diff options
Diffstat (limited to 'src/gf_w16.c')
| -rw-r--r-- | src/gf_w16.c | 127 |
1 files changed, 62 insertions, 65 deletions
diff --git a/src/gf_w16.c b/src/gf_w16.c index a62ea51..8316892 100644 --- a/src/gf_w16.c +++ b/src/gf_w16.c @@ -12,6 +12,7 @@ #include <stdio.h> #include <stdlib.h> #include "gf_w16.h" +#include "gf_cpu.h" #define AB2(ip, am1 ,am2, b, t1, t2) {\ t1 = (b << 1) & am1;\ @@ -391,6 +392,7 @@ gf_val_32_t gf_w16_matrix (gf_t *gf, gf_val_32_t b) extra memory. */ +#if defined(INTEL_SSE4_PCLMUL) static inline gf_val_32_t @@ -398,8 +400,6 @@ gf_w16_clm_multiply_2 (gf_t *gf, gf_val_32_t a16, gf_val_32_t b16) { gf_val_32_t rv = 0; -#if defined(INTEL_SSE4_PCLMUL) - __m128i a, b; __m128i result; __m128i prim_poly; @@ -433,11 +433,11 @@ gf_w16_clm_multiply_2 (gf_t *gf, gf_val_32_t a16, gf_val_32_t b16) rv = ((gf_val_32_t)_mm_extract_epi32(result, 0)); - -#endif return rv; } +#endif +#if defined(INTEL_SSE4_PCLMUL) static inline gf_val_32_t @@ -445,8 +445,6 @@ gf_w16_clm_multiply_3 (gf_t *gf, gf_val_32_t a16, gf_val_32_t b16) { gf_val_32_t rv = 0; -#if defined(INTEL_SSE4_PCLMUL) - __m128i a, b; __m128i result; __m128i prim_poly; @@ -473,11 +471,11 @@ gf_w16_clm_multiply_3 (gf_t *gf, gf_val_32_t a16, gf_val_32_t b16) rv = ((gf_val_32_t)_mm_extract_epi32(result, 0)); - -#endif return rv; } +#endif +#if defined(INTEL_SSE4_PCLMUL) static inline gf_val_32_t @@ -485,8 +483,6 @@ gf_w16_clm_multiply_4 (gf_t *gf, gf_val_32_t a16, gf_val_32_t b16) { gf_val_32_t rv = 0; -#if defined(INTEL_SSE4_PCLMUL) - __m128i a, b; __m128i result; __m128i prim_poly; @@ -515,10 +511,9 @@ gf_w16_clm_multiply_4 (gf_t *gf, gf_val_32_t a16, gf_val_32_t b16) rv = ((gf_val_32_t)_mm_extract_epi32(result, 0)); - -#endif return rv; } +#endif static @@ -556,25 +551,27 @@ static int gf_w16_cfm_init(gf_t *gf) { #if defined(INTEL_SSE4_PCLMUL) - gf_internal_t *h; + if (gf_cpu_supports_intel_pclmul) { + gf_internal_t *h; - h = (gf_internal_t *) gf->scratch; - - /*Ben: Determining how many reductions to do */ - - if ((0xfe00 & h->prim_poly) == 0) { - SET_FUNCTION(gf,multiply,w32,gf_w16_clm_multiply_2) - SET_FUNCTION(gf,multiply_region,w32,gf_w16_clm_multiply_region_from_single_2) - } else if((0xf000 & h->prim_poly) == 0) { - SET_FUNCTION(gf,multiply,w32,gf_w16_clm_multiply_3) - SET_FUNCTION(gf,multiply_region,w32,gf_w16_clm_multiply_region_from_single_3) - } else if ((0xe000 & h->prim_poly) == 0) { - SET_FUNCTION(gf,multiply,w32,gf_w16_clm_multiply_4) - SET_FUNCTION(gf,multiply_region,w32,gf_w16_clm_multiply_region_from_single_4) - } else { - return 0; - } - return 1; + h = (gf_internal_t *) gf->scratch; + + /*Ben: Determining how many reductions to do */ + + if ((0xfe00 & h->prim_poly) == 0) { + SET_FUNCTION(gf,multiply,w32,gf_w16_clm_multiply_2) + SET_FUNCTION(gf,multiply_region,w32,gf_w16_clm_multiply_region_from_single_2) + } else if((0xf000 & h->prim_poly) == 0) { + SET_FUNCTION(gf,multiply,w32,gf_w16_clm_multiply_3) + SET_FUNCTION(gf,multiply_region,w32,gf_w16_clm_multiply_region_from_single_3) + } else if ((0xe000 & h->prim_poly) == 0) { + SET_FUNCTION(gf,multiply,w32,gf_w16_clm_multiply_4) + SET_FUNCTION(gf,multiply_region,w32,gf_w16_clm_multiply_region_from_single_4) + } else { + return 0; + } + return 1; + } #endif return 0; @@ -688,10 +685,9 @@ int gf_w16_log_init(gf_t *gf) if (check) { if (h->mult_type != GF_MULT_LOG_TABLE) { - -#if defined(INTEL_SSE4_PCLMUL) - return gf_w16_cfm_init(gf); -#endif + if (gf_cpu_supports_intel_pclmul) { + return gf_w16_cfm_init(gf); + } return gf_w16_shift_init(gf); } else { _gf_errno = GF_E_LOGPOLY; @@ -948,11 +944,11 @@ gf_w16_table_lazy_multiply_region(gf_t *gf, void *src, void *dest, gf_val_32_t v gf_do_final_region_alignment(&rd); } +#ifdef INTEL_SSSE3 static void gf_w16_split_4_16_lazy_sse_multiply_region(gf_t *gf, void *src, void *dest, gf_val_32_t val, int bytes, int xor) { -#ifdef INTEL_SSSE3 uint64_t i, j, *s64, *d64, *top64;; uint64_t c, prod; uint8_t low[4][16]; @@ -1078,14 +1074,14 @@ gf_w16_split_4_16_lazy_sse_multiply_region(gf_t *gf, void *src, void *dest, gf_v } gf_do_final_region_alignment(&rd); -#endif } +#endif +#ifdef INTEL_SSSE3 static void gf_w16_split_4_16_lazy_sse_altmap_multiply_region(gf_t *gf, void *src, void *dest, gf_val_32_t val, int bytes, int xor) { -#ifdef INTEL_SSSE3 uint64_t i, j, *s64, *d64, *top64;; uint64_t c, prod; uint8_t low[4][16]; @@ -1187,8 +1183,8 @@ gf_w16_split_4_16_lazy_sse_altmap_multiply_region(gf_t *gf, void *src, void *des } gf_do_final_region_alignment(&rd); -#endif } +#endif uint32_t gf_w16_split_8_8_multiply(gf_t *gf, gf_val_32_t a, gf_val_32_t b) @@ -1216,21 +1212,11 @@ int gf_w16_split_init(gf_t *gf) { gf_internal_t *h; struct gf_w16_split_8_8_data *d8; - int i, j, exp, issse3; - int isneon = 0; + int i, j, exp; uint32_t p, basep, tmp; h = (gf_internal_t *) gf->scratch; -#ifdef INTEL_SSSE3 - issse3 = 1; -#else - issse3 = 0; -#endif -#ifdef ARM_NEON - isneon = 1; -#endif - if (h->arg1 == 8 && h->arg2 == 8) { d8 = (struct gf_w16_split_8_8_data *) h->private; basep = 1; @@ -1273,36 +1259,45 @@ int gf_w16_split_init(gf_t *gf) /* Defaults */ - if (issse3) { +#ifdef INTEL_SSSE3 + if (gf_cpu_supports_intel_ssse3) { SET_FUNCTION(gf,multiply_region,w32,gf_w16_split_4_16_lazy_sse_multiply_region) - } else if (isneon) { -#ifdef ARM_NEON + } else { +#elif ARM_NEON + if (gf_cpu_supports_arm_neon) { gf_w16_neon_split_init(gf); -#endif } else { +#endif SET_FUNCTION(gf,multiply_region,w32,gf_w16_split_8_16_lazy_multiply_region) +#if defined(INTEL_SSSE3) || defined(ARM_NEON) } - +#endif if ((h->arg1 == 8 && h->arg2 == 16) || (h->arg2 == 8 && h->arg1 == 16)) { SET_FUNCTION(gf,multiply_region,w32,gf_w16_split_8_16_lazy_multiply_region) } else if ((h->arg1 == 4 && h->arg2 == 16) || (h->arg2 == 4 && h->arg1 == 16)) { - if (issse3 || isneon) { +#if defined(INTEL_SSSE3) || defined(ARM_NEON) + if (gf_cpu_supports_intel_ssse3 || gf_cpu_supports_arm_neon) { if(h->region_type & GF_REGION_ALTMAP && h->region_type & GF_REGION_NOSIMD) SET_FUNCTION(gf,multiply_region,w32,gf_w16_split_4_16_lazy_nosse_altmap_multiply_region) else if(h->region_type & GF_REGION_NOSIMD) SET_FUNCTION(gf,multiply_region,w32,gf_w16_split_4_16_lazy_multiply_region) - else if(h->region_type & GF_REGION_ALTMAP && issse3) +#if defined(INTEL_SSSE3) + else if(h->region_type & GF_REGION_ALTMAP && gf_cpu_supports_intel_ssse3) SET_FUNCTION(gf,multiply_region,w32,gf_w16_split_4_16_lazy_sse_altmap_multiply_region) +#endif } else { +#endif if(h->region_type & GF_REGION_SIMD) return 0; else if(h->region_type & GF_REGION_ALTMAP) SET_FUNCTION(gf,multiply_region,w32,gf_w16_split_4_16_lazy_nosse_altmap_multiply_region) else SET_FUNCTION(gf,multiply_region,w32,gf_w16_split_4_16_lazy_multiply_region) +#if defined(INTEL_SSSE3) || defined(ARM_NEON) } +#endif } return 1; @@ -1846,26 +1841,28 @@ int gf_w16_bytwo_init(gf_t *gf) if (h->mult_type == GF_MULT_BYTWO_p) { SET_FUNCTION(gf,multiply,w32,gf_w16_bytwo_p_multiply) #ifdef INTEL_SSE2 - if (h->region_type & GF_REGION_NOSIMD) - SET_FUNCTION(gf,multiply_region,w32,gf_w16_bytwo_p_nosse_multiply_region) - else - SET_FUNCTION(gf,multiply_region,w32,gf_w16_bytwo_p_sse_multiply_region) - #else + if (gf_cpu_supports_intel_sse2 && !(h->region_type & GF_REGION_NOSIMD)) { + SET_FUNCTION(gf,multiply_region,w32,gf_w16_bytwo_p_sse_multiply_region) + } else { + #endif SET_FUNCTION(gf,multiply_region,w32,gf_w16_bytwo_p_nosse_multiply_region) if(h->region_type & GF_REGION_SIMD) return 0; + #ifdef INTEL_SSE2 + } #endif } else { SET_FUNCTION(gf,multiply,w32,gf_w16_bytwo_b_multiply) #ifdef INTEL_SSE2 - if (h->region_type & GF_REGION_NOSIMD) - SET_FUNCTION(gf,multiply_region,w32,gf_w16_bytwo_b_nosse_multiply_region) - else + if (gf_cpu_supports_intel_sse2 && !(h->region_type & GF_REGION_NOSIMD)) { SET_FUNCTION(gf,multiply_region,w32,gf_w16_bytwo_b_sse_multiply_region) - #else + } else { + #endif SET_FUNCTION(gf,multiply_region,w32,gf_w16_bytwo_b_nosse_multiply_region) if(h->region_type & GF_REGION_SIMD) return 0; + #ifdef INTEL_SSE2 + } #endif } |