ashikk
/
random_well512a_simd
mirror of https://github.com/ppettersson/random_well512a_simd.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Minor cleanup.

This commit is contained in:
Peter Pettersson 2013-04-22 22:14:28 +02:00
parent 542cb9a746
commit df14bb6ae5
4 changed files with 16 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
RandomWELL512a.h
View File

@ -28,8 +28,8 @@
#ifndef RANDOM_WELL512A_H #ifndef RANDOM_WELL512A_H
#define RANDOM_WELL512A_H #define RANDOM_WELL512A_H
#include <stdlib.h> // Needed for rand(). #include <stdlib.h> // Needed for rand().
#include <memory.h> // Needed for memcpy(). #include <memory.h> // Needed for memcpy().
class RandomWELL512a class RandomWELL512a
{ {
@ -63,9 +63,9 @@ inline RandomWELL512a::RandomWELL512a(unsigned *seed)
inline unsigned RandomWELL512a::GetUnsigned() inline unsigned RandomWELL512a::GetUnsigned()
{ {
#define MUTATE_LEFT(value, shift) value ^ (value << shift) #define MUTATE_LEFT(value, shift) value ^ (value << shift)
#define MUTATE_RIGHT(value, shift) value ^ (value >> shift) #define MUTATE_RIGHT(value, shift) value ^ (value >> shift)
#define MUTATE_LEFT_MIX(value, shift, mix) value ^ ((value << shift) & mix) #define MUTATE_LEFT_MIX(value, shift, mix) value ^ ((value << shift) & mix)
unsigned index_9 = (index + 9) & 15; unsigned index_9 = (index + 9) & 15;
unsigned index_13 = (index + 13) & 15; unsigned index_13 = (index + 13) & 15;

16
RandomWELL512a_SSE2.h
View File

@ -28,8 +28,8 @@
#ifndef RANDOM_WELL512A_SSE2_H #ifndef RANDOM_WELL512A_SSE2_H
#define RANDOM_WELL512A_SSE2_H #define RANDOM_WELL512A_SSE2_H
#include <emmintrin.h> // SSE2 instrinsics. #include <emmintrin.h> // SSE2 instrinsics.
#include <stdlib.h> // Needed for rand(). #include <stdlib.h> // Needed for rand().
class RandomWELL512a_SSE2 class RandomWELL512a_SSE2
{ {
@ -86,12 +86,12 @@ inline double RandomWELL512a_SSE2::GetDouble()
inline void RandomWELL512a_SSE2::GetUnsigned4(unsigned *result4) inline void RandomWELL512a_SSE2::GetUnsigned4(unsigned *result4)
{ {
unsigned index_15 = (index + 15) & 15; unsigned index_15 = (index + 15) & 15;
__m128i state_index = state[index]; __m128i state_index = state[index];
__m128i state_index_9 = state[(index + 9) & 15]; __m128i state_index_9 = state[(index + 9) & 15];
__m128i state_index_13 = state[(index + 13) & 15]; __m128i state_index_13 = state[(index + 13) & 15];
__m128i state_index_15 = state[index_15]; __m128i state_index_15 = state[index_15];
const __m128i kMix = _mm_set1_epi32(0xda442d24); const __m128i kMix = _mm_set1_epi32(0xda442d24);
__m128i left = _mm_xor_si128(state_index, _mm_slli_epi32(state_index, 16)); __m128i left = _mm_xor_si128(state_index, _mm_slli_epi32(state_index, 16));
__m128i right = _mm_xor_si128(state_index_13, _mm_slli_epi32(state_index_13, 15)); __m128i right = _mm_xor_si128(state_index_13, _mm_slli_epi32(state_index_13, 15));

5
gcc/Makefile
View File

@ -92,8 +92,8 @@ cleanall:
help: help:
@echo "BUILD = Build mode. One of:" @echo "BUILD = Build mode. One of:"
@echo " debug (default)" @echo " debug (no optimizations)"
@echo " release (optimizations)" @echo " release (optimizations, the default)"
@echo "APPS = Applications to build. Defaults to all." @echo "APPS = Applications to build. Defaults to all."
@echo "VERBOSE = Full output from commands if set." @echo "VERBOSE = Full output from commands if set."
@ -107,7 +107,6 @@ $(EXES):
$(BUILD_DIR)/%.a: $(BUILD_DIR)/%.a:
@mkdir -p $(@D) @mkdir -p $(@D)
$(HUSH_GENERATE) $(AR) $(ARFLAGS) $@ $^ $(HUSH_GENERATE) $(AR) $(ARFLAGS) $@ $^
# $(HUSH_GENERATE) $(AR) $(ARFLAGS) $@ $^ 2> /dev/null
$(BUILD_DIR)/%.o: $(SRC_ROOT)/%.c $(BUILD_DIR)/%.o: $(SRC_ROOT)/%.c
@mkdir -p $(@D) @mkdir -p $(@D)

2
readme.txt
View File

@ -29,7 +29,7 @@ Original C 0.223 0.19
=== REQUIREMENTS === === REQUIREMENTS ===
An SSE2 capable CPU (i.e. AMD Athlon 64, Intel Pentium 4 or newer. An SSE2 capable CPU (i.e. AMD Athlon 64, Intel Pentium 4 or newer).
It's been tested with Visual Studio 2012 and gcc 4.6.3 but should compile It's been tested with Visual Studio 2012 and gcc 4.6.3 but should compile
and run on other compilers as well. and run on other compilers as well.