Using fast_float library for faster parsing of 64 decimal strings. (#11884)

Fixes #8825 We're using the fast_float library[1] in our (compiled-in) floating-point fast_float_strtod implementation for faster and more portable parsing of 64 decimal strings. The single file fast_float.h is an amalgamation of the entire library, which can be (re)generated with the amalgamate.py script (from the fast_float repository) via the command: ``` python3 ./script/amalgamate.py --license=MIT > $REDIS_SRC/deps/fast_float/fast_float.h ``` [1]: https://github.com/fastfloat/fast_float The used commit from fast_float library was the one from https://github.com/fastfloat/fast_float/releases/tag/v3.10.1 --------- Co-authored-by: fcostaoliveira <filipe@redis.com>
2024-09-15 14:37:29 +01:00 · 2024-09-15 14:37:29 +01:00 · af7fca797a
parent 9146ac050b
commit af7fca797a
15 changed files with 3979 additions and 30 deletions
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@ -62,7 +62,7 @@ jobs:
    - uses: actions/checkout@v4
    - name: make
      run: |
-        sudo apt-get update && sudo apt-get install libc6-dev-i386
+        sudo apt-get update && sudo apt-get install libc6-dev-i386 gcc-multilib g++-multilib
        make REDIS_CFLAGS='-Werror' 32bit

  build-libc-malloc:
@ -79,7 +79,7 @@ jobs:
    - uses: actions/checkout@v4
    - name: make
      run: |
-        dnf -y install which gcc make
+        dnf -y install which gcc gcc-c++ make
        make REDIS_CFLAGS='-Werror'

  build-old-chain-jemalloc:
@ -96,6 +96,7 @@ jobs:
        apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 40976EAF437D05B5
        apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 3B4FE6ACC0B21F32
        apt-get update
-        apt-get install -y make gcc-4.8
+        apt-get install -y make gcc-4.8 g++-4.8
        update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.8 100
+        update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.8 100
        make CC=gcc REDIS_CFLAGS='-Werror'
--- a/.github/workflows/daily.yml
+++ b/.github/workflows/daily.yml
@ -94,8 +94,9 @@ jobs:
        ref: ${{ env.GITHUB_HEAD_REF }}
    - name: make
      run: |
-        apt-get update && apt-get install -y make gcc-13
+        apt-get update && apt-get install -y make gcc-13 g++-13
        update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 100
+        update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 100
        make CC=gcc REDIS_CFLAGS='-Werror -DREDIS_TEST -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=3'
    - name: testprep
      run: apt-get install -y tcl8.6 tclx procps
@ -211,7 +212,7 @@ jobs:
        ref: ${{ env.GITHUB_HEAD_REF }}
    - name: make
      run: |
-        sudo apt-get update && sudo apt-get install libc6-dev-i386
+        sudo apt-get update && sudo apt-get install libc6-dev-i386 g++ gcc-multilib g++-multilib
        make 32bit REDIS_CFLAGS='-Werror -DREDIS_TEST'
    - name: testprep
      run: sudo apt-get install tcl8.6 tclx
@ -456,7 +457,7 @@ jobs:
    - name: testprep
      run: |
        sudo apt-get update
-        sudo apt-get install tcl8.6 tclx valgrind -y
+        sudo apt-get install tcl8.6 tclx valgrind g++ -y
    - name: test
      if: true && !contains(github.event.inputs.skiptests, 'redis')
      run: ./runtest --valgrind --no-latency --verbose --clients 1 --timeout 2400 --dump-logs ${{github.event.inputs.test_args}}
@ -521,7 +522,7 @@ jobs:
    - name: testprep
      run: |
        sudo apt-get update
-        sudo apt-get install tcl8.6 tclx valgrind -y
+        sudo apt-get install tcl8.6 tclx valgrind g++ -y
    - name: test
      if: true && !contains(github.event.inputs.skiptests, 'redis')
      run: ./runtest --valgrind --no-latency --verbose --clients 1 --timeout 2400 --dump-logs ${{github.event.inputs.test_args}}
@ -678,7 +679,7 @@ jobs:
        ref: ${{ env.GITHUB_HEAD_REF }}
    - name: make
      run: |
-        dnf -y install which gcc make
+        dnf -y install which gcc make g++
        make REDIS_CFLAGS='-Werror'
    - name: testprep
      run: |
@ -720,7 +721,7 @@ jobs:
        ref: ${{ env.GITHUB_HEAD_REF }}
    - name: make
      run: |
-        dnf -y install which gcc make openssl-devel openssl
+        dnf -y install which gcc make openssl-devel openssl g++
        make BUILD_TLS=module REDIS_CFLAGS='-Werror'
    - name: testprep
      run: |
@ -767,7 +768,7 @@ jobs:
        ref: ${{ env.GITHUB_HEAD_REF }}
    - name: make
      run: |
-        dnf -y install which gcc make openssl-devel openssl
+        dnf -y install which gcc make openssl-devel openssl g++
        make BUILD_TLS=module REDIS_CFLAGS='-Werror'
    - name: testprep
      run: |
@ -907,6 +908,9 @@ jobs:
      (github.event_name == 'workflow_dispatch' || (github.event_name != 'workflow_dispatch' && github.repository == 'redis/redis')) &&
      !contains(github.event.inputs.skipjobs, 'freebsd')
    timeout-minutes: 14400
+    env:
+      CC: clang
+      CXX: clang++
    steps:
    - name: prep
      if: github.event_name == 'workflow_dispatch'
@ -925,7 +929,7 @@ jobs:
        version: 13.2
        shell: bash
        run: |
-          sudo pkg install -y bash gmake lang/tcl86 lang/tclx
+          sudo pkg install -y bash gmake lang/tcl86 lang/tclx gcc
          gmake
          ./runtest --single unit/keyspace --single unit/auth --single unit/networking --single unit/protocol

@ -1080,8 +1084,9 @@ jobs:
        apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 40976EAF437D05B5
        apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 3B4FE6ACC0B21F32
        apt-get update
-        apt-get install -y make gcc-4.8
+        apt-get install -y make gcc-4.8 g++-4.8
        update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.8 100
+        update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.8 100
        make CC=gcc REDIS_CFLAGS='-Werror'
    - name: testprep
      run: apt-get install -y tcl tcltls tclx
@ -1128,9 +1133,10 @@ jobs:
        apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 40976EAF437D05B5
        apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 3B4FE6ACC0B21F32
        apt-get update
-        apt-get install -y make gcc-4.8 openssl libssl-dev
+        apt-get install -y make gcc-4.8 g++-4.8 openssl libssl-dev
        update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.8 100
-        make CC=gcc BUILD_TLS=module REDIS_CFLAGS='-Werror'
+        update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.8 100
+        make CC=gcc CXX=g++ BUILD_TLS=module REDIS_CFLAGS='-Werror'
    - name: testprep
      run: |
        apt-get install -y tcl tcltls tclx
@ -1182,8 +1188,9 @@ jobs:
        apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 40976EAF437D05B5
        apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 3B4FE6ACC0B21F32
        apt-get update
-        apt-get install -y make gcc-4.8 openssl libssl-dev
+        apt-get install -y make gcc-4.8 g++-4.8 openssl libssl-dev
        update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.8 100
+        update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.8 100
        make BUILD_TLS=module CC=gcc REDIS_CFLAGS='-Werror'
    - name: testprep
      run: |
--- a/.gitignore
+++ b/.gitignore
@ -30,6 +30,7 @@ deps/lua/src/luac
 deps/lua/src/liblua.a
 deps/hdr_histogram/libhdrhistogram.a
 deps/fpconv/libfpconv.a
+deps/fast_float/libfast_float.a
 tests/tls/*
 .make-*
 .prerequisites
--- a/deps/Makefile
+++ b/deps/Makefile
@ -42,6 +42,7 @@ distclean:
 	-(cd jemalloc && [ -f Makefile ] && $(MAKE) distclean) > /dev/null || true
 	-(cd hdr_histogram && $(MAKE) clean) > /dev/null || true
 	-(cd fpconv && $(MAKE) clean) > /dev/null || true
+	-(cd fast_float && $(MAKE) clean) > /dev/null || true
 	-(rm -f .make-*)

 .PHONY: distclean
@ -74,6 +75,12 @@ fpconv: .make-prerequisites

 .PHONY: fpconv

+fast_float: .make-prerequisites
+	@printf '%b %b\n' $(MAKECOLOR)MAKE$(ENDCOLOR) $(BINCOLOR)$@$(ENDCOLOR)
+	cd fast_float && $(MAKE) libfast_float
+
+.PHONY: fast_float
+
 ifeq ($(uname_S),SunOS)
 	# Make isinf() available
 	LUA_CFLAGS= -D__C99FEATURES__=1
--- a/deps/fast_float/Makefile
+++ b/deps/fast_float/Makefile
@ -0,0 +1,24 @@
+# Fallback to gcc/g++ when $CC or $CXX is not in $PATH.
+CC ?= gcc
+CXX ?= g++
+
+CFLAGS=-Wall -O3
+# This avoids loosing the fastfloat specific compile flags when we override the CFLAGS via the main project
+FASTFLOAT_CFLAGS=-std=c++11 -DFASTFLOAT_ALLOWS_LEADING_PLUS
+LDFLAGS=
+
+libfast_float: fast_float_strtod.o
+	$(AR) -r libfast_float.a fast_float_strtod.o
+
+32bit: CFLAGS += -m32
+32bit: LDFLAGS += -m32
+32bit: libfast_float
+
+fast_float_strtod.o: fast_float_strtod.cpp
+	$(CXX) $(CFLAGS) $(FASTFLOAT_CFLAGS) -c fast_float_strtod.cpp $(LDFLAGS)
+
+clean:
+	rm -f *.o
+	rm -f *.a
+	rm -f *.h.gch
+	rm -rf *.dSYM
--- a/deps/fast_float/README.md
+++ b/deps/fast_float/README.md
@ -0,0 +1,21 @@
+README for fast_float v6.1.4
+
+----------------------------------------------
+
+We're using the fast_float library[1] in our (compiled-in)
+floating-point fast_float_strtod implementation for faster and more
+portable parsing of 64 decimal strings.
+
+The single file fast_float.h is an amalgamation of the entire library,
+which can be (re)generated with the amalgamate.py script (from the
+fast_float repository) via the command
+
+```
+git clone https://github.com/fastfloat/fast_float
+cd fast_float
+git checkout v6.1.4
+python3 ./script/amalgamate.py --license=MIT \
+  > $REDIS_SRC/deps/fast_float/fast_float.h
+```
+
+[1]: https://github.com/fastfloat/fast_float
--- a/deps/fast_float/fast_float.h
+++ b/deps/fast_float/fast_float.h
--- a/deps/fast_float/fast_float_strtod.cpp
+++ b/deps/fast_float/fast_float_strtod.cpp
@ -0,0 +1,32 @@
+#include "fast_float.h"
+#include <iostream>
+#include <string>
+#include <system_error>
+#include <cerrno>
+
+/* Convert NPTR to a double using the fast_float library.
+ *
+ * This function behaves similarly to the standard strtod function, converting
+ * the initial portion of the string pointed to by `nptr` to a `double` value,
+ * using the fast_float library for high performance. If the conversion fails,
+ * errno is set to EINVAL error code.
+ *
+ * @param nptr   A pointer to the null-terminated byte string to be interpreted.
+ * @param endptr A pointer to a pointer to character. If `endptr` is not NULL,
+ *               it will point to the character after the last character used
+ *               in the conversion.
+ * @return       The converted value as a double. If no valid conversion could
+ *               be performed, returns 0.0.
+ * If ENDPTR is not NULL, a pointer to the character after the last one used
+ * in the number is put in *ENDPTR.  */
+extern "C" double fast_float_strtod(const char *nptr, char **endptr) {
+  double result = 0.0;
+  auto answer = fast_float::from_chars(nptr, nptr + strlen(nptr), result);
+  if (answer.ec != std::errc()) {
+    errno = EINVAL;  // Fallback to  for other errors
+  }
+  if (endptr != NULL) {
+    *endptr = (char *)answer.ptr;
+  }
+  return result;
+}
--- a/deps/fast_float/fast_float_strtod.h
+++ b/deps/fast_float/fast_float_strtod.h
@ -0,0 +1,15 @@
+
+#ifndef __FAST_FLOAT_STRTOD_H__
+#define __FAST_FLOAT_STRTOD_H__
+
+#if defined(__cplusplus)
+extern "C"
+{
+#endif
+    double fast_float_strtod(const char *in, char **out);
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif /* __FAST_FLOAT_STRTOD_H__ */
--- a/src/Makefile
+++ b/src/Makefile
@ -34,7 +34,7 @@ endif
 ifneq ($(OPTIMIZATION),-O0)
 	OPTIMIZATION+=-fno-omit-frame-pointer
 endif
-DEPENDENCY_TARGETS=hiredis linenoise lua hdr_histogram fpconv
+DEPENDENCY_TARGETS=hiredis linenoise lua hdr_histogram fpconv fast_float
 NODEPS:=clean distclean

 # Default settings
@ -127,7 +127,7 @@ endif

 FINAL_CFLAGS=$(STD) $(WARN) $(OPT) $(DEBUG) $(CFLAGS) $(REDIS_CFLAGS)
 FINAL_LDFLAGS=$(LDFLAGS) $(OPT) $(REDIS_LDFLAGS) $(DEBUG)
-FINAL_LIBS=-lm
+FINAL_LIBS=-lm -lstdc++
 DEBUG=-g -ggdb

 # Linux ARM32 needs -latomic at linking time
@ -235,7 +235,7 @@ ifdef OPENSSL_PREFIX
 endif

 # Include paths to dependencies
-FINAL_CFLAGS+= -I../deps/hiredis -I../deps/linenoise -I../deps/lua/src -I../deps/hdr_histogram -I../deps/fpconv
+FINAL_CFLAGS+= -I../deps/hiredis -I../deps/linenoise -I../deps/lua/src -I../deps/hdr_histogram -I../deps/fpconv -I../deps/fast_float

 # Determine systemd support and/or build preference (defaulting to auto-detection)
 BUILD_WITH_SYSTEMD=no
@ -409,7 +409,7 @@ endif

 # redis-server
 $(REDIS_SERVER_NAME): $(REDIS_SERVER_OBJ)
-	$(REDIS_LD) -o $@ $^ ../deps/hiredis/libhiredis.a ../deps/lua/src/liblua.a ../deps/hdr_histogram/libhdrhistogram.a ../deps/fpconv/libfpconv.a $(FINAL_LIBS)
+	$(REDIS_LD) -o $@ $^ ../deps/hiredis/libhiredis.a ../deps/lua/src/liblua.a ../deps/hdr_histogram/libhdrhistogram.a ../deps/fpconv/libfpconv.a ../deps/fast_float/libfast_float.a $(FINAL_LIBS)

 # redis-sentinel
 $(REDIS_SENTINEL_NAME): $(REDIS_SERVER_NAME)
--- a/src/debug.c
+++ b/src/debug.c
@ -15,6 +15,7 @@
 #include "bio.h"
 #include "quicklist.h"
 #include "fpconv_dtoa.h"
+#include "fast_float_strtod.h"
 #include "cluster.h"
 #include "threads_mngr.h"
 #include "script.h"
@ -833,7 +834,7 @@ NULL
            addReplyError(c,"Wrong protocol type name. Please use one of the following: string|integer|double|bignum|null|array|set|map|attrib|push|verbatim|true|false");
        }
    } else if (!strcasecmp(c->argv[1]->ptr,"sleep") && c->argc == 3) {
-        double dtime = strtod(c->argv[2]->ptr,NULL);
+        double dtime = fast_float_strtod(c->argv[2]->ptr,NULL);
        long long utime = dtime*1000000;
        struct timespec tv;

--- a/src/resp_parser.c
+++ b/src/resp_parser.c
@ -34,6 +34,7 @@
 * ----------------------------------------------------------------------------------------
 */

+#include "fast_float_strtod.h"
 #include "resp_parser.h"
 #include "server.h"

@ -132,7 +133,7 @@ static int parseDouble(ReplyParser *parser, void *p_ctx) {
    if (len <= MAX_LONG_DOUBLE_CHARS) {
        memcpy(buf,proto+1,len);
        buf[len] = '\0';
-        d = strtod(buf,NULL); /* We expect a valid representation. */
+        d = fast_float_strtod(buf,NULL); /* We expect a valid representation. */
    } else {
        d = 0;
    }
--- a/src/sort.c
+++ b/src/sort.c
@ -7,7 +7,7 @@
 * (RSALv2) or the Server Side Public License v1 (SSPLv1).
 */

-
+#include "fast_float_strtod.h"
 #include "server.h"
 #include "pqsort.h" /* Partial qsort for SORT+LIMIT */
 #include <math.h> /* isnan() */
@ -472,7 +472,7 @@ void sortCommandGeneric(client *c, int readonly) {
                if (sdsEncodedObject(byval)) {
                    char *eptr;

-                    vector[j].u.score = strtod(byval->ptr,&eptr);
+                    vector[j].u.score = fast_float_strtod(byval->ptr,&eptr);
                    if (eptr[0] != '\0' || errno == ERANGE ||
                        isnan(vector[j].u.score))
                    {
--- a/src/t_zset.c
+++ b/src/t_zset.c
@ -55,7 +55,7 @@
 * c) there is a back pointer, so it's a doubly linked list with the back
 * pointers being only at "level 1". This allows to traverse the list
 * from tail to head, useful for ZREVRANGE. */
-
+#include "fast_float_strtod.h"
 #include "server.h"
 #include "intset.h"  /* Compact integer set structure */
 #include <math.h>
@ -566,11 +566,11 @@ static int zslParseRange(robj *min, robj *max, zrangespec *spec) {
        spec->min = (long)min->ptr;
    } else {
        if (((char*)min->ptr)[0] == '(') {
-            spec->min = strtod((char*)min->ptr+1,&eptr);
+            spec->min = fast_float_strtod((char*)min->ptr+1,&eptr);
            if (eptr[0] != '\0' || isnan(spec->min)) return C_ERR;
            spec->minex = 1;
        } else {
-            spec->min = strtod((char*)min->ptr,&eptr);
+            spec->min = fast_float_strtod((char*)min->ptr,&eptr);
            if (eptr[0] != '\0' || isnan(spec->min)) return C_ERR;
        }
    }
@ -578,11 +578,11 @@ static int zslParseRange(robj *min, robj *max, zrangespec *spec) {
        spec->max = (long)max->ptr;
    } else {
        if (((char*)max->ptr)[0] == '(') {
-            spec->max = strtod((char*)max->ptr+1,&eptr);
+            spec->max = fast_float_strtod((char*)max->ptr+1,&eptr);
            if (eptr[0] != '\0' || isnan(spec->max)) return C_ERR;
            spec->maxex = 1;
        } else {
-            spec->max = strtod((char*)max->ptr,&eptr);
+            spec->max = fast_float_strtod((char*)max->ptr,&eptr);
            if (eptr[0] != '\0' || isnan(spec->max)) return C_ERR;
        }
    }
@ -789,7 +789,7 @@ double zzlStrtod(unsigned char *vstr, unsigned int vlen) {
        vlen = sizeof(buf) - 1;
    memcpy(buf,vstr,vlen);
    buf[vlen] = '\0';
-    return strtod(buf,NULL);
+    return fast_float_strtod(buf,NULL);
 }

 double zzlGetScore(unsigned char *sptr) {
--- a/src/util.c
+++ b/src/util.c
@ -30,6 +30,7 @@

 #include "fmacros.h"
 #include "fpconv_dtoa.h"
+#include "fast_float_strtod.h"
 #include <stdlib.h>
 #include <stdio.h>
 #include <string.h>
@ -612,7 +613,7 @@ int string2ld(const char *s, size_t slen, long double *dp) {
 int string2d(const char *s, size_t slen, double *dp) {
    errno = 0;
    char *eptr;
-    *dp = strtod(s, &eptr);
+    *dp = fast_float_strtod(s, &eptr);
    if (slen == 0 ||
        isspace(((const char*)s)[0]) ||
        (size_t)(eptr-(char*)s) != slen ||