(), res.data.u32));
i->setSrc(1, NULL);
i->setSrc(2, NULL);
i->getSrc(0)->reg.data = res.data;
i->getSrc(0)->reg.type = i->dType;
i->getSrc(0)->reg.size = typeSizeof(i->dType);
i->op = OP_MOV;
}
void
ConstantFolding::unary(Instruction *i, const ImmediateValue &imm)
{
Storage res;
if (i->dType != TYPE_F32)
return;
switch (i->op) {
case OP_NEG: res.data.f32 = -imm.reg.data.f32; break;
case OP_ABS: res.data.f32 = fabsf(imm.reg.data.f32); break;
case OP_SAT: res.data.f32 = CLAMP(imm.reg.data.f32, 0.0f, 1.0f); break;
case OP_RCP: res.data.f32 = 1.0f / imm.reg.data.f32; break;
case OP_RSQ: res.data.f32 = 1.0f / sqrtf(imm.reg.data.f32); break;
case OP_LG2: res.data.f32 = log2f(imm.reg.data.f32); break;
case OP_EX2: res.data.f32 = exp2f(imm.reg.data.f32); break;
case OP_SIN: res.data.f32 = sinf(imm.reg.data.f32); break;
case OP_COS: res.data.f32 = cosf(imm.reg.data.f32); break;
case OP_SQRT: res.data.f32 = sqrtf(imm.reg.data.f32); break;
case OP_PRESIN:
case OP_PREEX2:
// these should be handled in subsequent OP_SIN/COS/EX2
res.data.f32 = imm.reg.data.f32;
break;
default:
return;
}
i->op = OP_MOV;
i->setSrc(0, new_ImmediateValue(i->bb->getProgram(), res.data.f32));
i->src(0).mod = Modifier(0);
}
void
ConstantFolding::tryCollapseChainedMULs(Instruction *mul2,
const int s, ImmediateValue& imm2)
{
const int t = s ? 0 : 1;
Instruction *insn;
Instruction *mul1 = NULL; // mul1 before mul2
int e = 0;
float f = imm2.reg.data.f32 * exp2f(mul2->postFactor);
ImmediateValue imm1;
assert(mul2->op == OP_MUL && mul2->dType == TYPE_F32);
if (mul2->getSrc(t)->refCount() == 1) {
insn = mul2->getSrc(t)->getInsn();
if (!mul2->src(t).mod && insn->op == OP_MUL && insn->dType == TYPE_F32)
mul1 = insn;
if (mul1 && !mul1->saturate) {
int s1;
if (mul1->src(s1 = 0).getImmediate(imm1) ||
mul1->src(s1 = 1).getImmediate(imm1)) {
bld.setPosition(mul1, false);
// a = mul r, imm1
// d = mul a, imm2 -> d = mul r, (imm1 * imm2)
mul1->setSrc(s1, bld.loadImm(NULL, f * imm1.reg.data.f32));
mul1->src(s1).mod = Modifier(0);
mul2->def(0).replace(mul1->getDef(0), false);
mul1->saturate = mul2->saturate;
} else
if (prog->getTarget()->isPostMultiplySupported(OP_MUL, f, e)) {
// c = mul a, b
// d = mul c, imm -> d = mul_x_imm a, b
mul1->postFactor = e;
mul2->def(0).replace(mul1->getDef(0), false);
if (f < 0)
mul1->src(0).mod *= Modifier(NV50_IR_MOD_NEG);
mul1->saturate = mul2->saturate;
}
return;
}
}
if (mul2->getDef(0)->refCount() == 1 && !mul2->saturate) {
// b = mul a, imm
// d = mul b, c -> d = mul_x_imm a, c
int s2, t2;
insn = (*mul2->getDef(0)->uses.begin())->getInsn();
if (!insn)
return;
mul1 = mul2;
mul2 = NULL;
s2 = insn->getSrc(0) == mul1->getDef(0) ? 0 : 1;
t2 = s2 ? 0 : 1;
if (insn->op == OP_MUL && insn->dType == TYPE_F32)
if (!insn->src(s2).mod && !insn->src(t2).getImmediate(imm1))
mul2 = insn;
if (mul2 && prog->getTarget()->isPostMultiplySupported(OP_MUL, f, e)) {
mul2->postFactor = e;
mul2->setSrc(s2, mul1->src(t));
if (f < 0)
mul2->src(s2).mod *= Modifier(NV50_IR_MOD_NEG);
}
}
}
void
ConstantFolding::opnd3(Instruction *i, ImmediateValue &imm2)
{
switch (i->op) {
case OP_MAD:
case OP_FMA:
if (imm2.isInteger(0)) {
i->op = OP_MUL;
i->setSrc(2, NULL);
foldCount++;
return;
}
break;
case OP_SHLADD:
if (imm2.isInteger(0)) {
i->op = OP_SHL;
i->setSrc(2, NULL);
foldCount++;
return;
}
break;
default:
return;
}
}
void
ConstantFolding::opnd(Instruction *i, ImmediateValue &imm0, int s)
{
const Target *target = prog->getTarget();
const int t = !s;
const operation op = i->op;
Instruction *newi = i;
switch (i->op) {
case OP_SPLIT: {
bld.setPosition(i, false);
uint8_t size = i->getDef(0)->reg.size;
uint8_t bitsize = size * 8;
uint32_t mask = (1ULL << bitsize) - 1;
assert(bitsize <= 32);
uint64_t val = imm0.reg.data.u64;
for (int8_t d = 0; i->defExists(d); ++d) {
Value *def = i->getDef(d);
assert(def->reg.size == size);
newi = bld.mkMov(def, bld.mkImm((uint32_t)(val & mask)), TYPE_U32);
val >>= bitsize;
}
delete_Instruction(prog, i);
break;
}
case OP_MUL:
if (i->dType == TYPE_F32)
tryCollapseChainedMULs(i, s, imm0);
if (i->subOp == NV50_IR_SUBOP_MUL_HIGH) {
assert(!isFloatType(i->sType));
if (imm0.isInteger(1) && i->dType == TYPE_S32) {
bld.setPosition(i, false);
// Need to set to the sign value, which is a compare.
newi = bld.mkCmp(OP_SET, CC_LT, TYPE_S32, i->getDef(0),
TYPE_S32, i->getSrc(t), bld.mkImm(0));
delete_Instruction(prog, i);
} else if (imm0.isInteger(0) || imm0.isInteger(1)) {
// The high bits can't be set in this case (either mul by 0 or
// unsigned by 1)
i->op = OP_MOV;
i->subOp = 0;
i->setSrc(0, new_ImmediateValue(prog, 0u));
i->src(0).mod = Modifier(0);
i->setSrc(1, NULL);
} else if (!imm0.isNegative() && imm0.isPow2()) {
// Translate into a shift
imm0.applyLog2();
i->op = OP_SHR;
i->subOp = 0;
imm0.reg.data.u32 = 32 - imm0.reg.data.u32;
i->setSrc(0, i->getSrc(t));
i->src(0).mod = i->src(t).mod;
i->setSrc(1, new_ImmediateValue(prog, imm0.reg.data.u32));
i->src(1).mod = 0;
}
} else
if (imm0.isInteger(0)) {
i->op = OP_MOV;
i->setSrc(0, new_ImmediateValue(prog, 0u));
i->src(0).mod = Modifier(0);
i->postFactor = 0;
i->setSrc(1, NULL);
} else
if (!i->postFactor && (imm0.isInteger(1) || imm0.isInteger(-1))) {
if (imm0.isNegative())
i->src(t).mod = i->src(t).mod ^ Modifier(NV50_IR_MOD_NEG);
i->op = i->src(t).mod.getOp();
if (s == 0) {
i->setSrc(0, i->getSrc(1));
i->src(0).mod = i->src(1).mod;
i->src(1).mod = 0;
}
if (i->op != OP_CVT)
i->src(0).mod = 0;
i->setSrc(1, NULL);
} else
if (!i->postFactor && (imm0.isInteger(2) || imm0.isInteger(-2))) {
if (imm0.isNegative())
i->src(t).mod = i->src(t).mod ^ Modifier(NV50_IR_MOD_NEG);
i->op = OP_ADD;
i->setSrc(s, i->getSrc(t));
i->src(s).mod = i->src(t).mod;
} else
if (!isFloatType(i->sType) && !imm0.isNegative() && imm0.isPow2()) {
i->op = OP_SHL;
imm0.applyLog2();
i->setSrc(0, i->getSrc(t));
i->src(0).mod = i->src(t).mod;
i->setSrc(1, new_ImmediateValue(prog, imm0.reg.data.u32));
i->src(1).mod = 0;
} else
if (i->postFactor && i->sType == TYPE_F32) {
/* Can't emit a postfactor with an immediate, have to fold it in */
i->setSrc(s, new_ImmediateValue(
prog, imm0.reg.data.f32 * exp2f(i->postFactor)));
i->postFactor = 0;
}
break;
case OP_FMA:
case OP_MAD:
if (imm0.isInteger(0)) {
i->setSrc(0, i->getSrc(2));
i->src(0).mod = i->src(2).mod;
i->setSrc(1, NULL);
i->setSrc(2, NULL);
i->op = i->src(0).mod.getOp();
if (i->op != OP_CVT)
i->src(0).mod = 0;
} else
if (i->subOp != NV50_IR_SUBOP_MUL_HIGH &&
(imm0.isInteger(1) || imm0.isInteger(-1))) {
if (imm0.isNegative())
i->src(t).mod = i->src(t).mod ^ Modifier(NV50_IR_MOD_NEG);
if (s == 0) {
i->setSrc(0, i->getSrc(1));
i->src(0).mod = i->src(1).mod;
}
i->setSrc(1, i->getSrc(2));
i->src(1).mod = i->src(2).mod;
i->setSrc(2, NULL);
i->op = OP_ADD;
} else
if (s == 1 && !imm0.isNegative() && imm0.isPow2() &&
!isFloatType(i->dType) &&
target->isOpSupported(OP_SHLADD, i->dType)) {
i->op = OP_SHLADD;
imm0.applyLog2();
i->setSrc(1, new_ImmediateValue(prog, imm0.reg.data.u32));
}
break;
case OP_SUB:
if (imm0.isInteger(0) && s == 0 && typeSizeof(i->dType) == 8 &&
!isFloatType(i->dType))
break;
/* fallthrough */
case OP_ADD:
if (i->usesFlags())
break;
if (imm0.isInteger(0)) {
if (s == 0) {
i->setSrc(0, i->getSrc(1));
i->src(0).mod = i->src(1).mod;
if (i->op == OP_SUB)
i->src(0).mod = i->src(0).mod ^ Modifier(NV50_IR_MOD_NEG);
}
i->setSrc(1, NULL);
i->op = i->src(0).mod.getOp();
if (i->op != OP_CVT)
i->src(0).mod = Modifier(0);
}
break;
case OP_DIV:
if (s != 1 || (i->dType != TYPE_S32 && i->dType != TYPE_U32))
break;
bld.setPosition(i, false);
if (imm0.reg.data.u32 == 0) {
break;
} else
if (imm0.reg.data.u32 == 1) {
i->op = OP_MOV;
i->setSrc(1, NULL);
} else
if (i->dType == TYPE_U32 && imm0.isPow2()) {
i->op = OP_SHR;
i->setSrc(1, bld.mkImm(util_logbase2(imm0.reg.data.u32)));
} else
if (i->dType == TYPE_U32) {
Instruction *mul;
Value *tA, *tB;
const uint32_t d = imm0.reg.data.u32;
uint32_t m;
int r, s;
uint32_t l = util_logbase2(d);
if (((uint32_t)1 << l) < d)
++l;
m = (((uint64_t)1 << 32) * (((uint64_t)1 << l) - d)) / d + 1;
r = l ? 1 : 0;
s = l ? (l - 1) : 0;
tA = bld.getSSA();
tB = bld.getSSA();
mul = bld.mkOp2(OP_MUL, TYPE_U32, tA, i->getSrc(0),
bld.loadImm(NULL, m));
mul->subOp = NV50_IR_SUBOP_MUL_HIGH;
bld.mkOp2(OP_SUB, TYPE_U32, tB, i->getSrc(0), tA);
tA = bld.getSSA();
if (r)
bld.mkOp2(OP_SHR, TYPE_U32, tA, tB, bld.mkImm(r));
else
tA = tB;
tB = s ? bld.getSSA() : i->getDef(0);
newi = bld.mkOp2(OP_ADD, TYPE_U32, tB, mul->getDef(0), tA);
if (s)
bld.mkOp2(OP_SHR, TYPE_U32, i->getDef(0), tB, bld.mkImm(s));
delete_Instruction(prog, i);
} else
if (imm0.reg.data.s32 == -1) {
i->op = OP_NEG;
i->setSrc(1, NULL);
} else {
LValue *tA, *tB;
LValue *tD;
const int32_t d = imm0.reg.data.s32;
int32_t m;
int32_t l = util_logbase2(static_cast<unsigned>(abs(d)));
if ((1 << l) < abs(d))
++l;
if (!l)
l = 1;
m = ((uint64_t)1 << (32 + l - 1)) / abs(d) + 1 - ((uint64_t)1 << 32);
tA = bld.getSSA();
tB = bld.getSSA();
bld.mkOp3(OP_MAD, TYPE_S32, tA, i->getSrc(0), bld.loadImm(NULL, m),
i->getSrc(0))->subOp = NV50_IR_SUBOP_MUL_HIGH;
if (l > 1)
bld.mkOp2(OP_SHR, TYPE_S32, tB, tA, bld.mkImm(l - 1));
else
tB = tA;
tA = bld.getSSA();
bld.mkCmp(OP_SET, CC_LT, TYPE_S32, tA, TYPE_S32, i->getSrc(0), bld.mkImm(0));
tD = (d < 0) ? bld.getSSA() : i->getDef(0)->asLValue();
newi = bld.mkOp2(OP_SUB, TYPE_U32, tD, tB, tA);
if (d < 0)
bld.mkOp1(OP_NEG, TYPE_S32, i->getDef(0), tB);
delete_Instruction(prog, i);
}
break;
case OP_MOD:
if (s == 1 && imm0.isPow2()) {
bld.setPosition(i, false);
if (i->sType == TYPE_U32) {
i->op = OP_AND;
i->setSrc(1, bld.loadImm(NULL, imm0.reg.data.u32 - 1));
} else if (i->sType == TYPE_S32) {
// Do it on the absolute value of the input, and then restore the
// sign. The only odd case is MIN_INT, but that should work out
// as well, since MIN_INT mod any power of 2 is 0.
//
// Technically we don't have to do any of this since MOD is
// undefined with negative arguments in GLSL, but this seems like
// the nice thing to do.
Value *abs = bld.mkOp1v(OP_ABS, TYPE_S32, bld.getSSA(), i->getSrc(0));
Value *neg, *v1, *v2;
bld.mkCmp(OP_SET, CC_LT, TYPE_S32,
(neg = bld.getSSA(1, prog->getTarget()->nativeFile(FILE_PREDICATE))),
TYPE_S32, i->getSrc(0), bld.loadImm(NULL, 0));
Value *mod = bld.mkOp2v(OP_AND, TYPE_U32, bld.getSSA(), abs,
bld.loadImm(NULL, imm0.reg.data.u32 - 1));
bld.mkOp1(OP_NEG, TYPE_S32, (v1 = bld.getSSA()), mod)
->setPredicate(CC_P, neg);
bld.mkOp1(OP_MOV, TYPE_S32, (v2 = bld.getSSA()), mod)
->setPredicate(CC_NOT_P, neg);
newi = bld.mkOp2(OP_UNION, TYPE_S32, i->getDef(0), v1, v2);
delete_Instruction(prog, i);
}
} else if (s == 1) {
// In this case, we still want the optimized lowering that we get
// from having division by an immediate.
//
// a % b == a - (a/b) * b
bld.setPosition(i, false);
Value *div = bld.mkOp2v(OP_DIV, i->sType, bld.getSSA(),
i->getSrc(0), i->getSrc(1));
newi = bld.mkOp2(OP_ADD, i->sType, i->getDef(0), i->getSrc(0),
bld.mkOp2v(OP_MUL, i->sType, bld.getSSA(), div, i->getSrc(1)));
// TODO: Check that target supports this. In this case, we know that
// all backends do.
newi->src(1).mod = Modifier(NV50_IR_MOD_NEG);
delete_Instruction(prog, i);
}
break;
case OP_SET: // TODO: SET_AND,OR,XOR
{
/* This optimizes the case where the output of a set is being compared
* to zero. Since the set can only produce 0/-1 (int) or 0/1 (float), we
* can be a lot cleverer in our comparison.
*/
CmpInstruction *si = findOriginForTestWithZero(i->getSrc(t));
CondCode cc, ccZ;
if (imm0.reg.data.u32 != 0 || !si)
return;
cc = si->setCond;
ccZ = (CondCode)((unsigned int)i->asCmp()->setCond & ~CC_U);
// We do everything assuming var (cmp) 0, reverse the condition if 0 is
// first.
if (s == 0)
ccZ = reverseCondCode(ccZ);
// If there is a negative modifier, we need to undo that, by flipping
// the comparison to zero.
if (i->src(t).mod.neg())
ccZ = reverseCondCode(ccZ);
// If this is a signed comparison, we expect the input to be a regular
// boolean, i.e. 0/-1. However the rest of the logic assumes that true
// is positive, so just flip the sign.
if (i->sType == TYPE_S32) {
assert(!isFloatType(si->dType));
ccZ = reverseCondCode(ccZ);
}
switch (ccZ) {
case CC_LT: cc = CC_FL; break; // bool < 0 -- this is never true
case CC_GE: cc = CC_TR; break; // bool >= 0 -- this is always true
case CC_EQ: cc = inverseCondCode(cc); break; // bool == 0 -- !bool
case CC_LE: cc = inverseCondCode(cc); break; // bool <= 0 -- !bool
case CC_GT: break; // bool > 0 -- bool
case CC_NE: break; // bool != 0 -- bool
default:
return;
}
// Update the condition of this SET to be identical to the origin set,
// but with the updated condition code. The original SET should get
// DCE'd, ideally.
i->op = si->op;
i->asCmp()->setCond = cc;
i->setSrc(0, si->src(0));
i->setSrc(1, si->src(1));
if (si->srcExists(2))
i->setSrc(2, si->src(2));
i->sType = si->sType;
}
break;
case OP_AND:
{
Instruction *src = i->getSrc(t)->getInsn();
ImmediateValue imm1;
if (imm0.reg.data.u32 == 0) {
i->op = OP_MOV;
i->setSrc(0, new_ImmediateValue(prog, 0u));
i->src(0).mod = Modifier(0);
i->setSrc(1, NULL);
} else if (imm0.reg.data.u32 == ~0U) {
i->op = i->src(t).mod.getOp();
if (t) {
i->setSrc(0, i->getSrc(t));
i->src(0).mod = i->src(t).mod;
}
i->setSrc(1, NULL);
} else if (src->asCmp()) {
CmpInstruction *cmp = src->asCmp();
if (!cmp || cmp->op == OP_SLCT || cmp->getDef(0)->refCount() > 1)
return;
if (!prog->getTarget()->isOpSupported(cmp->op, TYPE_F32))
return;
if (imm0.reg.data.f32 != 1.0)
return;
if (cmp->dType != TYPE_U32)
return;
cmp->dType = TYPE_F32;
if (i->src(t).mod != Modifier(0)) {
assert(i->src(t).mod == Modifier(NV50_IR_MOD_NOT));
i->src(t).mod = Modifier(0);
cmp->setCond = inverseCondCode(cmp->setCond);
}
i->op = OP_MOV;
i->setSrc(s, NULL);
if (t) {
i->setSrc(0, i->getSrc(t));
i->setSrc(t, NULL);
}
} else if (prog->getTarget()->isOpSupported(OP_EXTBF, TYPE_U32) &&
src->op == OP_SHR &&
src->src(1).getImmediate(imm1) &&
i->src(t).mod == Modifier(0) &&
util_is_power_of_two(imm0.reg.data.u32 + 1)) {
// low byte = offset, high byte = width
uint32_t ext = (util_last_bit(imm0.reg.data.u32) << 8) | imm1.reg.data.u32;
i->op = OP_EXTBF;
i->setSrc(0, src->getSrc(0));
i->setSrc(1, new_ImmediateValue(prog, ext));
} else if (src->op == OP_SHL &&
src->src(1).getImmediate(imm1) &&
i->src(t).mod == Modifier(0) &&
util_is_power_of_two(~imm0.reg.data.u32 + 1) &&
util_last_bit(~imm0.reg.data.u32) <= imm1.reg.data.u32) {
i->op = OP_MOV;
i->setSrc(s, NULL);
if (t) {
i->setSrc(0, i->getSrc(t));
i->setSrc(t, NULL);
}
}
}
break;
case OP_SHL:
{
if (s != 1 || i->src(0).mod != Modifier(0))
break;
// try to concatenate shifts
Instruction *si = i->getSrc(0)->getInsn();
if (!si)
break;
ImmediateValue imm1;
switch (si->op) {
case OP_SHL:
if (si->src(1).getImmediate(imm1)) {
bld.setPosition(i, false);
i->setSrc(0, si->getSrc(0));
i->setSrc(1, bld.loadImm(NULL, imm0.reg.data.u32 + imm1.reg.data.u32));
}
break;
case OP_SHR:
if (si->src(1).getImmediate(imm1) && imm0.reg.data.u32 == imm1.reg.data.u32) {
bld.setPosition(i, false);
i->op = OP_AND;
i->setSrc(0, si->getSrc(0));
i->setSrc(1, bld.loadImm(NULL, ~((1 << imm0.reg.data.u32) - 1)));
}
break;
case OP_MUL:
int muls;
if (isFloatType(si->dType))
return;
if (si->src(1).getImmediate(imm1))
muls = 1;
else if (si->src(0).getImmediate(imm1))
muls = 0;
else
return;
bld.setPosition(i, false);
i->op = OP_MUL;
i->setSrc(0, si->getSrc(!muls));
i->setSrc(1, bld.loadImm(NULL, imm1.reg.data.u32 << imm0.reg.data.u32));
break;
case OP_SUB:
case OP_ADD:
int adds;
if (isFloatType(si->dType))
return;
if (si->op != OP_SUB && si->src(0).getImmediate(imm1))
adds = 0;
else if (si->src(1).getImmediate(imm1))
adds = 1;
else
return;
if (si->src(!adds).mod != Modifier(0))
return;
// SHL(ADD(x, y), z) = ADD(SHL(x, z), SHL(y, z))
// This is more operations, but if one of x, y is an immediate, then
// we can get a situation where (a) we can use ISCADD, or (b)
// propagate the add bit into an indirect load.
bld.setPosition(i, false);
i->op = si->op;
i->setSrc(adds, bld.loadImm(NULL, imm1.reg.data.u32 << imm0.reg.data.u32));
i->setSrc(!adds, bld.mkOp2v(OP_SHL, i->dType,
bld.getSSA(i->def(0).getSize(), i->def(0).getFile()),
si->getSrc(!adds),
bld.mkImm(imm0.reg.data.u32)));
break;
default:
return;
}
}
break;
case OP_ABS:
case OP_NEG:
case OP_SAT:
case OP_LG2:
case OP_RCP:
case OP_SQRT:
case OP_RSQ:
case OP_PRESIN:
case OP_SIN:
case OP_COS:
case OP_PREEX2:
case OP_EX2:
unary(i, imm0);
break;
case OP_BFIND: {
int32_t res;
switch (i->dType) {
case TYPE_S32: res = util_last_bit_signed(imm0.reg.data.s32) - 1; break;
case TYPE_U32: res = util_last_bit(imm0.reg.data.u32) - 1; break;
default:
return;
}
if (i->subOp == NV50_IR_SUBOP_BFIND_SAMT && res >= 0)
res = 31 - res;
bld.setPosition(i, false); /* make sure bld is init'ed */
i->setSrc(0, bld.mkImm(res));
i->setSrc(1, NULL);
i->op = OP_MOV;
i->subOp = 0;
break;
}
case OP_POPCNT: {
// Only deal with 1-arg POPCNT here
if (i->srcExists(1))
break;
uint32_t res = util_bitcount(imm0.reg.data.u32);
i->setSrc(0, new_ImmediateValue(i->bb->getProgram(), res));
i->setSrc(1, NULL);
i->op = OP_MOV;
break;
}
case OP_CVT: {
Storage res;
// TODO: handle 64-bit values properly
if (typeSizeof(i->dType) == 8 || typeSizeof(i->sType) == 8)
return;
// TODO: handle single byte/word extractions
if (i->subOp)
return;
bld.setPosition(i, true); /* make sure bld is init'ed */
#define CASE(type, dst, fmin, fmax, imin, imax, umin, umax) \
case type: \
switch (i->sType) { \
case TYPE_F64: \
res.data.dst = util_iround(i->saturate ? \
CLAMP(imm0.reg.data.f64, fmin, fmax) : \
imm0.reg.data.f64); \
break; \
case TYPE_F32: \
res.data.dst = util_iround(i->saturate ? \
CLAMP(imm0.reg.data.f32, fmin, fmax) : \
imm0.reg.data.f32); \
break; \
case TYPE_S32: \
res.data.dst = i->saturate ? \
CLAMP(imm0.reg.data.s32, imin, imax) : \
imm0.reg.data.s32; \
break; \
case TYPE_U32: \
res.data.dst = i->saturate ? \
CLAMP(imm0.reg.data.u32, umin, umax) : \
imm0.reg.data.u32; \
break; \
case TYPE_S16: \
res.data.dst = i->saturate ? \
CLAMP(imm0.reg.data.s16, imin, imax) : \
imm0.reg.data.s16; \
break; \
case TYPE_U16: \
res.data.dst = i->saturate ? \
CLAMP(imm0.reg.data.u16, umin, umax) : \
imm0.reg.data.u16; \
break; \
default: return; \
} \
i->setSrc(0, bld.mkImm(res.data.dst)); \
break
switch(i->dType) {
CASE(TYPE_U16, u16, 0, UINT16_MAX, 0, UINT16_MAX, 0, UINT16_MAX);
CASE(TYPE_S16, s16, INT16_MIN, INT16_MAX, INT16_MIN, INT16_MAX, 0, INT16_MAX);
CASE(TYPE_U32, u32, 0, UINT32_MAX, 0, INT32_MAX, 0, UINT32_MAX);
CASE(TYPE_S32, s32, INT32_MIN, INT32_MAX, INT32_MIN, INT32_MAX, 0, INT32_MAX);
case TYPE_F32:
switch (i->sType) {
case TYPE_F64:
res.data.f32 = i->saturate ?
CLAMP(imm0.reg.data.f64, 0.0f, 1.0f) :
imm0.reg.data.f64;
break;
case TYPE_F32:
res.data.f32 = i->saturate ?
CLAMP(imm0.reg.data.f32, 0.0f, 1.0f) :
imm0.reg.data.f32;
break;
case TYPE_U16: res.data.f32 = (float) imm0.reg.data.u16; break;
case TYPE_U32: res.data.f32 = (float) imm0.reg.data.u32; break;
case TYPE_S16: res.data.f32 = (float) imm0.reg.data.s16; break;
case TYPE_S32: res.data.f32 = (float) imm0.reg.data.s32; break;
default:
return;
}
i->setSrc(0, bld.mkImm(res.data.f32));
break;
case TYPE_F64:
switch (i->sType) {
case TYPE_F64:
res.data.f64 = i->saturate ?
CLAMP(imm0.reg.data.f64, 0.0f, 1.0f) :
imm0.reg.data.f64;
break;
case TYPE_F32:
res.data.f64 = i->saturate ?
CLAMP(imm0.reg.data.f32, 0.0f, 1.0f) :
imm0.reg.data.f32;
break;
case TYPE_U16: res.data.f64 = (double) imm0.reg.data.u16; break;
case TYPE_U32: res.data.f64 = (double) imm0.reg.data.u32; break;
case TYPE_S16: res.data.f64 = (double) imm0.reg.data.s16; break;
case TYPE_S32: res.data.f64 = (double) imm0.reg.data.s32; break;
default:
return;
}
i->setSrc(0, bld.mkImm(res.data.f64));
break;
default:
return;
}
#undef CASE
i->setType(i->dType); /* Remove i->sType, which we don't need anymore */
i->op = OP_MOV;
i->saturate = 0;
i->src(0).mod = Modifier(0); /* Clear the already applied modifier */
break;
}
default:
return;
}
// This can get left behind some of the optimizations which simplify
// saturatable values.
if (newi->op == OP_MOV && newi->saturate) {
ImmediateValue tmp;
newi->saturate = 0;
newi->op = OP_SAT;
if (newi->src(0).getImmediate(tmp))
unary(newi, tmp);
}
if (newi->op != op)
foldCount++;
}
// =============================================================================
// Merge modifier operations (ABS, NEG, NOT) into ValueRefs where allowed.
class ModifierFolding : public Pass
{
private:
virtual bool visit(BasicBlock *);
};
bool
ModifierFolding::visit(BasicBlock *bb)
{
const Target *target = prog->getTarget();
Instruction *i, *next, *mi;
Modifier mod;
for (i = bb->getEntry(); i; i = next) {
next = i->next;
if (0 && i->op == OP_SUB) {
// turn "sub" into "add neg" (do we really want this ?)
i->op = OP_ADD;
i->src(0).mod = i->src(0).mod ^ Modifier(NV50_IR_MOD_NEG);
}
for (int s = 0; s < 3 && i->srcExists(s); ++s) {
mi = i->getSrc(s)->getInsn();
if (!mi ||
mi->predSrc >= 0 || mi->getDef(0)->refCount() > 8)
continue;
if (i->sType == TYPE_U32 && mi->dType == TYPE_S32) {
if ((i->op != OP_ADD &&
i->op != OP_MUL) ||
(mi->op != OP_ABS &&
mi->op != OP_NEG))
continue;
} else
if (i->sType != mi->dType) {
continue;
}
if ((mod = Modifier(mi->op)) == Modifier(0))
continue;
mod *= mi->src(0).mod;
if ((i->op == OP_ABS) || i->src(s).mod.abs()) {
// abs neg [abs] = abs
mod = mod & Modifier(~(NV50_IR_MOD_NEG | NV50_IR_MOD_ABS));
} else
if ((i->op == OP_NEG) && mod.neg()) {
assert(s == 0);
// neg as both opcode and modifier on same insn is prohibited
// neg neg abs = abs, neg neg = identity
mod = mod & Modifier(~NV50_IR_MOD_NEG);
i->op = mod.getOp();
mod = mod & Modifier(~NV50_IR_MOD_ABS);
if (mod == Modifier(0))
i->op = OP_MOV;
}
if (target->isModSupported(i, s, mod)) {
i->setSrc(s, mi->getSrc(0));
i->src(s).mod *= mod;
}
}
if (i->op == OP_SAT) {
mi = i->getSrc(0)->getInsn();
if (mi &&
mi->getDef(0)->refCount() <= 1 && target->isSatSupported(mi)) {
mi->saturate = 1;
mi->setDef(0, i->getDef(0));
delete_Instruction(prog, i);
}
}
}
return true;
}
// =============================================================================
// MUL + ADD -> MAD/FMA
// MIN/MAX(a, a) -> a, etc.
// SLCT(a, b, const) -> cc(const) ? a : b
// RCP(RCP(a)) -> a
// MUL(MUL(a, b), const) -> MUL_Xconst(a, b)
class AlgebraicOpt : public Pass
{
private:
virtual bool visit(BasicBlock *);
void handleABS(Instruction *);
bool handleADD(Instruction *);
bool tryADDToMADOrSAD(Instruction *, operation toOp);
void handleMINMAX(Instruction *);
void handleRCP(Instruction *);
void handleSLCT(Instruction *);
void handleLOGOP(Instruction *);
void handleCVT_NEG(Instruction *);
void handleCVT_CVT(Instruction *);
void handleCVT_EXTBF(Instruction *);
void handleSUCLAMP(Instruction *);
void handleNEG(Instruction *);
BuildUtil bld;
};
void
AlgebraicOpt::handleABS(Instruction *abs)
{
Instruction *sub = abs->getSrc(0)->getInsn();
DataType ty;
if (!sub ||
!prog->getTarget()->isOpSupported(OP_SAD, abs->dType))
return;
// expect not to have mods yet, if we do, bail
if (sub->src(0).mod || sub->src(1).mod)
return;
// hidden conversion ?
ty = intTypeToSigned(sub->dType);
if (abs->dType != abs->sType || ty != abs->sType)
return;
if ((sub->op != OP_ADD && sub->op != OP_SUB) ||
sub->src(0).getFile() != FILE_GPR || sub->src(0).mod ||
sub->src(1).getFile() != FILE_GPR || sub->src(1).mod)
return;
Value *src0 = sub->getSrc(0);
Value *src1 = sub->getSrc(1);
if (sub->op == OP_ADD) {
Instruction *neg = sub->getSrc(1)->getInsn();
if (neg && neg->op != OP_NEG) {
neg = sub->getSrc(0)->getInsn();
src0 = sub->getSrc(1);
}
if (!neg || neg->op != OP_NEG ||
neg->dType != neg->sType || neg->sType != ty)
return;
src1 = neg->getSrc(0);
}
// found ABS(SUB))
abs->moveSources(1, 2); // move sources >=1 up by 2
abs->op = OP_SAD;
abs->setType(sub->dType);
abs->setSrc(0, src0);
abs->setSrc(1, src1);
bld.setPosition(abs, false);
abs->setSrc(2, bld.loadImm(bld.getSSA(typeSizeof(ty)), 0));
}
bool
AlgebraicOpt::handleADD(Instruction *add)
{
Value *src0 = add->getSrc(0);
Value *src1 = add->getSrc(1);
if (src0->reg.file != FILE_GPR || src1->reg.file != FILE_GPR)
return false;
bool changed = false;
// we can't optimize to MAD if the add is precise
if (!add->precise && prog->getTarget()->isOpSupported(OP_MAD, add->dType))
changed = tryADDToMADOrSAD(add, OP_MAD);
if (!changed && prog->getTarget()->isOpSupported(OP_SAD, add->dType))
changed = tryADDToMADOrSAD(add, OP_SAD);
return changed;
}
// ADD(SAD(a,b,0), c) -> SAD(a,b,c)
// ADD(MUL(a,b), c) -> MAD(a,b,c)
bool
AlgebraicOpt::tryADDToMADOrSAD(Instruction *add, operation toOp)
{
Value *src0 = add->getSrc(0);
Value *src1 = add->getSrc(1);
Value *src;
int s;
const operation srcOp = toOp == OP_SAD ? OP_SAD : OP_MUL;
const Modifier modBad = Modifier(~((toOp == OP_MAD) ? NV50_IR_MOD_NEG : 0));
Modifier mod[4];
if (src0->refCount() == 1 &&
src0->getUniqueInsn() && src0->getUniqueInsn()->op == srcOp)
s = 0;
else
if (src1->refCount() == 1 &&
src1->getUniqueInsn() && src1->getUniqueInsn()->op == srcOp)
s = 1;
else
return false;
src = add->getSrc(s);
if (src->getUniqueInsn() && src->getUniqueInsn()->bb != add->bb)
return false;
if (src->getInsn()->saturate || src->getInsn()->postFactor ||
src->getInsn()->dnz || src->getInsn()->precise)
return false;
if (toOp == OP_SAD) {
ImmediateValue imm;
if (!src->getInsn()->src(2).getImmediate(imm))
return false;
if (!imm.isInteger(0))
return false;
}
if (typeSizeof(add->dType) != typeSizeof(src->getInsn()->dType) ||
isFloatType(add->dType) != isFloatType(src->getInsn()->dType))
return false;
mod[0] = add->src(0).mod;
mod[1] = add->src(1).mod;
mod[2] = src->getUniqueInsn()->src(0).mod;
mod[3] = src->getUniqueInsn()->src(1).mod;
if (((mod[0] | mod[1]) | (mod[2] | mod[3])) & modBad)
return false;
add->op = toOp;
add->subOp = src->getInsn()->subOp; // potentially mul-high
add->dnz = src->getInsn()->dnz;
add->dType = src->getInsn()->dType; // sign matters for imad hi
add->sType = src->getInsn()->sType;
add->setSrc(2, add->src(s ? 0 : 1));
add->setSrc(0, src->getInsn()->getSrc(0));
add->src(0).mod = mod[2] ^ mod[s];
add->setSrc(1, src->getInsn()->getSrc(1));
add->src(1).mod = mod[3];
return true;
}
void
AlgebraicOpt::handleMINMAX(Instruction *minmax)
{
Value *src0 = minmax->getSrc(0);
Value *src1 = minmax->getSrc(1);
if (src0 != src1 || src0->reg.file != FILE_GPR)
return;
if (minmax->src(0).mod == minmax->src(1).mod) {
if (minmax->def(0).mayReplace(minmax->src(0))) {
minmax->def(0).replace(minmax->src(0), false);
minmax->bb->remove(minmax);
} else {
minmax->op = OP_CVT;
minmax->setSrc(1, NULL);
}
} else {
// TODO:
// min(x, -x) = -abs(x)
// min(x, -abs(x)) = -abs(x)
// min(x, abs(x)) = x
// max(x, -abs(x)) = x
// max(x, abs(x)) = abs(x)
// max(x, -x) = abs(x)
}
}
void
AlgebraicOpt::handleRCP(Instruction *rcp)
{
Instruction *si = rcp->getSrc(0)->getUniqueInsn();
if (si && si->op == OP_RCP) {
Modifier mod = rcp->src(0).mod * si->src(0).mod;
rcp->op = mod.getOp();
rcp->setSrc(0, si->getSrc(0));
}
}
void
AlgebraicOpt::handleSLCT(Instruction *slct)
{
if (slct->getSrc(2)->reg.file == FILE_IMMEDIATE) {
if (slct->getSrc(2)->asImm()->compare(slct->asCmp()->setCond, 0.0f))
slct->setSrc(0, slct->getSrc(1));
} else
if (slct->getSrc(0) != slct->getSrc(1)) {
return;
}
slct->op = OP_MOV;
slct->setSrc(1, NULL);
slct->setSrc(2, NULL);
}
void
AlgebraicOpt::handleLOGOP(Instruction *logop)
{
Value *src0 = logop->getSrc(0);
Value *src1 = logop->getSrc(1);
if (src0->reg.file != FILE_GPR || src1->reg.file != FILE_GPR)
return;
if (src0 == src1) {
if ((logop->op == OP_AND || logop->op == OP_OR) &&
logop->def(0).mayReplace(logop->src(0))) {
logop->def(0).replace(logop->src(0), false);
delete_Instruction(prog, logop);
}
} else {
// try AND(SET, SET) -> SET_AND(SET)
Instruction *set0 = src0->getInsn();
Instruction *set1 = src1->getInsn();
if (!set0 || set0->fixed || !set1 || set1->fixed)
return;
if (set1->op != OP_SET) {
Instruction *xchg = set0;
set0 = set1;
set1 = xchg;
if (set1->op != OP_SET)
return;
}
operation redOp = (logop->op == OP_AND ? OP_SET_AND :
logop->op == OP_XOR ? OP_SET_XOR : OP_SET_OR);
if (!prog->getTarget()->isOpSupported(redOp, set1->sType))
return;
if (set0->op != OP_SET &&
set0->op != OP_SET_AND &&
set0->op != OP_SET_OR &&
set0->op != OP_SET_XOR)
return;
if (set0->getDef(0)->refCount() > 1 &&
set1->getDef(0)->refCount() > 1)
return;
if (set0->getPredicate() || set1->getPredicate())
return;
// check that they don't source each other
for (int s = 0; s < 2; ++s)
if (set0->getSrc(s) == set1->getDef(0) ||
set1->getSrc(s) == set0->getDef(0))
return;
set0 = cloneForward(func, set0);
set1 = cloneShallow(func, set1);
logop->bb->insertAfter(logop, set1);
logop->bb->insertAfter(logop, set0);
set0->dType = TYPE_U8;
set0->getDef(0)->reg.file = FILE_PREDICATE;
set0->getDef(0)->reg.size = 1;
set1->setSrc(2, set0->getDef(0));
set1->op = redOp;
set1->setDef(0, logop->getDef(0));
delete_Instruction(prog, logop);
}
}
// F2I(NEG(SET with result 1.0f/0.0f)) -> SET with result -1/0
// nv50:
// F2I(NEG(I2F(ABS(SET))))
void
AlgebraicOpt::handleCVT_NEG(Instruction *cvt)
{
Instruction *insn = cvt->getSrc(0)->getInsn();
if (cvt->sType != TYPE_F32 ||
cvt->dType != TYPE_S32 || cvt->src(0).mod != Modifier(0))
return;
if (!insn || insn->op != OP_NEG || insn->dType != TYPE_F32)
return;
if (insn->src(0).mod != Modifier(0))
return;
insn = insn->getSrc(0)->getInsn();
// check for nv50 SET(-1,0) -> SET(1.0f/0.0f) chain and nvc0's f32 SET
if (insn && insn->op == OP_CVT &&
insn->dType == TYPE_F32 &&
insn->sType == TYPE_S32) {
insn = insn->getSrc(0)->getInsn();
if (!insn || insn->op != OP_ABS || insn->sType != TYPE_S32 ||
insn->src(0).mod)
return;
insn = insn->getSrc(0)->getInsn();
if (!insn || insn->op != OP_SET || insn->dType != TYPE_U32)
return;
} else
if (!insn || insn->op != OP_SET || insn->dType != TYPE_F32) {
return;
}
Instruction *bset = cloneShallow(func, insn);
bset->dType = TYPE_U32;
bset->setDef(0, cvt->getDef(0));
cvt->bb->insertAfter(cvt, bset);
delete_Instruction(prog, cvt);
}
// F2I(TRUNC()) and so on can be expressed as a single CVT. If the earlier CVT
// does a type conversion, this becomes trickier as there might be range
// changes/etc. We could handle those in theory as long as the range was being
// reduced or kept the same.
void
AlgebraicOpt::handleCVT_CVT(Instruction *cvt)
{
Instruction *insn = cvt->getSrc(0)->getInsn();
RoundMode rnd = insn->rnd;
if (insn->saturate ||
insn->subOp ||
insn->dType != insn->sType ||
insn->dType != cvt->sType)
return;
switch (insn->op) {
case OP_CEIL:
rnd = ROUND_PI;
break;
case OP_FLOOR:
rnd = ROUND_MI;
break;
case OP_TRUNC:
rnd = ROUND_ZI;
break;
case OP_CVT:
break;
default:
return;
}
if (!isFloatType(cvt->dType) || !isFloatType(insn->sType))
rnd = (RoundMode)(rnd & 3);
cvt->rnd = rnd;
cvt->setSrc(0, insn->getSrc(0));
cvt->src(0).mod *= insn->src(0).mod;
cvt->sType = insn->sType;
}
// Some shaders extract packed bytes out of words and convert them to
// e.g. float. The Fermi+ CVT instruction can extract those directly, as can
// nv50 for word sizes.
//
// CVT(EXTBF(x, byte/word))
// CVT(AND(bytemask, x))
// CVT(AND(bytemask, SHR(x, 8/16/24)))
// CVT(SHR(x, 16/24))
void
AlgebraicOpt::handleCVT_EXTBF(Instruction *cvt)
{
Instruction *insn = cvt->getSrc(0)->getInsn();
ImmediateValue imm;
Value *arg = NULL;
unsigned width, offset;
if ((cvt->sType != TYPE_U32 && cvt->sType != TYPE_S32) || !insn)
return;
if (insn->op == OP_EXTBF && insn->src(1).getImmediate(imm)) {
width = (imm.reg.data.u32 >> 8) & 0xff;
offset = imm.reg.data.u32 & 0xff;
arg = insn->getSrc(0);
if (width != 8 && width != 16)
return;
if (width == 8 && offset & 0x7)
return;
if (width == 16 && offset & 0xf)
return;
} else if (insn->op == OP_AND) {
int s;
if (insn->src(0).getImmediate(imm))
s = 0;
else if (insn->src(1).getImmediate(imm))
s = 1;
else
return;
if (imm.reg.data.u32 == 0xff)
width = 8;
else if (imm.reg.data.u32 == 0xffff)
width = 16;
else
return;
arg = insn->getSrc(!s);
Instruction *shift = arg->getInsn();
offset = 0;
if (shift && shift->op == OP_SHR &&
shift->sType == cvt->sType &&
shift->src(1).getImmediate(imm) &&
((width == 8 && (imm.reg.data.u32 & 0x7) == 0) ||
(width == 16 && (imm.reg.data.u32 & 0xf) == 0))) {
arg = shift->getSrc(0);
offset = imm.reg.data.u32;
}
// We just AND'd the high bits away, which means this is effectively an
// unsigned value.
cvt->sType = TYPE_U32;
} else if (insn->op == OP_SHR &&
insn->sType == cvt->sType &&
insn->src(1).getImmediate(imm)) {
arg = insn->getSrc(0);
if (imm.reg.data.u32 == 24) {
width = 8;
offset = 24;
} else if (imm.reg.data.u32 == 16) {
width = 16;
offset = 16;
} else {
return;
}
}
if (!arg)
return;
// Irrespective of what came earlier, we can undo a shift on the argument
// by adjusting the offset.
Instruction *shift = arg->getInsn();
if (shift && shift->op == OP_SHL &&
shift->src(1).getImmediate(imm) &&
((width == 8 && (imm.reg.data.u32 & 0x7) == 0) ||
(width == 16 && (imm.reg.data.u32 & 0xf) == 0)) &&
imm.reg.data.u32 <= offset) {
arg = shift->getSrc(0);
offset -= imm.reg.data.u32;
}
// The unpackSnorm lowering still leaves a few shifts behind, but it's too
// annoying to detect them.
if (width == 8) {
cvt->sType = cvt->sType == TYPE_U32 ? TYPE_U8 : TYPE_S8;
} else {
assert(width == 16);
cvt->sType = cvt->sType == TYPE_U32 ? TYPE_U16 : TYPE_S16;
}
cvt->setSrc(0, arg);
cvt->subOp = offset >> 3;
}
// SUCLAMP dst, (ADD b imm), k, 0 -> SUCLAMP dst, b, k, imm (if imm fits s6)
void
AlgebraicOpt::handleSUCLAMP(Instruction *insn)
{
ImmediateValue imm;
int32_t val = insn->getSrc(2)->asImm()->reg.data.s32;
int s;
Instruction *add;
assert(insn->srcExists(0) && insn->src(0).getFile() == FILE_GPR);
// look for ADD (TODO: only count references by non-SUCLAMP)
if (insn->getSrc(0)->refCount() > 1)
return;
add = insn->getSrc(0)->getInsn();
if (!add || add->op != OP_ADD ||
(add->dType != TYPE_U32 &&
add->dType != TYPE_S32))
return;
// look for immediate
for (s = 0; s < 2; ++s)
if (add->src(s).getImmediate(imm))
break;
if (s >= 2)
return;
s = s ? 0 : 1;
// determine if immediate fits
val += imm.reg.data.s32;
if (val > 31 || val < -32)
return;
// determine if other addend fits
if (add->src(s).getFile() != FILE_GPR || add->src(s).mod != Modifier(0))
return;
bld.setPosition(insn, false); // make sure bld is init'ed
// replace sources
insn->setSrc(2, bld.mkImm(val));
insn->setSrc(0, add->getSrc(s));
}
// NEG(AND(SET, 1)) -> SET
void
AlgebraicOpt::handleNEG(Instruction *i) {
Instruction *src = i->getSrc(0)->getInsn();
ImmediateValue imm;
int b;
if (isFloatType(i->sType) || !src || src->op != OP_AND)
return;
if (src->src(0).getImmediate(imm))
b = 1;
else if (src->src(1).getImmediate(imm))
b = 0;
else
return;
if (!imm.isInteger(1))
return;
Instruction *set = src->getSrc(b)->getInsn();
if ((set->op == OP_SET || set->op == OP_SET_AND ||
set->op == OP_SET_OR || set->op == OP_SET_XOR) &&
!isFloatType(set->dType)) {
i->def(0).replace(set->getDef(0), false);
}
}
bool
AlgebraicOpt::visit(BasicBlock *bb)
{
Instruction *next;
for (Instruction *i = bb->getEntry(); i; i = next) {
next = i->next;
switch (i->op) {
case OP_ABS:
handleABS(i);
break;
case OP_ADD:
handleADD(i);
break;
case OP_RCP:
handleRCP(i);
break;
case OP_MIN:
case OP_MAX:
handleMINMAX(i);
break;
case OP_SLCT:
handleSLCT(i);
break;
case OP_AND:
case OP_OR:
case OP_XOR:
handleLOGOP(i);
break;
case OP_CVT:
handleCVT_NEG(i);
handleCVT_CVT(i);
if (prog->getTarget()->isOpSupported(OP_EXTBF, TYPE_U32))
handleCVT_EXTBF(i);
break;
case OP_SUCLAMP:
handleSUCLAMP(i);
break;
case OP_NEG:
handleNEG(i);
break;
default:
break;
}
}
return true;
}
// =============================================================================
// ADD(SHL(a, b), c) -> SHLADD(a, b, c)
class LateAlgebraicOpt : public Pass
{
private:
virtual bool visit(Instruction *);
void handleADD(Instruction *);
bool tryADDToSHLADD(Instruction *);
};
void
LateAlgebraicOpt::handleADD(Instruction *add)
{
Value *src0 = add->getSrc(0);
Value *src1 = add->getSrc(1);
if (src0->reg.file != FILE_GPR || src1->reg.file != FILE_GPR)
return;
if (prog->getTarget()->isOpSupported(OP_SHLADD, add->dType))
tryADDToSHLADD(add);
}
// ADD(SHL(a, b), c) -> SHLADD(a, b, c)
bool
LateAlgebraicOpt::tryADDToSHLADD(Instruction *add)
{
Value *src0 = add->getSrc(0);
Value *src1 = add->getSrc(1);
ImmediateValue imm;
Instruction *shl;
Value *src;
int s;
if (add->saturate || add->usesFlags() || typeSizeof(add->dType) == 8
|| isFloatType(add->dType))
return false;
if (src0->getUniqueInsn() && src0->getUniqueInsn()->op == OP_SHL)
s = 0;
else
if (src1->getUniqueInsn() && src1->getUniqueInsn()->op == OP_SHL)
s = 1;
else
return false;
src = add->getSrc(s);
shl = src->getUniqueInsn();
if (shl->bb != add->bb || shl->usesFlags() || shl->subOp || shl->src(0).mod)
return false;
if (!shl->src(1).getImmediate(imm))
return false;
add->op = OP_SHLADD;
add->setSrc(2, add->src(!s));
// SHL can't have any modifiers, but the ADD source may have had
// one. Preserve it.
add->setSrc(0, shl->getSrc(0));
if (s == 1)
add->src(0).mod = add->src(1).mod;
add->setSrc(1, new_ImmediateValue(shl->bb->getProgram(), imm.reg.data.u32));
add->src(1).mod = Modifier(0);
return true;
}
bool
LateAlgebraicOpt::visit(Instruction *i)
{
switch (i->op) {
case OP_ADD:
handleADD(i);
break;
default:
break;
}
return true;
}
// =============================================================================
// Split 64-bit MUL and MAD
class Split64BitOpPreRA : public Pass
{
private:
virtual bool visit(BasicBlock *);
void split64MulMad(Function *, Instruction *, DataType);
BuildUtil bld;
};
bool
Split64BitOpPreRA::visit(BasicBlock *bb)
{
Instruction *i, *next;
Modifier mod;
for (i = bb->getEntry(); i; i = next) {
next = i->next;
DataType hTy;
switch (i->dType) {
case TYPE_U64: hTy = TYPE_U32; break;
case TYPE_S64: hTy = TYPE_S32; break;
default:
continue;
}
if (i->op == OP_MAD || i->op == OP_MUL)
split64MulMad(func, i, hTy);
}
return true;
}
void
Split64BitOpPreRA::split64MulMad(Function *fn, Instruction *i, DataType hTy)
{
assert(i->op == OP_MAD || i->op == OP_MUL);
assert(!isFloatType(i->dType) && !isFloatType(i->sType));
assert(typeSizeof(hTy) == 4);
bld.setPosition(i, true);
Value *zero = bld.mkImm(0u);
Value *carry = bld.getSSA(1, FILE_FLAGS);
// We want to compute `d = a * b (+ c)?`, where a, b, c and d are 64-bit
// values (a, b and c might be 32-bit values), using 32-bit operations. This
// gives the following operations:
// * `d.low = low(a.low * b.low) (+ c.low)?`
// * `d.high = low(a.high * b.low) + low(a.low * b.high)
// + high(a.low * b.low) (+ c.high)?`
//
// To compute the high bits, we can split in the following operations:
// * `tmp1 = low(a.high * b.low) (+ c.high)?`
// * `tmp2 = low(a.low * b.high) + tmp1`
// * `d.high = high(a.low * b.low) + tmp2`
//
// mkSplit put lower bits at index 0 and higher bits at index 1
Value *op1[2];
if (i->getSrc(0)->reg.size == 8)
bld.mkSplit(op1, 4, i->getSrc(0));
else {
op1[0] = i->getSrc(0);
op1[1] = zero;
}
Value *op2[2];
if (i->getSrc(1)->reg.size == 8)
bld.mkSplit(op2, 4, i->getSrc(1));
else {
op2[0] = i->getSrc(1);
op2[1] = zero;
}
Value *op3[2] = { NULL, NULL };
if (i->op == OP_MAD) {
if (i->getSrc(2)->reg.size == 8)
bld.mkSplit(op3, 4, i->getSrc(2));
else {
op3[0] = i->getSrc(2);
op3[1] = zero;
}
}
Value *tmpRes1Hi = bld.getSSA();
if (i->op == OP_MAD)
bld.mkOp3(OP_MAD, hTy, tmpRes1Hi, op1[1], op2[0], op3[1]);
else
bld.mkOp2(OP_MUL, hTy, tmpRes1Hi, op1[1], op2[0]);
Value *tmpRes2Hi = bld.mkOp3v(OP_MAD, hTy, bld.getSSA(), op1[0], op2[1], tmpRes1Hi);
Value *def[2] = { bld.getSSA(), bld.getSSA() };
// If it was a MAD, add the carry from the low bits
// It is not needed if it was a MUL, since we added high(a.low * b.low) to
// d.high
if (i->op == OP_MAD)
bld.mkOp3(OP_MAD, hTy, def[0], op1[0], op2[0], op3[0])->setFlagsDef(1, carry);
else
bld.mkOp2(OP_MUL, hTy, def[0], op1[0], op2[0]);
Instruction *hiPart3 = bld.mkOp3(OP_MAD, hTy, def[1], op1[0], op2[0], tmpRes2Hi);
hiPart3->subOp = NV50_IR_SUBOP_MUL_HIGH;
if (i->op == OP_MAD)
hiPart3->setFlagsSrc(3, carry);
bld.mkOp2(OP_MERGE, i->dType, i->getDef(0), def[0], def[1]);
delete_Instruction(fn->getProgram(), i);
}
// =============================================================================
static inline void
updateLdStOffset(Instruction *ldst, int32_t offset, Function *fn)
{
if (offset != ldst->getSrc(0)->reg.data.offset) {
if (ldst->getSrc(0)->refCount() > 1)
ldst->setSrc(0, cloneShallow(fn, ldst->getSrc(0)));
ldst->getSrc(0)->reg.data.offset = offset;
}
}
// Combine loads and stores, forward stores to loads where possible.
class MemoryOpt : public Pass
{
private:
class Record
{
public:
Record *next;
Instruction *insn;
const Value *rel[2];
const Value *base;
int32_t offset;
int8_t fileIndex;
uint8_t size;
bool locked;
Record *prev;
bool overlaps(const Instruction *ldst) const;
inline void link(Record **);
inline void unlink(Record **);
inline void set(const Instruction *ldst);
};
public:
MemoryOpt();
Record *loads[DATA_FILE_COUNT];
Record *stores[DATA_FILE_COUNT];
MemoryPool recordPool;
private:
virtual bool visit(BasicBlock *);
bool runOpt(BasicBlock *);
Record **getList(const Instruction *);
Record *findRecord(const Instruction *, bool load, bool& isAdjacent) const;
// merge @insn into load/store instruction from @rec
bool combineLd(Record *rec, Instruction *ld);
bool combineSt(Record *rec, Instruction *st);
bool replaceLdFromLd(Instruction *ld, Record *ldRec);
bool replaceLdFromSt(Instruction *ld, Record *stRec);
bool replaceStFromSt(Instruction *restrict st, Record *stRec);
void addRecord(Instruction *ldst);
void purgeRecords(Instruction *const st, DataFile);
void lockStores(Instruction *const ld);
void reset();
private:
Record *prevRecord;
};
MemoryOpt::MemoryOpt() : recordPool(sizeof(MemoryOpt::Record), 6)
{
for (int i = 0; i < DATA_FILE_COUNT; ++i) {
loads[i] = NULL;
stores[i] = NULL;
}
prevRecord = NULL;
}
void
MemoryOpt::reset()
{
for (unsigned int i = 0; i < DATA_FILE_COUNT; ++i) {
Record *it, *next;
for (it = loads[i]; it; it = next) {
next = it->next;
recordPool.release(it);
}
loads[i] = NULL;
for (it = stores[i]; it; it = next) {
next = it->next;
recordPool.release(it);
}
stores[i] = NULL;
}
}
bool
MemoryOpt::combineLd(Record *rec, Instruction *ld)
{
int32_t offRc = rec->offset;
int32_t offLd = ld->getSrc(0)->reg.data.offset;
int sizeRc = rec->size;
int sizeLd = typeSizeof(ld->dType);
int size = sizeRc + sizeLd;
int d, j;
if (!prog->getTarget()->
isAccessSupported(ld->getSrc(0)->reg.file, typeOfSize(size)))
return false;
// no unaligned loads
if (((size == 0x8) && (MIN2(offLd, offRc) & 0x7)) ||
((size == 0xc) && (MIN2(offLd, offRc) & 0xf)))
return false;
// for compute indirect loads are not guaranteed to be aligned
if (prog->getType() == Program::TYPE_COMPUTE && rec->rel[0])
return false;
assert(sizeRc + sizeLd <= 16 && offRc != offLd);
// lock any stores that overlap with the load being merged into the
// existing record.
lockStores(ld);
for (j = 0; sizeRc; sizeRc -= rec->insn->getDef(j)->reg.size, ++j);
if (offLd < offRc) {
int sz;
for (sz = 0, d = 0; sz < sizeLd; sz += ld->getDef(d)->reg.size, ++d);
// d: nr of definitions in ld
// j: nr of definitions in rec->insn, move:
for (d = d + j - 1; j > 0; --j, --d)
rec->insn->setDef(d, rec->insn->getDef(j - 1));
if (rec->insn->getSrc(0)->refCount() > 1)
rec->insn->setSrc(0, cloneShallow(func, rec->insn->getSrc(0)));
rec->offset = rec->insn->getSrc(0)->reg.data.offset = offLd;
d = 0;
} else {
d = j;
}
// move definitions of @ld to @rec->insn
for (j = 0; sizeLd; ++j, ++d) {
sizeLd -= ld->getDef(j)->reg.size;
rec->insn->setDef(d, ld->getDef(j));
}
rec->size = size;
rec->insn->getSrc(0)->reg.size = size;
rec->insn->setType(typeOfSize(size));
delete_Instruction(prog, ld);
return true;
}
bool
MemoryOpt::combineSt(Record *rec, Instruction *st)
{
int32_t offRc = rec->offset;
int32_t offSt = st->getSrc(0)->reg.data.offset;
int sizeRc = rec->size;
int sizeSt = typeSizeof(st->dType);
int s = sizeSt / 4;
int size = sizeRc + sizeSt;
int j, k;
Value *src[4]; // no modifiers in ValueRef allowed for st
Value *extra[3];
if (!prog->getTarget()->
isAccessSupported(st->getSrc(0)->reg.file, typeOfSize(size)))
return false;
// no unaligned stores
if (size == 8 && MIN2(offRc, offSt) & 0x7)
return false;
// for compute indirect stores are not guaranteed to be aligned
if (prog->getType() == Program::TYPE_COMPUTE && rec->rel[0])
return false;
// remove any existing load/store records for the store being merged into
// the existing record.
purgeRecords(st, DATA_FILE_COUNT);
st->takeExtraSources(0, extra); // save predicate and indirect address
if (offRc < offSt) {
// save values from @st
for (s = 0; sizeSt; ++s) {
sizeSt -= st->getSrc(s + 1)->reg.size;
src[s] = st->getSrc(s + 1);
}
// set record's values as low sources of @st
for (j = 1; sizeRc; ++j) {
sizeRc -= rec->insn->getSrc(j)->reg.size;
st->setSrc(j, rec->insn->getSrc(j));
}
// set saved values as high sources of @st
for (k = j, j = 0; j < s; ++j)
st->setSrc(k++, src[j]);
updateLdStOffset(st, offRc, func);
} else {
for (j = 1; sizeSt; ++j)
sizeSt -= st->getSrc(j)->reg.size;
for (s = 1; sizeRc; ++j, ++s) {
sizeRc -= rec->insn->getSrc(s)->reg.size;
st->setSrc(j, rec->insn->getSrc(s));
}
rec->offset = offSt;
}
st->putExtraSources(0, extra); // restore pointer and predicate
delete_Instruction(prog, rec->insn);
rec->insn = st;
rec->size = size;
rec->insn->getSrc(0)->reg.size = size;
rec->insn->setType(typeOfSize(size));
return true;
}
void
MemoryOpt::Record::set(const Instruction *ldst)
{
const Symbol *mem = ldst->getSrc(0)->asSym();
fileIndex = mem->reg.fileIndex;
rel[0] = ldst->getIndirect(0, 0);
rel[1] = ldst->getIndirect(0, 1);
offset = mem->reg.data.offset;
base = mem->getBase();
size = typeSizeof(ldst->sType);
}
void
MemoryOpt::Record::link(Record **list)
{
next = *list;
if (next)
next->prev = this;
prev = NULL;
*list = this;
}
void
MemoryOpt::Record::unlink(Record **list)
{
if (next)
next->prev = prev;
if (prev)
prev->next = next;
else
*list = next;
}
MemoryOpt::Record **
MemoryOpt::getList(const Instruction *insn)
{
if (insn->op == OP_LOAD || insn->op == OP_VFETCH)
return &loads[insn->src(0).getFile()];
return &stores[insn->src(0).getFile()];
}
void
MemoryOpt::addRecord(Instruction *i)
{
Record **list = getList(i);
Record *it = reinterpret_cast<Record *>(recordPool.allocate());
it->link(list);
it->set(i);
it->insn = i;
it->locked = false;
}
MemoryOpt::Record *
MemoryOpt::findRecord(const Instruction *insn, bool load, bool& isAdj) const
{
const Symbol *sym = insn->getSrc(0)->asSym();
const int size = typeSizeof(insn->sType);
Record *rec = NULL;
Record *it = load ? loads[sym->reg.file] : stores[sym->reg.file];
for (; it; it = it->next) {
if (it->locked && insn->op != OP_LOAD && insn->op != OP_VFETCH)
continue;
if ((it->offset >> 4) != (sym->reg.data.offset >> 4) ||
it->rel[0] != insn->getIndirect(0, 0) ||
it->fileIndex != sym->reg.fileIndex ||
it->rel[1] != insn->getIndirect(0, 1))
continue;
if (it->offset < sym->reg.data.offset) {
if (it->offset + it->size >= sym->reg.data.offset) {
isAdj = (it->offset + it->size == sym->reg.data.offset);
if (!isAdj)
return it;
if (!(it->offset & 0x7))
rec = it;
}
} else {
isAdj = it->offset != sym->reg.data.offset;
if (size <= it->size && !isAdj)
return it;
else
if (!(sym->reg.data.offset & 0x7))
if (it->offset - size <= sym->reg.data.offset)
rec = it;
}
}
return rec;
}
bool
MemoryOpt::replaceLdFromSt(Instruction *ld, Record *rec)
{
Instruction *st = rec->insn;
int32_t offSt = rec->offset;
int32_t offLd = ld->getSrc(0)->reg.data.offset;
int d, s;
for (s = 1; offSt != offLd && st->srcExists(s); ++s)
offSt += st->getSrc(s)->reg.size;
if (offSt != offLd)
return false;
for (d = 0; ld->defExists(d) && st->srcExists(s); ++d, ++s) {
if (ld->getDef(d)->reg.size != st->getSrc(s)->reg.size)
return false;
if (st->getSrc(s)->reg.file != FILE_GPR)
return false;
ld->def(d).replace(st->src(s), false);
}
ld->bb->remove(ld);
return true;
}
bool
MemoryOpt::replaceLdFromLd(Instruction *ldE, Record *rec)
{
Instruction *ldR = rec->insn;
int32_t offR = rec->offset;
int32_t offE = ldE->getSrc(0)->reg.data.offset;
int dR, dE;
assert(offR <= offE);
for (dR = 0; offR < offE && ldR->defExists(dR); ++dR)
offR += ldR->getDef(dR)->reg.size;
if (offR != offE)
return false;
for (dE = 0; ldE->defExists(dE) && ldR->defExists(dR); ++dE, ++dR) {
if (ldE->getDef(dE)->reg.size != ldR->getDef(dR)->reg.size)
return false;
ldE->def(dE).replace(ldR->getDef(dR), false);
}
delete_Instruction(prog, ldE);
return true;
}
bool
MemoryOpt::replaceStFromSt(Instruction *restrict st, Record *rec)
{
const Instruction *const ri = rec->insn;
Value *extra[3];
int32_t offS = st->getSrc(0)->reg.data.offset;
int32_t offR = rec->offset;
int32_t endS = offS + typeSizeof(st->dType);
int32_t endR = offR + typeSizeof(ri->dType);
rec->size = MAX2(endS, endR) - MIN2(offS, offR);
st->takeExtraSources(0, extra);
if (offR < offS) {
Value *vals[10];
int s, n;
int k = 0;
// get non-replaced sources of ri
for (s = 1; offR < offS; offR += ri->getSrc(s)->reg.size, ++s)
vals[k++] = ri->getSrc(s);
n = s;
// get replaced sources of st
for (s = 1; st->srcExists(s); offS += st->getSrc(s)->reg.size, ++s)
vals[k++] = st->getSrc(s);
// skip replaced sources of ri
for (s = n; offR < endS; offR += ri->getSrc(s)->reg.size, ++s);
// get non-replaced sources after values covered by st
for (; offR < endR; offR += ri->getSrc(s)->reg.size, ++s)
vals[k++] = ri->getSrc(s);
assert((unsigned int)k <= ARRAY_SIZE(vals));
for (s = 0; s < k; ++s)
st->setSrc(s + 1, vals[s]);
st->setSrc(0, ri->getSrc(0));
} else
if (endR > endS) {
int j, s;
for (j = 1; offR < endS; offR += ri->getSrc(j++)->reg.size);
for (s = 1; offS < endS; offS += st->getSrc(s++)->reg.size);
for (; offR < endR; offR += ri->getSrc(j++)->reg.size)
st->setSrc(s++, ri->getSrc(j));
}
st->putExtraSources(0, extra);
delete_Instruction(prog, rec->insn);
rec->insn = st;
rec->offset = st->getSrc(0)->reg.data.offset;
st->setType(typeOfSize(rec->size));
return true;
}
bool
MemoryOpt::Record::overlaps(const Instruction *ldst) const
{
Record that;
that.set(ldst);
// This assumes that images/buffers can't overlap. They can.
// TODO: Plumb the restrict logic through, and only skip when it's a
// restrict situation, or there can implicitly be no writes.
if (this->fileIndex != that.fileIndex && this->rel[1] == that.rel[1])
return false;
if (this->rel[0] || that.rel[0])
return this->base == that.base;
return
(this->offset < that.offset + that.size) &&
(this->offset + this->size > that.offset);
}
// We must not eliminate stores that affect the result of @ld if
// we find later stores to the same location, and we may no longer
// merge them with later stores.
// The stored value can, however, still be used to determine the value
// returned by future loads.
void
MemoryOpt::lockStores(Instruction *const ld)
{
for (Record *r = stores[ld->src(0).getFile()]; r; r = r->next)
if (!r->locked && r->overlaps(ld))
r->locked = true;
}
// Prior loads from the location of @st are no longer valid.
// Stores to the location of @st may no longer be used to derive
// the value at it nor be coalesced into later stores.
void
MemoryOpt::purgeRecords(Instruction *const st, DataFile f)
{
if (st)
f = st->src(0).getFile();
for (Record *r = loads[f]; r; r = r->next)
if (!st || r->overlaps(st))
r->unlink(&loads[f]);
for (Record *r = stores[f]; r; r = r->next)
if (!st || r->overlaps(st))
r->unlink(&stores[f]);
}
bool
MemoryOpt::visit(BasicBlock *bb)
{
bool ret = runOpt(bb);
// Run again, one pass won't combine 4 32 bit ld/st to a single 128 bit ld/st
// where 96 bit memory operations are forbidden.
if (ret)
ret = runOpt(bb);
return ret;
}
bool
MemoryOpt::runOpt(BasicBlock *bb)
{
Instruction *ldst, *next;
Record *rec;
bool isAdjacent = true;
for (ldst = bb->getEntry(); ldst; ldst = next) {
bool keep = true;
bool isLoad = true;
next = ldst->next;
if (ldst->op == OP_LOAD || ldst->op == OP_VFETCH) {
if (ldst->isDead()) {
// might have been produced by earlier optimization
delete_Instruction(prog, ldst);
continue;
}
} else
if (ldst->op == OP_STORE || ldst->op == OP_EXPORT) {
if (typeSizeof(ldst->dType) == 4 &&
ldst->src(1).getFile() == FILE_GPR &&
ldst->getSrc(1)->getInsn()->op == OP_NOP) {
delete_Instruction(prog, ldst);
continue;
}
isLoad = false;
} else {
// TODO: maybe have all fixed ops act as barrier ?
if (ldst->op == OP_CALL ||
ldst->op == OP_BAR ||
ldst->op == OP_MEMBAR) {
purgeRecords(NULL, FILE_MEMORY_LOCAL);
purgeRecords(NULL, FILE_MEMORY_GLOBAL);
purgeRecords(NULL, FILE_MEMORY_SHARED);
purgeRecords(NULL, FILE_SHADER_OUTPUT);
} else
if (ldst->op == OP_ATOM || ldst->op == OP_CCTL) {
if (ldst->src(0).getFile() == FILE_MEMORY_GLOBAL) {
purgeRecords(NULL, FILE_MEMORY_LOCAL);
purgeRecords(NULL, FILE_MEMORY_GLOBAL);
purgeRecords(NULL, FILE_MEMORY_SHARED);
} else {
purgeRecords(NULL, ldst->src(0).getFile());
}
} else
if (ldst->op == OP_EMIT || ldst->op == OP_RESTART) {
purgeRecords(NULL, FILE_SHADER_OUTPUT);
}
continue;
}
if (ldst->getPredicate()) // TODO: handle predicated ld/st
continue;
if (ldst->perPatch) // TODO: create separate per-patch lists
continue;
if (isLoad) {
DataFile file = ldst->src(0).getFile();
// if ld l[]/g[] look for previous store to eliminate the reload
if (file == FILE_MEMORY_GLOBAL || file == FILE_MEMORY_LOCAL) {
// TODO: shared memory ?
rec = findRecord(ldst, false, isAdjacent);
if (rec && !isAdjacent)
keep = !replaceLdFromSt(ldst, rec);
}
// or look for ld from the same location and replace this one
rec = keep ? findRecord(ldst, true, isAdjacent) : NULL;
if (rec) {
if (!isAdjacent)
keep = !replaceLdFromLd(ldst, rec);
else
// or combine a previous load with this one
keep = !combineLd(rec, ldst);
}
if (keep)
lockStores(ldst);
} else {
rec = findRecord(ldst, false, isAdjacent);
if (rec) {
if (!isAdjacent)
keep = !replaceStFromSt(ldst, rec);
else
keep = !combineSt(rec, ldst);
}
if (keep)
purgeRecords(ldst, DATA_FILE_COUNT);
}
if (keep)
addRecord(ldst);
}
reset();
return true;
}
// =============================================================================
// Turn control flow into predicated instructions (after register allocation !).
// TODO:
// Could move this to before register allocation on NVC0 and also handle nested
// constructs.
class FlatteningPass : public Pass
{
private:
virtual bool visit(Function *);
virtual bool visit(BasicBlock *);
bool tryPredicateConditional(BasicBlock *);
void predicateInstructions(BasicBlock *, Value *pred, CondCode cc);
void tryPropagateBranch(BasicBlock *);
inline bool isConstantCondition(Value *pred);
inline bool mayPredicate(const Instruction *, const Value *pred) const;
inline void removeFlow(Instruction *);
uint8_t gpr_unit;
};
bool
FlatteningPass::isConstantCondition(Value *pred)
{
Instruction *insn = pred->getUniqueInsn();
assert(insn);
if (insn->op != OP_SET || insn->srcExists(2))
return false;
for (int s = 0; s < 2 && insn->srcExists(s); ++s) {
Instruction *ld = insn->getSrc(s)->getUniqueInsn();
DataFile file;
if (ld) {
if (ld->op != OP_MOV && ld->op != OP_LOAD)
return false;
if (ld->src(0).isIndirect(0))
return false;
file = ld->src(0).getFile();
} else {
file = insn->src(s).getFile();
// catch $r63 on NVC0 and $r63/$r127 on NV50. Unfortunately maxGPR is
// in register "units", which can vary between targets.
if (file == FILE_GPR) {
Value *v = insn->getSrc(s);
int bytes = v->reg.data.id * MIN2(v->reg.size, 4);
int units = bytes >> gpr_unit;
if (units > prog->maxGPR)
file = FILE_IMMEDIATE;
}
}
if (file != FILE_IMMEDIATE && file != FILE_MEMORY_CONST)
return false;
}
return true;
}
void
FlatteningPass::removeFlow(Instruction *insn)
{
FlowInstruction *term = insn ? insn->asFlow() : NULL;
if (!term)
return;
Graph::Edge::Type ty = term->bb->cfg.outgoing().getType();
if (term->op == OP_BRA) {
// TODO: this might get more difficult when we get arbitrary BRAs
if (ty == Graph::Edge::CROSS || ty == Graph::Edge::BACK)
return;
} else
if (term->op != OP_JOIN)
return;
Value *pred = term->getPredicate();
delete_Instruction(prog, term);
if (pred && pred->refCount() == 0) {
Instruction *pSet = pred->getUniqueInsn();
pred->join->reg.data.id = -1; // deallocate
if (pSet->isDead())
delete_Instruction(prog, pSet);
}
}
void
FlatteningPass::predicateInstructions(BasicBlock *bb, Value *pred, CondCode cc)
{
for (Instruction *i = bb->getEntry(); i; i = i->next) {
if (i->isNop())
continue;
assert(!i->getPredicate());
i->setPredicate(cc, pred);
}
removeFlow(bb->getExit());
}
bool
FlatteningPass::mayPredicate(const Instruction *insn, const Value *pred) const
{
if (insn->isPseudo())
return true;
// TODO: calls where we don't know which registers are modified
if (!prog->getTarget()->mayPredicate(insn, pred))
return false;
for (int d = 0; insn->defExists(d); ++d)
if (insn->getDef(d)->equals(pred))
return false;
return true;
}
// If we jump to BRA/RET/EXIT, replace the jump with it.
// NOTE: We do not update the CFG anymore here !
//
// TODO: Handle cases where we skip over a branch (maybe do that elsewhere ?):
// BB:0
// @p0 bra BB:2 -> @!p0 bra BB:3 iff (!) BB:2 immediately adjoins BB:1
// BB1:
// bra BB:3
// BB2:
// ...
// BB3:
// ...
void
FlatteningPass::tryPropagateBranch(BasicBlock *bb)
{
for (Instruction *i = bb->getExit(); i && i->op == OP_BRA; i = i->prev) {
BasicBlock *bf = i->asFlow()->target.bb;
if (bf->getInsnCount() != 1)
continue;
FlowInstruction *bra = i->asFlow();
FlowInstruction *rep = bf->getExit()->asFlow();
if (!rep || rep->getPredicate())
continue;
if (rep->op != OP_BRA &&
rep->op != OP_JOIN &&
rep->op != OP_EXIT)
continue;
// TODO: If there are multiple branches to @rep, only the first would
// be replaced, so only remove them after this pass is done ?
// Also, need to check all incident blocks for fall-through exits and
// add the branch there.
bra->op = rep->op;
bra->target.bb = rep->target.bb;
if (bf->cfg.incidentCount() == 1)
bf->remove(rep);
}
}
bool
FlatteningPass::visit(Function *fn)
{
gpr_unit = prog->getTarget()->getFileUnit(FILE_GPR);
return true;
}
bool
FlatteningPass::visit(BasicBlock *bb)
{
if (tryPredicateConditional(bb))
return true;
// try to attach join to previous instruction
if (prog->getTarget()->hasJoin) {
Instruction *insn = bb->getExit();
if (insn && insn->op == OP_JOIN && !insn->getPredicate()) {
insn = insn->prev;
if (insn && !insn->getPredicate() &&
!insn->asFlow() &&
insn->op != OP_DISCARD &&
insn->op != OP_TEXBAR &&
!isTextureOp(insn->op) && // probably just nve4
!isSurfaceOp(insn->op) && // not confirmed
insn->op != OP_LINTERP && // probably just nve4
insn->op != OP_PINTERP && // probably just nve4
((insn->op != OP_LOAD && insn->op != OP_STORE && insn->op != OP_ATOM) ||
(typeSizeof(insn->dType) <= 4 && !insn->src(0).isIndirect(0))) &&
!insn->isNop()) {
insn->join = 1;
bb->remove(bb->getExit());
return true;
}
}
}
tryPropagateBranch(bb);
return true;
}
bool
FlatteningPass::tryPredicateConditional(BasicBlock *bb)
{
BasicBlock *bL = NULL, *bR = NULL;
unsigned int nL = 0, nR = 0, limit = 12;
Instruction *insn;
unsigned int mask;
mask = bb->initiatesSimpleConditional();
if (!mask)
return false;
assert(bb->getExit());
Value *pred = bb->getExit()->getPredicate();
assert(pred);
if (isConstantCondition(pred))
limit = 4;
Graph::EdgeIterator ei = bb->cfg.outgoing();
if (mask & 1) {
bL = BasicBlock::get(ei.getNode());
for (insn = bL->getEntry(); insn; insn = insn->next, ++nL)
if (!mayPredicate(insn, pred))
return false;
if (nL > limit)
return false; // too long, do a real branch
}
ei.next();
if (mask & 2) {
bR = BasicBlock::get(ei.getNode());
for (insn = bR->getEntry(); insn; insn = insn->next, ++nR)
if (!mayPredicate(insn, pred))
return false;
if (nR > limit)
return false; // too long, do a real branch
}
if (bL)
predicateInstructions(bL, pred, bb->getExit()->cc);
if (bR)
predicateInstructions(bR, pred, inverseCondCode(bb->getExit()->cc));
if (bb->joinAt) {
bb->remove(bb->joinAt);
bb->joinAt = NULL;
}
removeFlow(bb->getExit()); // delete the branch/join at the fork point
// remove potential join operations at the end of the conditional
if (prog->getTarget()->joinAnterior) {
bb = BasicBlock::get((bL ? bL : bR)->cfg.outgoing().getNode());
if (bb->getEntry() && bb->getEntry()->op == OP_JOIN)
removeFlow(bb->getEntry());
}
return true;
}
// =============================================================================
// Fold Immediate into MAD; must be done after register allocation due to
// constraint SDST == SSRC2
// TODO:
// Does NVC0+ have other situations where this pass makes sense?
class PostRaLoadPropagation : public Pass
{
private:
virtual bool visit(Instruction *);
void handleMADforNV50(Instruction *);
void handleMADforNVC0(Instruction *);
};
static bool
post_ra_dead(Instruction *i)
{
for (int d = 0; i->defExists(d); ++d)
if (i->getDef(d)->refCount())
return false;
return true;
}
// Fold Immediate into MAD; must be done after register allocation due to
// constraint SDST == SSRC2
void
PostRaLoadPropagation::handleMADforNV50(Instruction *i)
{
if (i->def(0).getFile() != FILE_GPR ||
i->src(0).getFile() != FILE_GPR ||
i->src(1).getFile() != FILE_GPR ||
i->src(2).getFile() != FILE_GPR ||
i->getDef(0)->reg.data.id != i->getSrc(2)->reg.data.id)
return;
if (i->getDef(0)->reg.data.id >= 64 ||
i->getSrc(0)->reg.data.id >= 64)
return;
if (i->flagsSrc >= 0 && i->getSrc(i->flagsSrc)->reg.data.id != 0)
return;
if (i->getPredicate())
return;
Value *vtmp;
Instruction *def = i->getSrc(1)->getInsn();
if (def && def->op == OP_SPLIT && typeSizeof(def->sType) == 4)
def = def->getSrc(0)->getInsn();
if (def && def->op == OP_MOV && def->src(0).getFile() == FILE_IMMEDIATE) {
vtmp = i->getSrc(1);
if (isFloatType(i->sType)) {
i->setSrc(1, def->getSrc(0));
} else {
ImmediateValue val;
// getImmediate() has side-effects on the argument so this *shouldn't*
// be folded into the assert()
MAYBE_UNUSED bool ret = def->src(0).getImmediate(val);
assert(ret);
if (i->getSrc(1)->reg.data.id & 1)
val.reg.data.u32 >>= 16;
val.reg.data.u32 &= 0xffff;
i->setSrc(1, new_ImmediateValue(prog, val.reg.data.u32));
}
/* There's no post-RA dead code elimination, so do it here
* XXX: if we add more code-removing post-RA passes, we might
* want to create a post-RA dead-code elim pass */
if (post_ra_dead(vtmp->getInsn())) {
Value *src = vtmp->getInsn()->getSrc(0);
// Careful -- splits will have already been removed from the
// functions. Don't double-delete.
if (vtmp->getInsn()->bb)
delete_Instruction(prog, vtmp->getInsn());
if (src->getInsn() && post_ra_dead(src->getInsn()))
delete_Instruction(prog, src->getInsn());
}
}
}
void
PostRaLoadPropagation::handleMADforNVC0(Instruction *i)
{
if (i->def(0).getFile() != FILE_GPR ||
i->src(0).getFile() != FILE_GPR ||
i->src(1).getFile() != FILE_GPR ||
i->src(2).getFile() != FILE_GPR ||
i->getDef(0)->reg.data.id != i->getSrc(2)->reg.data.id)
return;
// TODO: gm107 can also do this for S32, maybe other chipsets as well
if (i->dType != TYPE_F32)
return;
if ((i->src(2).mod | Modifier(NV50_IR_MOD_NEG)) != Modifier(NV50_IR_MOD_NEG))
return;
ImmediateValue val;
int s;
if (i->src(0).getImmediate(val))
s = 1;
else if (i->src(1).getImmediate(val))
s = 0;
else
return;
if ((i->src(s).mod | Modifier(NV50_IR_MOD_NEG)) != Modifier(NV50_IR_MOD_NEG))
return;
if (s == 1)
i->swapSources(0, 1);
Instruction *imm = i->getSrc(1)->getInsn();
i->setSrc(1, imm->getSrc(0));
if (post_ra_dead(imm))
delete_Instruction(prog, imm);
}
bool
PostRaLoadPropagation::visit(Instruction *i)
{
switch (i->op) {
case OP_FMA:
case OP_MAD:
if (prog->getTarget()->getChipset() < 0xc0)
handleMADforNV50(i);
else
handleMADforNVC0(i);
break;
default:
break;
}
return true;
}
// =============================================================================
// Common subexpression elimination. Stupid O^2 implementation.
class LocalCSE : public Pass
{
private:
virtual bool visit(BasicBlock *);
inline bool tryReplace(Instruction **, Instruction *);
DLList ops[OP_LAST + 1];
};
class GlobalCSE : public Pass
{
private:
virtual bool visit(BasicBlock *);
};
bool
Instruction::isActionEqual(const Instruction *that) const
{
if (this->op != that->op ||
this->dType != that->dType ||
this->sType != that->sType)
return false;
if (this->cc != that->cc)
return false;
if (this->asTex()) {
if (memcmp(&this->asTex()->tex,
&that->asTex()->tex,
sizeof(this->asTex()->tex)))
return false;
} else
if (this->asCmp()) {
if (this->asCmp()->setCond != that->asCmp()->setCond)
return false;
} else
if (this->asFlow()) {
return false;
} else {
if (this->ipa != that->ipa ||
this->lanes != that->lanes ||
this->perPatch != that->perPatch)
return false;
if (this->postFactor != that->postFactor)
return false;
}
if (this->subOp != that->subOp ||
this->saturate != that->saturate ||
this->rnd != that->rnd ||
this->ftz != that->ftz ||
this->dnz != that->dnz ||
this->cache != that->cache ||
this->mask != that->mask)
return false;
return true;
}
bool
Instruction::isResultEqual(const Instruction *that) const
{
unsigned int d, s;
// NOTE: location of discard only affects tex with liveOnly and quadops
if (!this->defExists(0) && this->op != OP_DISCARD)
return false;
if (!isActionEqual(that))
return false;
if (this->predSrc != that->predSrc)
return false;
for (d = 0; this->defExists(d); ++d) {
if (!that->defExists(d) ||
!this->getDef(d)->equals(that->getDef(d), false))
return false;
}
if (that->defExists(d))
return false;
for (s = 0; this->srcExists(s); ++s) {
if (!that->srcExists(s))
return false;
if (this->src(s).mod != that->src(s).mod)
return false;
if (!this->getSrc(s)->equals(that->getSrc(s), true))
return false;
}
if (that->srcExists(s))
return false;
if (op == OP_LOAD || op == OP_VFETCH || op == OP_ATOM) {
switch (src(0).getFile()) {
case FILE_MEMORY_CONST:
case FILE_SHADER_INPUT:
return true;
case FILE_SHADER_OUTPUT:
return bb->getProgram()->getType() == Program::TYPE_TESSELLATION_EVAL;
default:
return false;
}
}
return true;
}
// pull through common expressions from different in-blocks
bool
GlobalCSE::visit(BasicBlock *bb)
{
Instruction *phi, *next, *ik;
int s;
// TODO: maybe do this with OP_UNION, too
for (phi = bb->getPhi(); phi && phi->op == OP_PHI; phi = next) {
next = phi->next;
if (phi->getSrc(0)->refCount() > 1)
continue;
ik = phi->getSrc(0)->getInsn();
if (!ik)
continue; // probably a function input
if (ik->defCount(0xff) > 1)
continue; // too painful to check if we can really push this forward
for (s = 1; phi->srcExists(s); ++s) {
if (phi->getSrc(s)->refCount() > 1)
break;
if (!phi->getSrc(s)->getInsn() ||
!phi->getSrc(s)->getInsn()->isResultEqual(ik))
break;
}
if (!phi->srcExists(s)) {
Instruction *entry = bb->getEntry();
ik->bb->remove(ik);
if (!entry || entry->op != OP_JOIN)
bb->insertHead(ik);
else
bb->insertAfter(entry, ik);
ik->setDef(0, phi->getDef(0));
delete_Instruction(prog, phi);
}
}
return true;
}
bool
LocalCSE::tryReplace(Instruction **ptr, Instruction *i)
{
Instruction *old = *ptr;
// TODO: maybe relax this later (causes trouble with OP_UNION)
if (i->isPredicated())
return false;
if (!old->isResultEqual(i))
return false;
for (int d = 0; old->defExists(d); ++d)
old->def(d).replace(i->getDef(d), false);
delete_Instruction(prog, old);
*ptr = NULL;
return true;
}
bool
LocalCSE::visit(BasicBlock *bb)
{
unsigned int replaced;
do {
Instruction *ir, *next;
replaced = 0;
// will need to know the order of instructions
int serial = 0;
for (ir = bb->getFirst(); ir; ir = ir->next)
ir->serial = serial++;
for (ir = bb->getFirst(); ir; ir = next) {
int s;
Value *src = NULL;
next = ir->next;
if (ir->fixed) {
ops[ir->op].insert(ir);
continue;
}
for (s = 0; ir->srcExists(s); ++s)
if (ir->getSrc(s)->asLValue())
if (!src || ir->getSrc(s)->refCount() < src->refCount())
src = ir->getSrc(s);
if (src) {
for (Value::UseIterator it = src->uses.begin();
it != src->uses.end(); ++it) {
Instruction *ik = (*it)->getInsn();
if (ik && ik->bb == ir->bb && ik->serial < ir->serial)
if (tryReplace(&ir, ik))
break;
}
} else {
DLLIST_FOR_EACH(&ops[ir->op], iter)
{
Instruction *ik = reinterpret_cast<Instruction *>(iter.get());
if (tryReplace(&ir, ik))
break;
}
}
if (ir)
ops[ir->op].insert(ir);
else
++replaced;
}
for (unsigned int i = 0; i <= OP_LAST; ++i)
ops[i].clear();
} while (replaced);
return true;
}
// =============================================================================
// Remove computations of unused values.
class DeadCodeElim : public Pass
{
public:
bool buryAll(Program *);
private:
virtual bool visit(BasicBlock *);
void checkSplitLoad(Instruction *ld); // for partially dead loads
unsigned int deadCount;
};
bool
DeadCodeElim::buryAll(Program *prog)
{
do {
deadCount = 0;
if (!this->run(prog, false, false))
return false;
} while (deadCount);
return true;
}
bool
DeadCodeElim::visit(BasicBlock *bb)
{
Instruction *prev;
for (Instruction *i = bb->getExit(); i; i = prev) {
prev = i->prev;
if (i->isDead()) {
++deadCount;
delete_Instruction(prog, i);
} else
if (i->defExists(1) &&
i->subOp == 0 &&
(i->op == OP_VFETCH || i->op == OP_LOAD)) {
checkSplitLoad(i);
} else
if (i->defExists(0) && !i->getDef(0)->refCount()) {
if (i->op == OP_ATOM ||
i->op == OP_SUREDP ||
i->op == OP_SUREDB) {
i->setDef(0, NULL);
if (i->op == OP_ATOM && i->subOp == NV50_IR_SUBOP_ATOM_EXCH) {
i->cache = CACHE_CV;
i->op = OP_STORE;
i->subOp = 0;
}
} else if (i->op == OP_LOAD && i->subOp == NV50_IR_SUBOP_LOAD_LOCKED) {
i->setDef(0, i->getDef(1));
i->setDef(1, NULL);
}
}
}
return true;
}
// Each load can go into up to 4 destinations, any of which might potentially
// be dead (i.e. a hole). These can always be split into 2 loads, independent
// of where the holes are. We find the first contiguous region, put it into
// the first load, and then put the second contiguous region into the second
// load. There can be at most 2 contiguous regions.
//
// Note that there are some restrictions, for example it's not possible to do
// a 64-bit load that's not 64-bit aligned, so such a load has to be split
// up. Also hardware doesn't support 96-bit loads, so those also have to be
// split into a 64-bit and 32-bit load.
void
DeadCodeElim::checkSplitLoad(Instruction *ld1)
{
Instruction *ld2 = NULL; // can get at most 2 loads
Value *def1[4];
Value *def2[4];
int32_t addr1, addr2;
int32_t size1, size2;
int d, n1, n2;
uint32_t mask = 0xffffffff;
for (d = 0; ld1->defExists(d); ++d)
if (!ld1->getDef(d)->refCount() && ld1->getDef(d)->reg.data.id < 0)
mask &= ~(1 << d);
if (mask == 0xffffffff)
return;
addr1 = ld1->getSrc(0)->reg.data.offset;
n1 = n2 = 0;
size1 = size2 = 0;
// Compute address/width for first load
for (d = 0; ld1->defExists(d); ++d) {
if (mask & (1 << d)) {
if (size1 && (addr1 & 0x7))
break;
def1[n1] = ld1->getDef(d);
size1 += def1[n1++]->reg.size;
} else
if (!n1) {
addr1 += ld1->getDef(d)->reg.size;
} else {
break;
}
}
// Scale back the size of the first load until it can be loaded. This
// typically happens for TYPE_B96 loads.
while (n1 &&
!prog->getTarget()->isAccessSupported(ld1->getSrc(0)->reg.file,
typeOfSize(size1))) {
size1 -= def1[--n1]->reg.size;
d--;
}
// Compute address/width for second load
for (addr2 = addr1 + size1; ld1->defExists(d); ++d) {
if (mask & (1 << d)) {
assert(!size2 || !(addr2 & 0x7));
def2[n2] = ld1->getDef(d);
size2 += def2[n2++]->reg.size;
} else if (!n2) {
assert(!n2);
addr2 += ld1->getDef(d)->reg.size;
} else {
break;
}
}
// Make sure that we've processed all the values
for (; ld1->defExists(d); ++d)
assert(!(mask & (1 << d)));
updateLdStOffset(ld1, addr1, func);
ld1->setType(typeOfSize(size1));
for (d = 0; d < 4; ++d)
ld1->setDef(d, (d < n1) ? def1[d] : NULL);
if (!n2)
return;
ld2 = cloneShallow(func, ld1);
updateLdStOffset(ld2, addr2, func);
ld2->setType(typeOfSize(size2));
for (d = 0; d < 4; ++d)
ld2->setDef(d, (d < n2) ? def2[d] : NULL);
ld1->bb->insertAfter(ld1, ld2);
}
// =============================================================================
#define RUN_PASS(l, n, f) \
if (level >= (l)) { \
if (dbgFlags & NV50_IR_DEBUG_VERBOSE) \
INFO("PEEPHOLE: %s\n", #n); \
n pass; \
if (!pass.f(this)) \
return false; \
}
bool
Program::optimizeSSA(int level)
{
RUN_PASS(1, DeadCodeElim, buryAll);
RUN_PASS(1, CopyPropagation, run);
RUN_PASS(1, MergeSplits, run);
RUN_PASS(2, GlobalCSE, run);
RUN_PASS(1, LocalCSE, run);
RUN_PASS(2, AlgebraicOpt, run);
RUN_PASS(2, ModifierFolding, run); // before load propagation -> less checks
RUN_PASS(1, ConstantFolding, foldAll);
RUN_PASS(1, Split64BitOpPreRA, run);
RUN_PASS(1, LoadPropagation, run);
RUN_PASS(1, IndirectPropagation, run);
RUN_PASS(2, MemoryOpt, run);
RUN_PASS(2, LateAlgebraicOpt, run);
RUN_PASS(2, LocalCSE, run);
RUN_PASS(0, DeadCodeElim, buryAll);
return true;
}
bool
Program::optimizePostRA(int level)
{
RUN_PASS(2, FlatteningPass, run);
RUN_PASS(2, PostRaLoadPropagation, run);
return true;
}
}