syoyo

The clock cycle count for "expf()" (reference) is 141645.

The clock cycle count for "expapprox()" is 74.

The clock cycle count for "expapprox4()" is 127 (/4 = 31).

// GCC
#define RESTRICT __restrict__

// Disable range check makes faster evaluation of exp().

syoyo

// code From http://gallium.inria.fr/blog/fast-vectorizable-math-approx/

$ e-gcc compile options: -O3 -mno-soft-cmpsf -mcmove -mfp-mode=truncate

// 73 clocks
00000f40 <_expapprox>:
     f40:       200b 0002       mov r1,0x0
     f44:       476b 0aa2       mov r2,0xaa3b
     f48:       470b 14b2       movt r2,0x4b38
     f4c:       2fcb 14e2       movt r1,0x4e7e

syoyo

// Writing (&x)[i] in operator[] is safe/correct C++ code or not? 
// The following is the reduced code fragment(Thus not work by just copy&paste) which Intel C++ compier(ver 13 and 15) miscompiles(Release build only) the code for the access to real3 object through operator[] inside OpenMP loop.
// clang and gcc are OK to compile&run

typedef float real;

struct real3 {
  real3() {}
  real3(real xx, real yy, real zz) {
    x = xx;

syoyo

CL_DEVICE_NAME: pthread-Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz
CL_DEVICE_VENDOR: 
CL_DEVICE_OPENCL_C_VERSION: OpenCL C 1.2
CL_DEVICE_PROFILE: FULL_PROFILE
CL_DEVICE_VERSION: OpenCL 1.2 pocl
CL_DRIVER_VERSION: 0.9
CL_DEVICE_EXTENSIONS: cl_khr_fp64 cl_khr_fp16 cl_khr_byte_addressable_store
CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS: 3
CL_DEVICE_IMAGE_SUPPORT: 1
CL_DEVICE_IMAGE2D_MAX_WIDTH: 8192

syoyo

Demo settings:
  SelectedDemo=1, demoname = BoxBox
  x_dim=30, y_dim=30, z_dim=30
  x_gap=16.299999, y_gap=6.300000, z_gap=16.299999

OpenCL settings:
  Preferred cl_device index 1
  Preferred cl_platform index-1

Platform info:

syoyo

#!/usr/bin/env python

import ninja_syntax
import glob
import os

cxx_files = [
    "src/OptionParser.cpp"
  , "src/easywsclient.cpp"
  , "src/json_to_eson.cc"

syoyo

typedef float float4 __attribute__((ext_vector_type(4)));

void
clang_test(float* out, float* in)
{
  float4 a, b;
  
  b = *((float4*)in);
  a = b * b;

syoyo

void
clang_test(float* out, float* in)
{
  for (int i = 0; i < 4; i++) {
    out[i] = in[i] * in[i];
  }
}

// asm output

syoyo

==20402== Memcheck, a memory error detector
==20402== Copyright (C) 2002-2009, and GNU GPL'd, by Julian Seward et al.
==20402== Using Valgrind-3.5.0 and LibVEX; rerun with -h for copyright info
==20402== Command: ../../ispc --emit-c++ --target=generic-16 -h ao_ispc.h -o bora.cc ao.ispc
==20402== 
==20402== Conditional jump or move depends on uninitialised value(s)
==20402==    at 0xBC0085: clang::SourceManager::getColumnNumber(clang::FileID, unsigned int, bool*) const (in /home/syoyo/work/ispc/ispc)
==20402==    by 0xBC3838: clang::SourceManager::getPresumedLoc(clang::SourceLocation) const (in /home/syoyo/work/ispc/ispc)
==20402==    by 0x50734D: (anonymous namespace)::PrintPPOutputPPCallbacks::FileChanged(clang::SourceLocation, clang::PPCallbacks::FileChangeReason, clang::SrcMgr::CharacteristicKind, clang::FileID) (in /home/syoyo/work/ispc/ispc)
==20402==    by 0xC17F3B: clang::Preprocessor::HandleDigitDirective(clang::Token&) (in /home/syoyo/work/ispc/ispc)

syoyo

std::vector<std::thread> workers;
std::atomic<unsigned int> i(0);

for (auto t = 0; t < std::thread::hardware_concurrency(); t++){
    workers.push_back(std::thread([&,t](){
        int index = 0;
        while((index = i++) < numItems){
           ...
        }
    }));