Optimized division by constants

This commit is contained in:
tevador 2019-01-11 14:08:21 +01:00
parent c02ee4291d
commit 451dfc5730
4 changed files with 385 additions and 5 deletions

View file

@ -11,7 +11,7 @@ SRCDIR=src
OBJDIR=obj
LDFLAGS=-lpthread
TOBJS=$(addprefix $(OBJDIR)/,instructionsPortable.o TestAluFpu.o)
ROBJS=$(addprefix $(OBJDIR)/,argon2_core.o argon2_ref.o AssemblyGeneratorX86.o blake2b.o CompiledVirtualMachine.o dataset.o JitCompilerX86.o instructionsPortable.o Instruction.o InterpretedVirtualMachine.o main.o Program.o softAes.o VirtualMachine.o t1ha2.o Cache.o virtualMemory.o)
ROBJS=$(addprefix $(OBJDIR)/,argon2_core.o argon2_ref.o AssemblyGeneratorX86.o blake2b.o CompiledVirtualMachine.o dataset.o JitCompilerX86.o instructionsPortable.o Instruction.o InterpretedVirtualMachine.o main.o Program.o softAes.o VirtualMachine.o t1ha2.o Cache.o virtualMemory.o divideByConstantCodegen.o)
ifeq ($(PLATFORM),x86_64)
ROBJS += $(OBJDIR)/JitCompilerX86-static.o
endif
@ -57,6 +57,9 @@ $(OBJDIR)/CompiledVirtualMachine.o: $(addprefix $(SRCDIR)/,CompiledVirtualMachin
$(OBJDIR)/dataset.o: $(addprefix $(SRCDIR)/,dataset.cpp common.hpp Pcg32.hpp) | $(OBJDIR)
$(CXX) $(CXXFLAGS) -c $(SRCDIR)/dataset.cpp -o $@
$(OBJDIR)/divideByConstantCodegen.o: $(addprefix $(SRCDIR)/,divideByConstantCodegen.c divideByConstantCodegen.h) | $(OBJDIR)
$(CC) $(CCFLAGS) -c $(SRCDIR)/divideByConstantCodegen.c -o $@
$(OBJDIR)/JitCompilerX86.o: $(addprefix $(SRCDIR)/,JitCompilerX86.cpp JitCompilerX86.hpp Instruction.hpp instructionWeights.hpp) | $(OBJDIR)
$(CXX) $(CXXFLAGS) -c $(SRCDIR)/JitCompilerX86.cpp -o $@

View file

@ -17,10 +17,14 @@ You should have received a copy of the GNU General Public License
along with RandomX. If not, see<http://www.gnu.org/licenses/>.
*/
//#define TRACE
//#define MAGIC_DIVISION
#include "AssemblyGeneratorX86.hpp"
#include "Pcg32.hpp"
#include "common.hpp"
#include "instructions.hpp"
#ifdef MAGIC_DIVISION
#include "divideByConstantCodegen.h"
#endif
namespace RandomX {
@ -315,34 +319,118 @@ namespace RandomX {
void AssemblyGeneratorX86::h_DIV_64(Instruction& instr, int i) {
genar(instr, i);
if ((instr.locb & 7) >= 6) {
#ifdef MAGIC_DIVISION
if (instr.imm32 != 0) {
uint32_t divisor = instr.imm32;
asmCode << "\t; magic divide by " << divisor << std::endl;
if (divisor & (divisor - 1)) {
magicu_info mi = compute_unsigned_magic_info(divisor, sizeof(uint64_t) * 8);
if (mi.pre_shift > 0)
asmCode << "\tshr rax, " << mi.pre_shift << std::endl;
if (mi.increment) {
asmCode << "\tadd rax, 1" << std::endl;
asmCode << "\tsbb rax, 0" << std::endl;
}
asmCode << "\tmov rcx, " << mi.multiplier << std::endl;
asmCode << "\tmul rcx" << std::endl;
asmCode << "\tmov rax, rdx" << std::endl;
if (mi.post_shift > 0)
asmCode << "\tshr rax, " << mi.post_shift << std::endl;
}
else { //divisor is a power of two
int shift = 0;
while (divisor >>= 1)
++shift;
if(shift > 0)
asmCode << "\tshr rax, " << shift << std::endl;
}
}
#else
if (instr.imm32 == 0) {
asmCode << "\tmov ecx, 1" << std::endl;
}
else {
asmCode << "\tmov ecx, " << instr.imm32 << std::endl;
}
#endif
}
else {
asmCode << "\tmov ecx, 1" << std::endl;
asmCode << "\tmov edx, " << regR32[instr.regb % RegistersCount] << std::endl;
asmCode << "\ttest edx, edx" << std::endl;
asmCode << "\tcmovne ecx, edx" << std::endl;
#ifdef MAGIC_DIVISION
asmCode << "\txor edx, edx" << std::endl;
asmCode << "\tdiv rcx" << std::endl;
#endif
}
#ifndef MAGIC_DIVISION
asmCode << "\txor edx, edx" << std::endl;
asmCode << "\tdiv rcx" << std::endl;
#endif
gencr(instr);
}
void AssemblyGeneratorX86::h_IDIV_64(Instruction& instr, int i) {
genar(instr, i);
#ifdef MAGIC_DIVISION
if ((instr.locb & 7) >= 6) {
int64_t divisor = instr.imm32;
asmCode << "\t; magic divide by " << divisor << std::endl;
if ((divisor & -divisor) == divisor || (divisor & -divisor) == -divisor) {
// +/- power of two
bool negative = divisor < 0;
if (negative)
divisor = -divisor;
int shift = 0;
uint64_t unsignedDivisor = divisor;
while (unsignedDivisor >>= 1)
++shift;
if (shift > 0) {
asmCode << "\tmov rcx, rax" << std::endl;
asmCode << "\tsar rcx, 63" << std::endl;
uint32_t mask = (1ULL << shift) + 0xFFFFFFFF;
asmCode << "\tand ecx, 0" << std::hex << mask << std::dec << "h" << std::endl;
asmCode << "\tadd rax, rcx" << std::endl;
asmCode << "\tsar rax, " << shift << std::endl;
}
if(negative)
asmCode << "\tneg rax" << std::endl;
} else if(divisor != 0) {
magics_info mi = compute_signed_magic_info(divisor);
if ((divisor >= 0) != (mi.multiplier >= 0))
asmCode << "\tmov rcx, rax" << std::endl;
asmCode << "\tmov rdx, " << mi.multiplier << std::endl;
asmCode << "\timul rdx" << std::endl;
asmCode << "\tmov rax, rdx" << std::endl;
asmCode << "\txor edx, edx" << std::endl;
bool haveSF = false;
if (divisor > 0 && mi.multiplier < 0) {
asmCode << "\tadd rax, rcx" << std::endl;
haveSF = true;
}
if (divisor < 0 && mi.multiplier > 0) {
asmCode << "\tsub rax, rcx" << std::endl;
haveSF = true;
}
if (mi.shift > 0) {
asmCode << "\tsar rax, " << mi.shift << std::endl;
haveSF = true;
}
if (!haveSF)
asmCode << "\ttest rax, rax" << std::endl;
asmCode << "\tsets dl" << std::endl;
asmCode << "\tadd rax, rdx" << std::endl;
}
}
else {
#endif
asmCode << "\tmov edx, ";
genbr132(instr);
asmCode << "\tcmp edx, -1" << std::endl;
asmCode << "\tjne short safe_idiv_" << i << std::endl;
asmCode << "\tmov rcx, rax" << std::endl;
asmCode << "\trol rcx, 1" << std::endl;
asmCode << "\tdec rcx" << std::endl;
asmCode << "\tjz short result_idiv_" << i << std::endl;
asmCode << "\tneg rax" << std::endl;
asmCode << "\tjmp short result_idiv_" << i << std::endl;
asmCode << "safe_idiv_" << i << ":" << std::endl;
asmCode << "\tmov ecx, 1" << std::endl;
asmCode << "\ttest edx, edx" << std::endl;
@ -351,6 +439,9 @@ namespace RandomX {
asmCode << "\tcqo" << std::endl;
asmCode << "\tidiv rcx" << std::endl;
asmCode << "result_idiv_" << i << ":" << std::endl;
#ifdef MAGIC_DIVISION
}
#endif
gencr(instr);
}

View file

@ -0,0 +1,169 @@
/*
Reference implementations of computing and using the "magic number" approach to dividing
by constants, including codegen instructions. The unsigned division incorporates the
"round down" optimization per ridiculous_fish.
This is free and unencumbered software. Any copyright is dedicated to the Public Domain.
*/
#include <limits.h> //for CHAR_BIT
#include <assert.h>
#include "divideByConstantCodegen.h"
struct magicu_info compute_unsigned_magic_info(uint D, unsigned num_bits) {
//The numerator must fit in a uint
assert(num_bits > 0 && num_bits <= sizeof(uint) * CHAR_BIT);
// D must be larger than zero and not a power of 2
assert(D & (D - 1));
// The eventual result
struct magicu_info result;
// Bits in a uint
const unsigned UINT_BITS = sizeof(uint) * CHAR_BIT;
// The extra shift implicit in the difference between UINT_BITS and num_bits
const unsigned extra_shift = UINT_BITS - num_bits;
// The initial power of 2 is one less than the first one that can possibly work
const uint initial_power_of_2 = (uint)1 << (UINT_BITS - 1);
// The remainder and quotient of our power of 2 divided by d
uint quotient = initial_power_of_2 / D, remainder = initial_power_of_2 % D;
// ceil(log_2 D)
unsigned ceil_log_2_D;
// The magic info for the variant "round down" algorithm
uint down_multiplier = 0;
unsigned down_exponent = 0;
int has_magic_down = 0;
// Compute ceil(log_2 D)
ceil_log_2_D = 0;
uint tmp;
for (tmp = D; tmp > 0; tmp >>= 1)
ceil_log_2_D += 1;
// Begin a loop that increments the exponent, until we find a power of 2 that works.
unsigned exponent;
for (exponent = 0; ; exponent++) {
// Quotient and remainder is from previous exponent; compute it for this exponent.
if (remainder >= D - remainder) {
// Doubling remainder will wrap around D
quotient = quotient * 2 + 1;
remainder = remainder * 2 - D;
}
else {
// Remainder will not wrap
quotient = quotient * 2;
remainder = remainder * 2;
}
// We're done if this exponent works for the round_up algorithm.
// Note that exponent may be larger than the maximum shift supported,
// so the check for >= ceil_log_2_D is critical.
if ((exponent + extra_shift >= ceil_log_2_D) || (D - remainder) <= ((uint)1 << (exponent + extra_shift)))
break;
// Set magic_down if we have not set it yet and this exponent works for the round_down algorithm
if (!has_magic_down && remainder <= ((uint)1 << (exponent + extra_shift))) {
has_magic_down = 1;
down_multiplier = quotient;
down_exponent = exponent;
}
}
if (exponent < ceil_log_2_D) {
// magic_up is efficient
result.multiplier = quotient + 1;
result.pre_shift = 0;
result.post_shift = exponent;
result.increment = 0;
}
else if (D & 1) {
// Odd divisor, so use magic_down, which must have been set
assert(has_magic_down);
result.multiplier = down_multiplier;
result.pre_shift = 0;
result.post_shift = down_exponent;
result.increment = 1;
}
else {
// Even divisor, so use a prefix-shifted dividend
unsigned pre_shift = 0;
uint shifted_D = D;
while ((shifted_D & 1) == 0) {
shifted_D >>= 1;
pre_shift += 1;
}
result = compute_unsigned_magic_info(shifted_D, num_bits - pre_shift);
assert(result.increment == 0 && result.pre_shift == 0); //expect no increment or pre_shift in this path
result.pre_shift = pre_shift;
}
return result;
}
struct magics_info compute_signed_magic_info(sint D) {
// D must not be zero and must not be a power of 2 (or its negative)
assert(D != 0 && (D & -D) != D && (D & -D) != -D);
// Our result
struct magics_info result;
// Bits in an sint
const unsigned SINT_BITS = sizeof(sint) * CHAR_BIT;
// Absolute value of D (we know D is not the most negative value since that's a power of 2)
const uint abs_d = (D < 0 ? -D : D);
// The initial power of 2 is one less than the first one that can possibly work
// "two31" in Warren
unsigned exponent = SINT_BITS - 1;
const uint initial_power_of_2 = (uint)1 << exponent;
// Compute the absolute value of our "test numerator,"
// which is the largest dividend whose remainder with d is d-1.
// This is called anc in Warren.
const uint tmp = initial_power_of_2 + (D < 0);
const uint abs_test_numer = tmp - 1 - tmp % abs_d;
// Initialize our quotients and remainders (q1, r1, q2, r2 in Warren)
uint quotient1 = initial_power_of_2 / abs_test_numer, remainder1 = initial_power_of_2 % abs_test_numer;
uint quotient2 = initial_power_of_2 / abs_d, remainder2 = initial_power_of_2 % abs_d;
uint delta;
// Begin our loop
do {
// Update the exponent
exponent++;
// Update quotient1 and remainder1
quotient1 *= 2;
remainder1 *= 2;
if (remainder1 >= abs_test_numer) {
quotient1 += 1;
remainder1 -= abs_test_numer;
}
// Update quotient2 and remainder2
quotient2 *= 2;
remainder2 *= 2;
if (remainder2 >= abs_d) {
quotient2 += 1;
remainder2 -= abs_d;
}
// Keep going as long as (2**exponent) / abs_d <= delta
delta = abs_d - remainder2;
} while (quotient1 < delta || (quotient1 == delta && remainder1 == 0));
result.multiplier = quotient2 + 1;
if (D < 0) result.multiplier = -result.multiplier;
result.shift = exponent - SINT_BITS;
return result;
}

View file

@ -0,0 +1,117 @@
/*
Copyright (c) 2018 tevador
This file is part of RandomX.
RandomX is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
RandomX is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with RandomX. If not, see<http://www.gnu.org/licenses/>.
*/
#pragma once
#include <stdint.h>
#if defined(__cplusplus)
extern "C" {
#endif
typedef uint64_t uint;
typedef int64_t sint;
/* Computes "magic info" for performing signed division by a fixed integer D.
The type 'sint' is assumed to be defined as a signed integer type large enough
to hold both the dividend and the divisor.
Here >> is arithmetic (signed) shift, and >>> is logical shift.
To emit code for n/d, rounding towards zero, use the following sequence:
m = compute_signed_magic_info(D)
emit("result = (m.multiplier * n) >> SINT_BITS");
if d > 0 and m.multiplier < 0: emit("result += n")
if d < 0 and m.multiplier > 0: emit("result -= n")
if m.post_shift > 0: emit("result >>= m.shift")
emit("result += (result < 0)")
The shifts by SINT_BITS may be "free" if the high half of the full multiply
is put in a separate register.
The final add can of course be implemented via the sign bit, e.g.
result += (result >>> (SINT_BITS - 1))
or
result -= (result >> (SINT_BITS - 1))
This code is heavily indebted to Hacker's Delight by Henry Warren.
See http://www.hackersdelight.org/HDcode/magic.c.txt
Used with permission from http://www.hackersdelight.org/permissions.htm
*/
struct magics_info {
sint multiplier; // the "magic number" multiplier
unsigned shift; // shift for the dividend after multiplying
};
struct magics_info compute_signed_magic_info(sint D);
/* Computes "magic info" for performing unsigned division by a fixed positive integer D.
The type 'uint' is assumed to be defined as an unsigned integer type large enough
to hold both the dividend and the divisor. num_bits can be set appropriately if n is
known to be smaller than the largest uint; if this is not known then pass
(sizeof(uint) * CHAR_BIT) for num_bits.
Assume we have a hardware register of width UINT_BITS, a known constant D which is
not zero and not a power of 2, and a variable n of width num_bits (which may be
up to UINT_BITS). To emit code for n/d, use one of the two following sequences
(here >>> refers to a logical bitshift):
m = compute_unsigned_magic_info(D, num_bits)
if m.pre_shift > 0: emit("n >>>= m.pre_shift")
if m.increment: emit("n = saturated_increment(n)")
emit("result = (m.multiplier * n) >>> UINT_BITS")
if m.post_shift > 0: emit("result >>>= m.post_shift")
or
m = compute_unsigned_magic_info(D, num_bits)
if m.pre_shift > 0: emit("n >>>= m.pre_shift")
emit("result = m.multiplier * n")
if m.increment: emit("result = result + m.multiplier")
emit("result >>>= UINT_BITS")
if m.post_shift > 0: emit("result >>>= m.post_shift")
The shifts by UINT_BITS may be "free" if the high half of the full multiply
is put in a separate register.
saturated_increment(n) means "increment n unless it would wrap to 0," i.e.
if n == (1 << UINT_BITS)-1: result = n
else: result = n+1
A common way to implement this is with the carry bit. For example, on x86:
add 1
sbb 0
Some invariants:
1: At least one of pre_shift and increment is zero
2: multiplier is never zero
This code incorporates the "round down" optimization per ridiculous_fish.
*/
struct magicu_info {
uint multiplier; // the "magic number" multiplier
unsigned pre_shift; // shift for the dividend before multiplying
unsigned post_shift; //shift for the dividend after multiplying
int increment; // 0 or 1; if set then increment the numerator, using one of the two strategies
};
struct magicu_info compute_unsigned_magic_info(uint D, unsigned num_bits);
#if defined(__cplusplus)
}
#endif