nkurz

###  compare times for sample.int() vs internal function sample2()
compareSampleTimes = function(popSizeList=c(1e5, 1e6, 1e7, 1e8, 1e9),
    sampleSizeList=c(10, 100, 1000, 10000),
    numReplications=1000) {
    for (sampleSize in sampleSizeList) {
        for (popSize in popSizeList)  {
            elapsed1 = system.time(replicate(numReplications, sample.int(popSize, sampleSize)))[["elapsed"]]
            elapsed2 = system.time(replicate(numReplications, .Internal(sample2(popSize, sampleSize))))[["elapsed"]]
            cat(sprintf("Sample %d from %.0e: %f vs %f seconds\n", sampleSize, popSize, elapsed1, elapsed2))
        }

nkurz

diff --git a/DESCRIPTION b/DESCRIPTION
index cc23502..4d6e10c 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,6 +1,6 @@
 Package: data.table
-Version: 1.9.5
-Title: Extension of Data.frame
+Version: 1.9.5.1
+Title: Extension of Data.frame (+KeepSource -ByteCompile)

nkurz

--- R-3.2.1/src/main/memory.c.orig      2015-07-31 23:15:07.017151621 -0700
+++ R-3.2.1/src/main/memory.c   2015-07-31 23:17:10.185150073 -0700
@@ -3724,11 +3724,21 @@
 static FILE *R_MemReportingOutfile;
 static R_size_t R_MemReportingThreshold;

+static void printLineNum(FILE *file, SEXP srcref) {
+  if (srcref && !isNull(srcref)) {
+    int line = asInteger(srcref);
+    fprintf(file, "#%d ", line);

nkurz

// gcc -march=native -g -std=gnu99 -Wall -Wextra -O3 symmetric.c -o symmetric -DUSE_ALG
//   (where USE_ALG is one of USE_NATE, USE_KARIM, USE_BASIC, or USE_CONDITIONAL

// Or if using https://code.google.com/p/likwid/ with -m markers:
// gcc -march=native -g -std=gnu99 -Wall -Wextra -O3 symmetric.c -o symmetric -DLIKWID -llikwid -lpthread -lm -DUSE_ALG

#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>

nkurz

// C implementation for Pathfinding Benchmark by [email protected]
// See https://github.com/logicchains/LPATHBench for details

// Summary of benchmarks (see bottom for full numbers)
// 8981 LANGUAGE C 623
// 8981 LANGUAGE C++/clang 734
// 8981 LANGUAGE C++/gcc 755
// Best results compiling with GCC 4.7 or 4.8 -O2
// clang, icc and GCC 4.9 slightly worse with -O1, -O2, -O3, -Ofast
// -O3 and -Ofast much worse for all GCC.  -O1 mixed but worse.

nkurz

// Calculate cycles spent on overhead of function calls
// See http://cs.coloradocollege.edu/~bylvisaker/CallReturn/
// gcc -g -std=gnu99 -O3 -Wall -Wextra call-return.c -o call-return

#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>

#define DEFAULT_LOOP_COUNT (1000 * 1000)

nkurz

// cc  -fno-inline -g -march=native -std=gnu99 -O3 -Wall -Wextra broadcast.c -o broadcast
// works with 'gcc 4.8.2' and 'icc 14.03', but crashes with 'clang 3.4' because of alignment
// usage: broadcast [-r repeat] [-s size]

#ifdef LIKWID
#include <likwid.h>
#else
#define likwid_markerInit()
#define likwid_markerThreadInit()
#define likwid_markerStartRegion(name)

nkurz

; Minimal example, see also http://stackoverflow.com/q/26266953/3766665
; To build (Linux):
;   nasm -felf64 func.asm
;   ld func.o
; Then run:
;   perf stat -r10 ./a.out
; On Haswell and Sandy Bridge, observed runtime varies 
; ~15% depending on whether sub or sbb is used in the loop
section .text
global _start

nkurz

// gcc -std=gnu99 -O3 -Wall -Wextra same-function.c -o same-function
// Identical loops that execute in different but consistent times

#if COPY_AND_RUN_TO_TEST
for n in 0 1 2 3 4 5 6 7 8 9;
do echo same-function ${n}:;
/usr/bin/time -f "%e seconds" same-function ${n};
/usr/bin/time -f "%e seconds" same-function ${n};
/usr/bin/time -f "%e seconds" same-function ${n};
done

nkurz

GCC 4.8.1 
  4019f0:       66 0f 6f 00             movdqa (%rax),%xmm0
  4019f4:       48 83 c0 10             add    $0x10,%rax
  4019f8:       48 39 c5                cmp    %rax,%rbp
  4019fb:       66 0f 6f c8             movdqa %xmm0,%xmm1
  4019ff:       66 0f 6f e0             movdqa %xmm0,%xmm4
  401a03:       66 0f 62 c8             punpckldq %xmm0,%xmm1
  401a07:       66 0f 6a e0             punpckhdq %xmm0,%xmm4
  401a0b:       66 0f f4 cb             pmuludq %xmm3,%xmm1
  401a0f:       66 0f f4 e3             pmuludq %xmm3,%xmm4