tommy/libpostal
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[revert] reverting the commits from #578, leaving header file in repo for the moment

Browse Source
This commit is contained in:
Al
2023-07-06 01:49:02 -04:00
parent c76d020c18
commit 57eaa414ce
6 changed files with 29 additions and 103 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@@ -113,6 +113,8 @@ brew install curl autoconf automake libtool pkg-config
Then to install the C library: Then to install the C library:
If you're using an M1 Mac, add `--disable-sse2` to the `./configure` command. This will result in poorer performance but the build will succeed.
``` ```
git clone https://github.com/openvenues/libpostal git clone https://github.com/openvenues/libpostal
cd libpostal cd libpostal

55
configure.ac
View File

@@ -84,57 +84,20 @@ AS_IF([test "x$FOUND_SHUF" = xyes], [AC_DEFINE([HAVE_SHUF], [1], [shuf availabl
AS_IF([test "x$FOUND_GSHUF" = xyes], [AC_DEFINE([HAVE_GSHUF], [1], [gshuf available])]) AS_IF([test "x$FOUND_GSHUF" = xyes], [AC_DEFINE([HAVE_GSHUF], [1], [gshuf available])])
# ------------------------------------------------------------------ # ------------------------------------------------------------------
# Architecture-specific options # Checks for SSE2 build
# ------------------------------------------------------------------ # ------------------------------------------------------------------
# allow enabling hardware optimization on any system:
case "$host_cpu" in
arm*|aarch64*)
enable_arm_neon=yes
enable_intel_sse=no
AC_DEFINE([ARM_NEON], [1],
[Enable ARM_NEON optimizations])
;;
i?86|x86_64)
enable_intel_sse=yes
enable_arm_neon=no
AC_DEFINE([INTEL_SSE], [1],
[Enable Intel SSE optimizations])
;;
esac
AC_ARG_ENABLE([neon],
AS_HELP_STRING([[[--disable-neon]]],
[Disable ARM NEON hardware optimizations]),
[
enable_arm_neon=no
AC_DEFINE([ARM_NEON], [0],
[Disable ARM_NEON optimizations])
])
AC_ARG_ENABLE([sse2], AC_ARG_ENABLE([sse2],
AS_HELP_STRING([[[--disable-sse2]]], AS_HELP_STRING(
[Disable Intel SSE2 hardware optimizations]), [--disable-sse2],
[ [disable SSE2 optimization routines]
enable_intel_sse=no )
AC_DEFINE([INTEL_SSE], [0], )
[Disable INTEL_SSE optimizations])
AS_IF([test "x$enable_sse2" != "xno"], [
CFLAGS="-mfpmath=sse -msse2 -DUSE_SSE ${CFLAGS}"
]) ])
SIMDFLAGS=""
AS_IF([test "x$enable_intel_sse" != "xno"], [
SIMDFLAGS="-mfpmath=sse -msse2 -DINTEL_SSE"
])
AS_IF([test "x$enable_arm_neon" != "xno"], [
SIMDFLAGS="-march=armv8-a+fp+simd+crypto+crc -DARM_NEON"
])
CFLAGS="${SIMDFLAGS} ${CFLAGS}"
AC_SUBST([SIMDFLAGS], [$SIMDFLAGS])
AC_CHECK_HEADER(cblas.h, [AX_CBLAS]) AC_CHECK_HEADER(cblas.h, [AX_CBLAS])
AC_ARG_ENABLE([data-download], AC_ARG_ENABLE([data-download],

16
src/crf_context.c
View File

@@ -40,7 +40,7 @@ crf_context_t *crf_context_new(int flag, size_t L, size_t T) {
} }
if (context->flag & CRF_CONTEXT_MARGINALS) { if (context->flag & CRF_CONTEXT_MARGINALS) {
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
context->exp_state = double_matrix_new_aligned(T, L, 32); context->exp_state = double_matrix_new_aligned(T, L, 32);
if (context->exp_state == NULL) goto exit_context_created; if (context->exp_state == NULL) goto exit_context_created;
double_matrix_zero(context->exp_state); double_matrix_zero(context->exp_state);
@@ -52,7 +52,7 @@ crf_context_t *crf_context_new(int flag, size_t L, size_t T) {
context->mexp_state = double_matrix_new_zeros(T, L); context->mexp_state = double_matrix_new_zeros(T, L);
if (context->mexp_state == NULL) goto exit_context_created; if (context->mexp_state == NULL) goto exit_context_created;
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
context->exp_state_trans = double_matrix_new_aligned(T, L * L, 32); context->exp_state_trans = double_matrix_new_aligned(T, L * L, 32);
if (context->exp_state_trans == NULL) goto exit_context_created; if (context->exp_state_trans == NULL) goto exit_context_created;
double_matrix_zero(context->exp_state_trans); double_matrix_zero(context->exp_state_trans);
@@ -64,7 +64,7 @@ crf_context_t *crf_context_new(int flag, size_t L, size_t T) {
context->mexp_state_trans = double_matrix_new_zeros(T, L * L); context->mexp_state_trans = double_matrix_new_zeros(T, L * L);
if (context->mexp_state_trans == NULL) goto exit_context_created; if (context->mexp_state_trans == NULL) goto exit_context_created;
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
context->exp_trans = double_matrix_new_aligned(L, L, 32); context->exp_trans = double_matrix_new_aligned(L, L, 32);
if (context->exp_trans == NULL) goto exit_context_created; if (context->exp_trans == NULL) goto exit_context_created;
double_matrix_zero(context->exp_trans); double_matrix_zero(context->exp_trans);
@@ -130,13 +130,13 @@ bool crf_context_set_num_items(crf_context_t *self, size_t T) {
if (self->flag & CRF_CONTEXT_MARGINALS && if (self->flag & CRF_CONTEXT_MARGINALS &&
( (
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
!double_matrix_resize_aligned(self->exp_state, T, L, 32) || !double_matrix_resize_aligned(self->exp_state, T, L, 32) ||
#else #else
!double_matrix_resize(self->exp_state, T, L) || !double_matrix_resize(self->exp_state, T, L) ||
#endif #endif
!double_matrix_resize(self->mexp_state, T, L) || !double_matrix_resize(self->mexp_state, T, L) ||
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
!double_matrix_resize_aligned(self->exp_state_trans, T, L * L, 32) || !double_matrix_resize_aligned(self->exp_state_trans, T, L * L, 32) ||
#else #else
!double_matrix_resize(self->exp_state_trans, T, L * L) || !double_matrix_resize(self->exp_state_trans, T, L * L) ||
@@ -184,7 +184,7 @@ void crf_context_destroy(crf_context_t *self) {
} }
if (self->exp_state != NULL) { if (self->exp_state != NULL) {
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
double_matrix_destroy_aligned(self->exp_state); double_matrix_destroy_aligned(self->exp_state);
#else #else
double_matrix_destroy(self->exp_state); double_matrix_destroy(self->exp_state);
@@ -200,7 +200,7 @@ void crf_context_destroy(crf_context_t *self) {
} }
if (self->exp_state_trans != NULL) { if (self->exp_state_trans != NULL) {
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
double_matrix_destroy_aligned(self->exp_state_trans); double_matrix_destroy_aligned(self->exp_state_trans);
#else #else
double_matrix_destroy(self->exp_state_trans); double_matrix_destroy(self->exp_state_trans);
@@ -216,7 +216,7 @@ void crf_context_destroy(crf_context_t *self) {
} }
if (self->exp_trans != NULL) { if (self->exp_trans != NULL) {
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
double_matrix_destroy_aligned(self->exp_trans); double_matrix_destroy_aligned(self->exp_trans);
#else #else
double_matrix_destroy(self->exp_trans); double_matrix_destroy(self->exp_trans);

6
src/vector_math.h
View File

@@ -8,10 +8,8 @@
#define ks_lt_index(a, b) ((a).value < (b).value) #define ks_lt_index(a, b) ((a).value < (b).value)
#if defined(INTEL_SSE) #if defined(USE_SSE)
#include <emmintrin.h> #include <emmintrin.h>
#elif defined(ARM_NEON)
#include "sse2neon.h"
#endif #endif
/* /*
@@ -340,7 +338,7 @@
#if defined(INTEL_SSE) || defined(ARM_NEON) #if defined(USE_SSE)
/* /*
From https://github.com/herumi/fmath/blob/master/fastexp.cpp From https://github.com/herumi/fmath/blob/master/fastexp.cpp

2
test/Makefile.am
View File

@@ -5,7 +5,7 @@ CFLAGS_O2 = $(CFLAGS_BASE) -O2
CFLAGS_O3 = $(CFLAGS_BASE) -O3 CFLAGS_O3 = $(CFLAGS_BASE) -O3
DEFAULT_INCLUDES = -I.. -I/usr/local/include DEFAULT_INCLUDES = -I.. -I/usr/local/include
CFLAGS = $(SIMDFLAGS) $(CFLAGS_BASE) CFLAGS = $(CFLAGS_BASE)
TESTS = test_libpostal TESTS = test_libpostal
noinst_PROGRAMS = test_libpostal noinst_PROGRAMS = test_libpostal

55
windows/configure.ac
View File

@@ -73,57 +73,20 @@ AS_IF([test "x$FOUND_SHUF" = xyes], [AC_DEFINE([HAVE_SHUF], [1], [shuf availabl
AS_IF([test "x$FOUND_GSHUF" = xyes], [AC_DEFINE([HAVE_GSHUF], [1], [gshuf available])]) AS_IF([test "x$FOUND_GSHUF" = xyes], [AC_DEFINE([HAVE_GSHUF], [1], [gshuf available])])
# ------------------------------------------------------------------ # ------------------------------------------------------------------
# Architecture-specific options # Checks for SSE2 build
# ------------------------------------------------------------------ # ------------------------------------------------------------------
# allow enabling hardware optimization on any system:
case "$host_cpu" in
arm*|aarch64*)
enable_arm_neon=yes
enable_intel_sse=no
AC_DEFINE([ARM_NEON], [1],
[Enable ARM_NEON optimizations])
;;
i?86|x86_64)
enable_intel_sse=yes
enable_arm_neon=no
AC_DEFINE([INTEL_SSE], [1],
[Enable Intel SSE optimizations])
;;
esac
AC_ARG_ENABLE([neon],
AS_HELP_STRING([[[--disable-neon]]],
[Disable ARM NEON hardware optimizations]),
[
enable_arm_neon=no
AC_DEFINE([ARM_NEON], [0],
[Disable ARM_NEON optimizations])
])
AC_ARG_ENABLE([sse2], AC_ARG_ENABLE([sse2],
AS_HELP_STRING([[[--disable-sse2]]], AS_HELP_STRING(
[Disable Intel SSE2 hardware optimizations]), [--disable-sse2],
[ [disable SSE2 optimization routines]
enable_intel_sse=no )
AC_DEFINE([INTEL_SSE], [0], )
[Disable INTEL_SSE optimizations])
AS_IF([test "x$enable_sse2" != "xno"], [
CFLAGS="-mfpmath=sse -msse2 -DUSE_SSE ${CFLAGS}"
]) ])
SIMDFLAGS=""
AS_IF([test "x$enable_intel_sse" != "xno"], [
SIMDFLAGS="-mfpmath=sse -msse2 -DINTEL_SSE"
])
AS_IF([test "x$enable_arm_neon" != "xno"], [
SIMDFLAGS="-march=armv8-a+fp+simd+crypto+crc -DARM_NEON"
])
CFLAGS="${SIMDFLAGS} ${CFLAGS}"
AC_SUBST([SIMDFLAGS], [$SIMDFLAGS])
AC_CHECK_HEADER(cblas.h, [AX_CBLAS]) AC_CHECK_HEADER(cblas.h, [AX_CBLAS])
AC_ARG_ENABLE([data-download], AC_ARG_ENABLE([data-download],