-

Not all of these events have been tested and they may be broken
USE AT YOUR OWN RISK!

CBO (Last Level Cache Slice) CACHE Events
  CBO.LLC_LOOKUP                 Cache Lookups                           
  CBO.LLC_LOOKUP.ANY             Cache Lookups                           
  CBO.LLC_LOOKUP.DATA_READ       Cache Lookups                           
  CBO.LLC_LOOKUP.NID             Cache Lookups                           
  CBO.LLC_LOOKUP.READ            Cache Lookups                           

androm3da

#!/bin/bash -ex

CC="clang"
CXX="clang++"

SRCTOP=$(readlink -f ${PWD})
INSTALL=${SRCTOP}/install
if [[ ! -d ${SRCTOP}/llvm ]]; then
    echo Expected to find the source in ${SRCTOP}/llvm but it is missing
    exit 3

androm3da

#!/bin/bash -ex

CC="clang"
CXX="clang++"
export PATH=/local/mnt/workspace/install/binutils-2.27/bin:${PATH}

SRCTOP=$(readlink -f ${PWD})
INSTALL=${1-${SRCTOP}/install}
if [[ ! -d ${SRCTOP}/llvm ]]; then
    echo Expected to find the source in ${SRCTOP}/llvm but it is missing

martinmoene

// This tells Catch to provide a main() - only do this in one cpp file:
#define CATCH_CONFIG_MAIN  
#include "catch.hpp"

rygorous

So the general idea with FMA3 is that it's designed to have enough degrees of
freedom so you can just have a more general FMA4 op in the IR and pick the right
FMA3 op *very* late (ideally, after register allocation!).

The generalized FMA op looks like this:

  dst = FMA ±src0, ±src1, ±src2

where at most one of the src's can be a memory operand. This computes
(±src0 * ±src1) ± src2. The two signs for src0 and src1 combine

nkurz

nate@haswell:~/src$ likwid-perfctr -m -g UOPS_ISSUED_ANY:PMC0,UOPS_EXECUTED_CORE:PMC1,UOPS_RETIRED_ALL:PMC2,BR_INST_RETIRED_NEAR_TAKEN:PMC3 -C 1 fusion
-------------------------------------------------------------
-------------------------------------------------------------
CPU type:	Intel Core Haswell processor
CPU clock:	3.39 GHz
-------------------------------------------------------------
fusion
two_micro_two_macro: sum1=10000000, sum2=9999999
one_micro_two_macro: sum1=10000000, sum2=9999999
one_micro_one_macro: sum1=10000000, sum2=9999999

withzombies

#!/usr/bin/env python
# Copyright 2017 Ryan Stortz (@withzombies)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software

rygorous

Brief explanation what I did to get the speed-up, and the thought process behind it.

The original code went:

EnterLoop:
			movaps		xmm3, xmmword ptr [edx+ecx*2]		// Box1YZ
			cmpnltps	xmm3, xmm2
			movmskps	eax, xmm3

			cmp			eax, 0Ch

pervognsen

• Intel 4004, first microprocessor: http://www.computerhistory.org/collections/catalog/102658187
• Intel 8008: http://www.computerhistory.org/collections/catalog/102657982
• Intel 8080: http://www.computerhistory.org/collections/catalog/102658123
• Z80: http://www.computerhistory.org/collections/catalog/102658073
• Federico Faggin, SGT inventor, chip designer for 4004, Z80: http://www.computerhistory.org/collections/catalog/102658025
• Bill Mensch, chip designer on 6800/6502/65C02/65816: http://www.computerhistory.org/collections/catalog/102739969
• Motorolla 68000: http://www.computerhistory.org/collections/catalog/102658109
• 3dfx, Voodoo, the seminal GPU: http://www.computerhistory.org/collections/catalog/102746834
• LSI Logic, EDA/fabless innovator: http://www.computerhistory.org/collections/catalog/102746194
• VLSI Technologies, EDA/fabless innovator: http://www.computerhistory.org/collections/catalog/102746456

vegard

#if 0
(g++-9 $0 || g++ $0) && \
	./a.out > output.tex && \
	pdflatex output && \
	exec convert -density 400 -flatten output.pdf -resize 25% output.png
exit 1
#endif

#include <cmath>
#include <cstdio>