vector_sse 0.0.1.pre → 0.0.2.pre

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: 98eb4a76855923685b4f2857b834900304bb85c8
-  data.tar.gz: 5ca280566cecbd8cf474e6950ebf21dfe3104199
+  metadata.gz: fdabd1511ca1ab68168a0723f06d26afe3312627
+  data.tar.gz: 9bb853808e5c0a8ec8e8b1592e84a0537954f6dc
 SHA512:
-  metadata.gz: 39cbc928cf7f4fc555d8edca11caa0eefb28d34b23356c837f2ef453afd06d6f1b6798c802ad4cad76f65fe3ae25fd6dee459f32fc731bfaad386e4ecf9dbf49
-  data.tar.gz: df243d6d5260ac8f414ccfb5b2c68dfc0a090304cd7c0d49947dc0a6d880dbe966a9cba265d6bedb7caadd41fd6d7bc55e9776929ab453870d85d3c17112b04f
+  metadata.gz: 213dffc68fc752e4520f0d5f7e910467b3fb51faba6ed5f6052b38d6a677928ead0b899b7658427fa78e08c7400f9ac874112b8c325174a328a9c137cdf34d6d
+  data.tar.gz: 10b7386669a7a1acbbcad0a9f24323e703487997d18969f9b02a6f0d5ba02ac2bd9ff1c50ba5df71f33d18413e38545454eb48700570046bc2c2c5fb1f54f712

data/.gitignore

@@ -1,4 +1,8 @@
 test_vectorops.rb
 vector_sse-*.gem
 tmp/
-*.sublime-*
+*.sublime-*
+Gemfile.lock
+vendor/
+*DS_Store
+.bundle/

data/Gemfile

@@ -0,0 +1,5 @@
+source 'https://rubygems.org'
+gem 'rake', '10.4.2'
+gem 'rake-compiler', '0.9.5'
+gem 'rspec'
+

data/ext/.gitignore

@@ -0,0 +1 @@
+*DS_Store

data/ext/vector_sse/extconf.rb

@@ -3,7 +3,7 @@ require 'mkmf'
 # Give it a name
 extension_name = 'vector_sse'
 
-$CFLAGS << ' -O3'
+$CFLAGS << ' -O3 -msse -msse2 -msse3 -msse4.1 -msse4.2'
 
 # Check for dependencies
 have_header( 'emmintrin.h' )

data/ext/vector_sse/vector_sse_add.c

@@ -32,6 +32,7 @@
 
 #include <emmintrin.h>
 #include "vector_sse_add.h"
+#include "vector_sse_common.h"
 
 #define  TEMPLATE_ADD_S( FUNC_NAME, TYPE, OFTYPE, TYPE_SIZE, CONV_IN, CONV_OUT, EL_PER_VEC, ADD ) \
 VALUE FUNC_NAME( VALUE self, VALUE left, VALUE right ) \
@@ -118,10 +119,10 @@ VALUE FUNC_NAME( VALUE self, VALUE left, VALUE right ) \
 
 TEMPLATE_ADD_S( method_vec_add_s32, int32_t, int32_t, 32, NUM2INT, INT2NUM, 4, _mm_add_epi32 );
 TEMPLATE_ADD_S( method_vec_add_s64, int64_t, int64_t, 64, NUM2LL, LL2NUM, 2, _mm_add_epi64 );
-TEMPLATE_ADD_S( method_vec_add_f32, float, int32_t, 32, NUM2DBL, DBL2NUM, 4, _mm_add_ps );
-TEMPLATE_ADD_S( method_vec_add_f64, double, int64_t, 64, NUM2DBL, DBL2NUM, 2, _mm_add_pd );
+TEMPLATE_ADD_S( method_vec_add_f32, float, int32_t, 32, NUM2DBL, DBL2NUM, 4, add_f32 );
+TEMPLATE_ADD_S( method_vec_add_f64, double, int64_t, 64, NUM2DBL, DBL2NUM, 2, add_f64 );
 
 TEMPLATE_ADD_S( method_vec_sub_s32, int32_t, int32_t, 32, NUM2INT, INT2NUM, 4, _mm_sub_epi32 );
 TEMPLATE_ADD_S( method_vec_sub_s64, int64_t, int64_t, 64, NUM2LL, LL2NUM, 2, _mm_sub_epi64 );
-TEMPLATE_ADD_S( method_vec_sub_f32, float, int32_t, 32, NUM2DBL, DBL2NUM, 4, _mm_sub_ps );
-TEMPLATE_ADD_S( method_vec_sub_f64, double, int64_t, 64, NUM2DBL, DBL2NUM, 2, _mm_sub_pd );
+TEMPLATE_ADD_S( method_vec_sub_f32, float, int32_t, 32, NUM2DBL, DBL2NUM, 4, sub_f32 );
+TEMPLATE_ADD_S( method_vec_sub_f64, double, int64_t, 64, NUM2DBL, DBL2NUM, 2, sub_f64 );

data/ext/vector_sse/vector_sse_common.c

@@ -0,0 +1,31 @@
+#include "vector_sse_common.h"
+
+__m128i add_f32( const __m128i left, const __m128i right )
+{
+   return _mm_castps_si128( _mm_add_ps( _mm_castsi128_ps( left ), _mm_castsi128_ps( right ) ) );
+}
+
+__m128i add_f64( const __m128i left, const __m128i right )
+{
+   return _mm_cvtpd_epi32( _mm_add_pd( _mm_castsi128_pd( left ), _mm_castsi128_pd( right ) ) );
+}
+
+__m128i sub_f32( const __m128i left, const __m128i right )
+{
+   return _mm_castps_si128( _mm_sub_ps( _mm_castsi128_ps( left ), _mm_castsi128_ps( right ) ) );
+}
+
+__m128i sub_f64( const __m128i left, const __m128i right )
+{
+   return _mm_cvtpd_epi32( _mm_sub_pd( _mm_castsi128_pd( left ), _mm_castsi128_pd( right ) ) );
+}
+
+__m128i mul_f32( const __m128i left, const __m128i right )
+{
+   return _mm_castps_si128( _mm_mul_ps( _mm_castsi128_ps( left ), _mm_castsi128_ps( right ) ) );
+}
+
+__m128i mul_f64( const __m128i left, const __m128i right )
+{
+   return _mm_cvtpd_epi32( _mm_mul_pd( _mm_castsi128_pd( left ), _mm_castsi128_pd( right ) ) );
+}

data/ext/vector_sse/vector_sse_common.h

@@ -0,0 +1,13 @@
+#ifndef  VECTOR_SSE_COMMON_H
+#define  VECTOR_SSE_COMMON_H
+
+#include <emmintrin.h>
+
+__m128i add_f32( const __m128i left, const __m128i right );
+__m128i add_f64( const __m128i left, const __m128i right );
+__m128i sub_f32( const __m128i left, const __m128i right );
+__m128i sub_f64( const __m128i left, const __m128i right );
+__m128i mul_f32( const __m128i left, const __m128i right );
+__m128i mul_f64( const __m128i left, const __m128i right );
+
+#endif // VECTOR_SSE_COMMON_H

data/ext/vector_sse/vector_sse_mul.c

@@ -32,6 +32,7 @@
 
 #include <emmintrin.h>
 #include "vector_sse_mul.h"
+#include "vector_sse_common.h"
 
 #define  SSE_VECTOR_WIDTH    (4)
 
@@ -40,7 +41,6 @@ VALUE method_mat_mul_s32( VALUE self, VALUE left, VALUE left_rows_rb, VALUE left
    uint32_t left_row = 0;
    uint32_t right_col = 0;
    uint32_t common = 0;
-   uint32_t vector_pos = 0;
    uint32_t input_index = 0;
    uint32_t pos = 0;
 
@@ -148,17 +148,8 @@ VALUE method_mat_mul_s64( VALUE self, VALUE left, VALUE left_rows_rb, VALUE left
    uint32_t left_row = 0;
    uint32_t right_col = 0;
    uint32_t common = 0;
-   uint32_t vector_pos = 0;
-   uint32_t input_index = 0;
    uint32_t pos = 0;
 
-   int64_t left_segment[ SSE_VECTOR_WIDTH ];
-   int64_t right_segment[ SSE_VECTOR_WIDTH ];
-
-   __m128i* left_vec = NULL;
-   __m128i* right_vec = NULL;
-   __m128i result_vec;
-
    VALUE result = Qnil;
 
    int64_t* result_native = NULL;
@@ -233,7 +224,6 @@ VALUE method_mat_mul_f32( VALUE self, VALUE left, VALUE left_rows_rb, VALUE left
    uint32_t left_row = 0;
    uint32_t right_col = 0;
    uint32_t common = 0;
-   uint32_t vector_pos = 0;
    uint32_t input_index = 0;
    uint32_t pos = 0;
 
@@ -303,7 +293,7 @@ VALUE method_mat_mul_f32( VALUE self, VALUE left, VALUE left_rows_rb, VALUE left
 
             left_vec   = ( __m128i *)left_segment;
             right_vec  = ( __m128i *)right_segment;
-            result_vec = _mm_mul_ps( *left_vec, *right_vec );
+            result_vec = mul_f32( *left_vec, *right_vec );
 
             _mm_store_si128( (__m128i*)result_segment, result_vec );
             for ( pos = 0; pos < SSE_VECTOR_WIDTH; ++pos )

data/ext/vector_sse/vector_sse_sum.c

@@ -32,9 +32,31 @@
 
 #include <string.h>
 #include <emmintrin.h>
+#include <ruby.h>
 #include "vector_sse_sum.h"
+#include "vector_sse_common.h"
 
-#define  TEMPLATE_SUM_S( FUNC_NAME, TYPE, OFTYPE, TYPE_SIZE, CONV_IN, CONV_OUT, EL_PER_VEC, ADD ) \
+
+// Check for overflow
+// __m128i sign_left;
+// __m128i sign_right;
+// const int32_t OVERFLOW_MASK = ( (int32_t)0x1 << (32-1) );
+// int32_t overflow[ 4 ];
+// __m128i* overflow_vec = (__m128i*)overflow;
+// sign_left = _mm_xor_si128(result_vec, left_vec);
+// sign_right = _mm_xor_si128(result_vec, right_vec);
+// *overflow_vec = _mm_and_si128(sign_left, sign_right);
+
+// for ( vector_pos = 0; vector_pos < 4; ++vector_pos )
+// {
+//    if ( ( (int32_t)overflow[ vector_pos ] & OVERFLOW_MASK ) )
+//    {
+//       rb_raise( rb_eRuntimeError, "Vector addition overflow" );
+//    }
+// }
+
+
+#define  TEMPLATE_SUM_S( FUNC_NAME, TYPE, CONV_IN, CONV_OUT, EL_PER_VEC, ADDER ) \
 VALUE FUNC_NAME( VALUE self, VALUE vector ) \
 { \
    uint32_t length      = 0; \
@@ -44,20 +66,12 @@ VALUE FUNC_NAME( VALUE self, VALUE vector ) \
 \
    TYPE  result = 0; \
 \
-   TYPE left_segment[ EL_PER_VEC ]; \
-   TYPE right_segment[ EL_PER_VEC ]; \
    TYPE result_segment[ EL_PER_VEC ]; \
    TYPE vector_segment[ EL_PER_VEC ]; \
 \
-   __m128i left_vec; \
-   __m128i right_vec; \
+   __m128i left_vec;   \
+   __m128i right_vec;  \
    __m128i result_vec; \
-\
-   __m128i sign_left; \
-   __m128i sign_right; \
-   const OFTYPE OVERFLOW_MASK = ( (OFTYPE)0x1 << (TYPE_SIZE-1) ); \
-   OFTYPE overflow[ EL_PER_VEC ]; \
-   __m128i* overflow_vec = (__m128i*)overflow; \
 \
    Check_Type( vector, T_ARRAY ); \
 \
@@ -85,19 +99,7 @@ VALUE FUNC_NAME( VALUE self, VALUE vector ) \
          right_vec = _mm_loadu_si128( (const __m128i *)vector_segment ); \
          left_vec  = _mm_loadu_si128( &result_vec ); \
 \
-         result_vec = ADD( left_vec, right_vec ); \
-\
-         sign_left = _mm_xor_si128(result_vec, left_vec); \
-         sign_right = _mm_xor_si128(result_vec, right_vec); \
-         *overflow_vec = _mm_and_si128(sign_left, sign_right); \
-\
-         for ( vector_pos = 0; vector_pos < EL_PER_VEC; ++vector_pos ) \
-         { \
-            if ( ( (OFTYPE)overflow[ vector_pos ] & OVERFLOW_MASK ) ) \
-            { \
-               rb_raise( rb_eRuntimeError, "Vector addition overflow" ); \
-            } \
-         } \
+         result_vec = ADDER( left_vec, right_vec ); \
       } \
 \
       _mm_store_si128( (__m128i*)result_segment, result_vec ); \
@@ -111,8 +113,8 @@ VALUE FUNC_NAME( VALUE self, VALUE vector ) \
    return CONV_OUT( result ); \
 }
 
-TEMPLATE_SUM_S( method_vec_sum_s32, int32_t, int32_t, 32, NUM2INT, INT2NUM, 4, _mm_add_epi32 );
-TEMPLATE_SUM_S( method_vec_sum_s64, int64_t, int64_t, 64, NUM2LL, LL2NUM, 2, _mm_add_epi64 );
-TEMPLATE_SUM_S( method_vec_sum_f32, float, int32_t, 32, NUM2DBL, DBL2NUM, 4, _mm_add_ps );
-TEMPLATE_SUM_S( method_vec_sum_f64, double, int64_t, 32, NUM2DBL, DBL2NUM, 2, _mm_add_pd );
+TEMPLATE_SUM_S( method_vec_sum_s32, int32_t, NUM2INT, INT2NUM, 4, _mm_add_epi32 );
+TEMPLATE_SUM_S( method_vec_sum_s64, int64_t, NUM2LL, LL2NUM, 2, _mm_add_epi64 );
+TEMPLATE_SUM_S( method_vec_sum_f32, float, NUM2DBL, DBL2NUM, 4, add_f32 );
+TEMPLATE_SUM_S( method_vec_sum_f64, double, NUM2DBL, DBL2NUM, 2, add_f64 );
 

data/ext/vector_sse/vector_sse_vec_mul.c

@@ -36,6 +36,7 @@
 #include <smmintrin.h>
 #endif
 #include "vector_sse_vec_mul.h"
+#include "vector_sse_common.h"
 
 #define  SSE_VECTOR_WIDTH    (4)
 // #define  EL_PER_VEC    SSE_VECTOR_WIDTH
@@ -66,14 +67,14 @@ static inline __m128i mul_s64( const __m128i* left_vec, const __m128i* right_vec
    return _mm_loadu_si128( (const __m128i *)result );
 }
 
-static inline __m128i mul_f32(const __m128i* a, const __m128i* b )
+static inline __m128i mul_f32_ptr(const __m128i* a, const __m128i* b )
 {
-   return _mm_mul_ps( *a, *b );
+   return mul_f32( *a, *b );
 }
 
-static inline __m128i mul_f64(const __m128i* a, const __m128i* b )
+static inline __m128i mul_f64_ptr(const __m128i* a, const __m128i* b )
 {
-   return _mm_mul_pd( *a, *b );
+   return mul_f64( *a, *b );
 }
 
 
@@ -95,9 +96,6 @@ VALUE FUNC_NAME( VALUE self, VALUE left, VALUE right ) \
 \
    TYPE result_segment[ EL_PER_VEC ]; \
    __m128i result_vec; \
-\
-   __m128i sign_left;  \
-   __m128i sign_right; \
 \
    Check_Type( left, T_ARRAY );  \
    Check_Type( right, T_ARRAY ); \
@@ -152,6 +150,6 @@ VALUE FUNC_NAME( VALUE self, VALUE left, VALUE right ) \
 
 TEMPLATE_VEC_MUL_S( method_vec_mul_s32, int32_t, 32, NUM2INT, INT2NUM, 4, mul_s32 );
 TEMPLATE_VEC_MUL_S( method_vec_mul_s64, int64_t, 64, NUM2LL, LL2NUM, 2, mul_s64 );
-TEMPLATE_VEC_MUL_S( method_vec_mul_f32, float, 32, NUM2DBL, DBL2NUM, 4, mul_f32 );
-TEMPLATE_VEC_MUL_S( method_vec_mul_f64, double, 64, NUM2DBL, DBL2NUM, 2, mul_f64 );
+TEMPLATE_VEC_MUL_S( method_vec_mul_f32, float, 32, NUM2DBL, DBL2NUM, 4, mul_f32_ptr );
+TEMPLATE_VEC_MUL_S( method_vec_mul_f64, double, 64, NUM2DBL, DBL2NUM, 2, mul_f64_ptr );
 

data/lib/vector_sse.rb

@@ -36,7 +36,7 @@ require File.join( bin_root, 'vector_sse.so' )
 
 module VectorSSE
 
-   VERSION = "0.0.1.pre"
+   VERSION = "0.0.2.pre"
 
    module Type
       S32 = 0

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: vector_sse
 version: !ruby/object:Gem::Version
-  version: 0.0.1.pre
+  version: 0.0.2.pre
 platform: ruby
 authors:
 - Robert Glissmann
@@ -14,28 +14,28 @@ dependencies:
   name: rake-compiler
   requirement: !ruby/object:Gem::Requirement
     requirements:
-    - - ~>
+    - - "~>"
       - !ruby/object:Gem::Version
         version: 0.9.5
   type: :development
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
-    - - ~>
+    - - "~>"
       - !ruby/object:Gem::Version
         version: 0.9.5
 - !ruby/object:Gem::Dependency
   name: rspec
   requirement: !ruby/object:Gem::Requirement
     requirements:
-    - - ~>
+    - - "~>"
       - !ruby/object:Gem::Version
         version: 3.1.0
   type: :development
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
-    - - ~>
+    - - "~>"
       - !ruby/object:Gem::Version
         version: 3.1.0
 description: VectorSse employs x86 Streaming SIMD Extensions (SSE), v3 or greater,
@@ -46,15 +46,19 @@ extensions:
 - ext/vector_sse/extconf.rb
 extra_rdoc_files: []
 files:
-- .gitignore
+- ".gitignore"
+- Gemfile
 - LICENSE.txt
 - README.md
 - Rakefile
+- ext/.gitignore
 - ext/vector_sse/.gitignore
 - ext/vector_sse/extconf.rb
 - ext/vector_sse/vector_sse.c
 - ext/vector_sse/vector_sse_add.c
 - ext/vector_sse/vector_sse_add.h
+- ext/vector_sse/vector_sse_common.c
+- ext/vector_sse/vector_sse_common.h
 - ext/vector_sse/vector_sse_mul.c
 - ext/vector_sse/vector_sse_mul.h
 - ext/vector_sse/vector_sse_sum.c
@@ -76,17 +80,17 @@ require_paths:
 - lib
 required_ruby_version: !ruby/object:Gem::Requirement
   requirements:
-  - - '>='
+  - - ">="
     - !ruby/object:Gem::Version
       version: '0'
 required_rubygems_version: !ruby/object:Gem::Requirement
   requirements:
-  - - '>'
+  - - ">"
     - !ruby/object:Gem::Version
       version: 1.3.1
 requirements: []
 rubyforge_project: 
-rubygems_version: 2.0.14
+rubygems_version: 2.4.8
 signing_key: 
 specification_version: 4
 summary: SIMD accelerated vector and matrix operations