Re-introduce puffs_base__limit1.

It's not immediately obvious why this data structure performs differently, but
on net, the benchmark numbers seem better. It also won't require any
PUFFS_USE_NO_OP_PERFORMANCE_HACKS or "redundant but mysteriously performant"
use_limit fields.

name                             old speed      new speed      delta

puffs_adler32_10k/clang          2.46GB/s ± 0%  2.43GB/s ± 3%     ~     (p=0.513 n=5+10)
puffs_adler32_100k/clang         2.46GB/s ± 0%  2.44GB/s ± 2%     ~     (p=0.241 n=5+10)
puffs_flate_decode_1k/clang       122MB/s ± 1%   119MB/s ± 2%   -1.86%  (p=0.013 n=5+10)
puffs_flate_decode_10k/clang      147MB/s ± 0%   142MB/s ± 2%   -3.47%  (p=0.001 n=5+10)
puffs_flate_decode_100k/clang     161MB/s ± 0%   161MB/s ± 4%     ~     (p=0.635 n=4+10)
puffs_zlib_decode_10k/clang       139MB/s ± 0%   135MB/s ± 1%   -2.69%  (p=0.001 n=5+10)
puffs_zlib_decode_100k/clang      151MB/s ± 0%   152MB/s ± 3%     ~     (p=1.000 n=5+10)

puffs_adler32_10k/gcc            3.21GB/s ± 1%  3.18GB/s ± 2%     ~     (p=0.310 n=5+10)
puffs_adler32_100k/gcc           3.22GB/s ± 0%  3.19GB/s ± 2%     ~     (p=0.075 n=5+10)
puffs_flate_decode_1k/gcc         121MB/s ± 4%   126MB/s ± 1%   +4.54%  (p=0.002 n=5+8)
puffs_flate_decode_10k/gcc        144MB/s ± 1%   141MB/s ± 1%   -2.52%  (p=0.001 n=5+9)
puffs_flate_decode_100k/gcc       154MB/s ± 1%   158MB/s ± 0%   +2.39%  (p=0.001 n=5+9)
puffs_zlib_decode_10k/gcc         137MB/s ± 2%   135MB/s ± 1%     ~     (p=0.055 n=5+10)
puffs_zlib_decode_100k/gcc        146MB/s ± 2%   150MB/s ± 0%   +2.87%  (p=0.001 n=5+10)

puffs_gif_lzw_decode_20k/clang    132MB/s ± 0%   125MB/s ± 0%   -5.24%  (p=0.001 n=5+10)
puffs_gif_lzw_decode_100k/clang   176MB/s ± 0%   164MB/s ± 0%   -6.82%  (p=0.001 n=5+9)
puffs_gif_decode_1k/clang         363MB/s ± 0%   339MB/s ± 1%   -6.58%  (p=0.002 n=4+10)
puffs_gif_decode_10k/clang        138MB/s ± 0%   133MB/s ± 1%   -3.77%  (p=0.001 n=5+10)
puffs_gif_decode_100k/clang       125MB/s ± 0%   119MB/s ± 0%   -4.79%  (p=0.001 n=5+10)
puffs_gif_decode_1000k/clang      127MB/s ± 0%   121MB/s ± 0%   -5.04%  (p=0.001 n=5+10)

puffs_gif_lzw_decode_20k/gcc      141MB/s ± 0%   140MB/s ± 2%     ~     (p=1.000 n=5+10)
puffs_gif_lzw_decode_100k/gcc     182MB/s ± 0%   180MB/s ± 2%   -1.56%  (p=0.001 n=5+10)
puffs_gif_decode_1k/gcc           363MB/s ± 0%   409MB/s ± 3%  +12.71%  (p=0.001 n=5+10)
puffs_gif_decode_10k/gcc          127MB/s ± 0%   161MB/s ± 3%  +26.15%  (p=0.001 n=5+10)
puffs_gif_decode_100k/gcc         112MB/s ± 0%   136MB/s ± 2%  +21.95%  (p=0.001 n=5+10)
puffs_gif_decode_1000k/gcc        114MB/s ± 0%   140MB/s ± 2%  +22.78%  (p=0.001 n=5+10)
This commit is contained in:
Nigel Tao committed 2017-11-03 10:09:57 +11:00
1 parent 4158d7eae7
commit e3822f9951
14 files changed
+444 -258

No files matched your search

+13 -10
View File
@@ -76,6 +76,17 @@ typedef struct {
bool closed; // No further writes are expected.
} puffs_base__buf1;
// puffs_base__limit1 provides a limited view of a 1-dimensional byte stream:
// its first N bytes. That N can be greater than a buffer's current read or
// write capacity. N decreases naturally over time as bytes are read from or
// written to the stream.
//
// A value with all fields NULL or zero is a valid, unlimited view.
typedef struct puffs_base__limit1 {
uint64_t* ptr_to_len; // Pointer to N.
struct puffs_base__limit1* next; // Linked list of limits.
} puffs_base__limit1;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
typedef struct {
void* always_null0;
@@ -91,12 +102,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
struct {
puffs_base__paired_nulls* noph0;
@@ -114,12 +121,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
} private_impl;
} puffs_base__writer1;
+10 -37
View File
@@ -278,31 +278,21 @@ static inline uint32_t puffs_base__writer1__copy_from_slice32(
// public (in base-header.h). We assume that, at the boundary between user code
// and Puffs code, the reader1 and writer1's private_impl fields (including
// limit and mark) are NULL. Otherwise, some internal assumptions break down.
// For example, the limit field is represented as a pointer, even though
// conceptually it is a count, but that pointer-to-count correspondence becomes
// invalid if a buffer is re-used (e.g. on resuming a coroutine).
// For example, limits could be represented as pointers, even though
// conceptually they are counts, but that pointer-to-count correspondence
// becomes invalid if a buffer is re-used (e.g. on resuming a coroutine).
//
// Admittedly, some of the Puffs test code calls these methods, but that test
// code is still Puffs code, not user code. Other Puffs test code modifies
// private_impl fields directly.
static inline puffs_base__empty_struct puffs_base__reader1__limit(
static inline puffs_base__reader1 puffs_base__reader1__limit(
puffs_base__reader1* o,
uint8_t* ptr,
uint64_t limit) {
if (!o->buf) {
o->private_impl.limit = NULL;
} else {
uint64_t len = o->buf->wi - o->buf->ri;
if (limit > len) {
limit = len;
}
if (!o->private_impl.limit || (limit < (o->private_impl.limit - ptr))) {
o->private_impl.limit = ptr + limit;
}
}
o->private_impl.use_limit = true;
return ((puffs_base__empty_struct){});
uint64_t* ptr_to_len) {
puffs_base__reader1 ret = *o;
ret.private_impl.limit.ptr_to_len = ptr_to_len;
ret.private_impl.limit.next = &o->private_impl.limit;
return ret;
}
static inline puffs_base__empty_struct puffs_base__reader1__mark(
@@ -312,24 +302,7 @@ static inline puffs_base__empty_struct puffs_base__reader1__mark(
return ((puffs_base__empty_struct){});
}
static inline puffs_base__empty_struct puffs_base__writer1__limit(
puffs_base__writer1* o,
uint8_t* ptr,
uint64_t limit) {
if (!o->buf) {
o->private_impl.limit = NULL;
} else {
uint64_t len = o->buf->len - o->buf->wi;
if (limit > len) {
limit = len;
}
if (!o->private_impl.limit || (limit < (o->private_impl.limit - ptr))) {
o->private_impl.limit = ptr + limit;
}
}
o->private_impl.use_limit = true;
return ((puffs_base__empty_struct){});
}
// TODO: static inline puffs_base__writer1 puffs_base__writer1__limit()
static inline puffs_base__empty_struct puffs_base__writer1__mark(
puffs_base__writer1* o,
+6 -7
View File
@@ -19,9 +19,9 @@ package cgen
const baseHeader = "" +
"#ifndef PUFFS_BASE_HEADER_H\n#define PUFFS_BASE_HEADER_H\n\n// Copyright 2017 The Puffs Authors.\n//\n// Licensed under the Apache License, Version 2.0 (the \"License\");\n// you may not use this file except in compliance with the License.\n// You may obtain a copy of the License at\n//\n// https://www.apache.org/licenses/LICENSE-2.0\n//\n// Unless required by applicable law or agreed to in writing, software\n// distributed under the License is distributed on an \"AS IS\" BASIS,\n// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n// See the License for the specific language governing permissions and\n// limitations under the License.\n\n#include <stdbool.h>\n#include <stdint.h>\n#include <string.h>\n\n// Puffs requires a word size of at least 32 bits because it assumes that\n// converting a u32 to usize will never overflow. For example, the size of a\n// decoded image is often represented, explicitly or implicitly in an image\n// file, as a u32, and it is convenient to compare that to a buffer size.\n//\n// Si" +
"milarly, the word size is at most 64 bits because it assumes that\n// converting a usize to u64 will never overflow.\n#if __WORDSIZE < 32\n#error \"Puffs requires a word size of at least 32 bits\"\n#elif __WORDSIZE > 64\n#error \"Puffs requires a word size of at most 64 bits\"\n#endif\n\n// PUFFS_VERSION is the major.minor version number as a uint32. The major\n// number is the high 16 bits. The minor number is the low 16 bits.\n//\n// The intention is to bump the version number at least on every API / ABI\n// backwards incompatible change.\n//\n// For now, the API and ABI are simply unstable and can change at any time.\n//\n// TODO: don't hard code this in base-header.h.\n#define PUFFS_VERSION (0x00001)\n\n// PUFFS_USE_NO_OP_PERFORMANCE_HACKS enables code paths that look like\n// redundant no-ops, but for reasons to be investigated, can have dramatic\n// performance effects with gcc 4.8.4 (e.g. 1.2x on some benchmarks).\n//\n// TODO: investigate; delete these hacks (without regressing performance).\n// The order matters here. Clang als" +
"o defines \"__GNUC__\".\n#if defined(__clang__)\n#elif defined(__GNUC__)\n#define PUFFS_USE_NO_OP_PERFORMANCE_HACKS 1\n#endif\n\n// puffs_base__slice_u8 is a 1-dimensional buffer (a pointer and length).\n//\n// A value with all fields NULL or zero is a valid, empty slice.\ntypedef struct {\n uint8_t* ptr;\n size_t len;\n} puffs_base__slice_u8;\n\n// puffs_base__buf1 is a 1-dimensional buffer (a pointer and length), plus\n// additional indexes into that buffer, plus an opened / closed flag.\n//\n// A value with all fields NULL or zero is a valid, empty buffer.\ntypedef struct {\n uint8_t* ptr; // Pointer.\n size_t len; // Length.\n size_t wi; // Write index. Invariant: wi <= len.\n size_t ri; // Read index. Invariant: ri <= wi.\n bool closed; // No further writes are expected.\n} puffs_base__buf1;\n\n#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS\ntypedef struct {\n void* always_null0;\n void* always_null1;\n} puffs_base__paired_nulls;\n#endif\n\ntypedef struct {\n // TODO: move buf into private_impl? As it is, it looks like u" +
"sers can modify\n // the buf field to point to a different buffer, which can turn the limit and\n // mark fields into dangling pointers.\n puffs_base__buf1* buf;\n // Do not access the private_impl's fields directly. There is no API/ABI\n // compatibility or safety guarantee if you do so.\n struct {\n uint8_t* limit;\n uint8_t* mark;\n // use_limit is redundant, in that it always equals (limit != NULL), but\n // having a separate bool can have a significant performance effect with\n // gcc 4.8.4 (e.g. 1.1x on some benchmarks).\n bool use_limit;\n#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS\n struct {\n puffs_base__paired_nulls* noph0;\n uint32_t noph1;\n } * no_op_performance_hacks;\n#endif\n } private_impl;\n} puffs_base__reader1;\n\ntypedef struct {\n // TODO: move buf into private_impl? As it is, it looks like users can modify\n // the buf field to point to a different buffer, which can turn the limit and\n // mark fields into dangling pointers.\n puffs_base__buf1* buf;\n // Do not access th" +
"e private_impl's fields directly. There is no API/ABI\n // compatibility or safety guarantee if you do so.\n struct {\n uint8_t* limit;\n uint8_t* mark;\n // use_limit is redundant, in that it always equals (limit != NULL), but\n // having a separate bool can have a significant performance effect with\n // gcc 4.8.4 (e.g. 1.1x on some benchmarks).\n bool use_limit;\n } private_impl;\n} puffs_base__writer1;\n\n#endif // PUFFS_BASE_HEADER_H\n" +
"o defines \"__GNUC__\".\n#if defined(__clang__)\n#elif defined(__GNUC__)\n#define PUFFS_USE_NO_OP_PERFORMANCE_HACKS 1\n#endif\n\n// puffs_base__slice_u8 is a 1-dimensional buffer (a pointer and length).\n//\n// A value with all fields NULL or zero is a valid, empty slice.\ntypedef struct {\n uint8_t* ptr;\n size_t len;\n} puffs_base__slice_u8;\n\n// puffs_base__buf1 is a 1-dimensional buffer (a pointer and length), plus\n// additional indexes into that buffer, plus an opened / closed flag.\n//\n// A value with all fields NULL or zero is a valid, empty buffer.\ntypedef struct {\n uint8_t* ptr; // Pointer.\n size_t len; // Length.\n size_t wi; // Write index. Invariant: wi <= len.\n size_t ri; // Read index. Invariant: ri <= wi.\n bool closed; // No further writes are expected.\n} puffs_base__buf1;\n\n// puffs_base__limit1 provides a limited view of a 1-dimensional byte stream:\n// its first N bytes. That N can be greater than a buffer's current read or\n// write capacity. N decreases naturally over time as bytes are r" +
"ead from or\n// written to the stream.\n//\n// A value with all fields NULL or zero is a valid, unlimited view.\ntypedef struct puffs_base__limit1 {\n uint64_t* ptr_to_len; // Pointer to N.\n struct puffs_base__limit1* next; // Linked list of limits.\n} puffs_base__limit1;\n\n#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS\ntypedef struct {\n void* always_null0;\n void* always_null1;\n} puffs_base__paired_nulls;\n#endif\n\ntypedef struct {\n // TODO: move buf into private_impl? As it is, it looks like users can modify\n // the buf field to point to a different buffer, which can turn the limit and\n // mark fields into dangling pointers.\n puffs_base__buf1* buf;\n // Do not access the private_impl's fields directly. There is no API/ABI\n // compatibility or safety guarantee if you do so.\n struct {\n puffs_base__limit1 limit;\n uint8_t* mark;\n#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS\n struct {\n puffs_base__paired_nulls* noph0;\n uint32_t noph1;\n } * no_op_performance_hacks;\n#endif\n } private_impl;" +
"\n} puffs_base__reader1;\n\ntypedef struct {\n // TODO: move buf into private_impl? As it is, it looks like users can modify\n // the buf field to point to a different buffer, which can turn the limit and\n // mark fields into dangling pointers.\n puffs_base__buf1* buf;\n // Do not access the private_impl's fields directly. There is no API/ABI\n // compatibility or safety guarantee if you do so.\n struct {\n puffs_base__limit1 limit;\n uint8_t* mark;\n } private_impl;\n} puffs_base__writer1;\n\n#endif // PUFFS_BASE_HEADER_H\n" +
""
const baseImpl = "" +
@@ -33,9 +33,8 @@ const baseImpl = "" +
"_ij(\n puffs_base__slice_u8 s,\n uint64_t i,\n uint64_t j) {\n if ((i <= j) && (j <= SIZE_MAX) && (j <= s.len)) {\n return ((puffs_base__slice_u8){\n .ptr = s.ptr + i,\n .len = j - i,\n });\n }\n return ((puffs_base__slice_u8){});\n}\n\n// puffs_base__slice_u8__prefix returns up to the first up_to bytes of s.\nstatic inline puffs_base__slice_u8 puffs_base__slice_u8__prefix(\n puffs_base__slice_u8 s,\n uint64_t up_to) {\n if ((uint64_t)(s.len) > up_to) {\n s.len = up_to;\n }\n return s;\n}\n\n// puffs_base__slice_u8__suffix returns up to the last up_to bytes of s.\nstatic inline puffs_base__slice_u8 puffs_base__slice_u8_suffix(\n puffs_base__slice_u8 s,\n uint64_t up_to) {\n if ((uint64_t)(s.len) > up_to) {\n s.ptr += (uint64_t)(s.len) - up_to;\n s.len = up_to;\n }\n return s;\n}\n\n// puffs_base__slice_u8__copy_from_slice calls memmove(dst.ptr, src.ptr,\n// length) where length is the minimum of dst.len and src.len.\n//\n// Passing a puffs_base__slice_u8 with all fields NULL or zero (a v" +
"alid, empty\n// slice) is valid and results in a no-op.\nstatic inline uint64_t puffs_base__slice_u8__copy_from_slice(\n puffs_base__slice_u8 dst,\n puffs_base__slice_u8 src) {\n size_t length = dst.len < src.len ? dst.len : src.len;\n if (length > 0) {\n memmove(dst.ptr, src.ptr, length);\n }\n return length;\n}\n\nstatic inline uint32_t puffs_base__writer1__copy_from_history32(\n uint8_t** ptr_ptr,\n uint8_t* start, // May be NULL, meaning an unmarked writer1.\n uint8_t* end,\n uint32_t distance,\n uint32_t length) {\n if (!start) {\n return 0;\n }\n uint8_t* ptr = *ptr_ptr;\n size_t d = ptr - start;\n if ((d == 0) || (d < (size_t)(distance))) {\n return 0;\n }\n start = ptr - distance;\n size_t n = end - ptr;\n if ((size_t)(length) > n) {\n length = n;\n } else {\n n = length;\n }\n // TODO: is manual unrolling actually helpful?\n for (; n >= 8; n -= 8) {\n *ptr++ = *start++;\n *ptr++ = *start++;\n *ptr++ = *start++;\n *ptr++ = *start++;\n *ptr++ = *start++;\n *ptr++ = *star" +
"t++;\n *ptr++ = *start++;\n *ptr++ = *start++;\n }\n for (; n; n--) {\n *ptr++ = *start++;\n }\n *ptr_ptr = ptr;\n return length;\n}\n\nstatic inline uint32_t puffs_base__writer1__copy_from_reader32(\n uint8_t** ptr_wptr,\n uint8_t* wend,\n uint8_t** ptr_rptr,\n uint8_t* rend,\n uint32_t length) {\n uint8_t* wptr = *ptr_wptr;\n size_t n = length;\n if (n > wend - wptr) {\n n = wend - wptr;\n }\n uint8_t* rptr = *ptr_rptr;\n if (n > rend - rptr) {\n n = rend - rptr;\n }\n if (n > 0) {\n memmove(wptr, rptr, n);\n *ptr_wptr += n;\n *ptr_rptr += n;\n }\n return n;\n}\n\nstatic inline uint64_t puffs_base__writer1__copy_from_slice(\n uint8_t** ptr_wptr,\n uint8_t* wend,\n puffs_base__slice_u8 src) {\n uint8_t* wptr = *ptr_wptr;\n size_t n = src.len;\n if (n > wend - wptr) {\n n = wend - wptr;\n }\n if (n > 0) {\n memmove(wptr, src.ptr, n);\n *ptr_wptr += n;\n }\n return n;\n}\n\nstatic inline uint32_t puffs_base__writer1__copy_from_slice32(\n uint8_t** ptr_wptr,\n uint8_t* wend,\n " +
" puffs_base__slice_u8 src,\n uint32_t length) {\n uint8_t* wptr = *ptr_wptr;\n size_t n = src.len;\n if (n > length) {\n n = length;\n }\n if (n > wend - wptr) {\n n = wend - wptr;\n }\n if (n > 0) {\n memmove(wptr, src.ptr, n);\n *ptr_wptr += n;\n }\n return n;\n}\n\n// Note that the *__limit and *__mark methods are private (in base-impl.h) not\n// public (in base-header.h). We assume that, at the boundary between user code\n// and Puffs code, the reader1 and writer1's private_impl fields (including\n// limit and mark) are NULL. Otherwise, some internal assumptions break down.\n// For example, the limit field is represented as a pointer, even though\n// conceptually it is a count, but that pointer-to-count correspondence becomes\n// invalid if a buffer is re-used (e.g. on resuming a coroutine).\n//\n// Admittedly, some of the Puffs test code calls these methods, but that test\n// code is still Puffs code, not user code. Other Puffs test code modifies\n// private_impl fields directly.\n\nstatic inline puffs_base__" +
"empty_struct puffs_base__reader1__limit(\n puffs_base__reader1* o,\n uint8_t* ptr,\n uint64_t limit) {\n if (!o->buf) {\n o->private_impl.limit = NULL;\n } else {\n uint64_t len = o->buf->wi - o->buf->ri;\n if (limit > len) {\n limit = len;\n }\n if (!o->private_impl.limit || (limit < (o->private_impl.limit - ptr))) {\n o->private_impl.limit = ptr + limit;\n }\n }\n o->private_impl.use_limit = true;\n return ((puffs_base__empty_struct){});\n}\n\nstatic inline puffs_base__empty_struct puffs_base__reader1__mark(\n puffs_base__reader1* o,\n uint8_t* mark) {\n o->private_impl.mark = mark;\n return ((puffs_base__empty_struct){});\n}\n\nstatic inline puffs_base__empty_struct puffs_base__writer1__limit(\n puffs_base__writer1* o,\n uint8_t* ptr,\n uint64_t limit) {\n if (!o->buf) {\n o->private_impl.limit = NULL;\n } else {\n uint64_t len = o->buf->len - o->buf->wi;\n if (limit > len) {\n limit = len;\n }\n if (!o->private_impl.limit || (limit < (o->private_impl.limit - ptr" +
"))) {\n o->private_impl.limit = ptr + limit;\n }\n }\n o->private_impl.use_limit = true;\n return ((puffs_base__empty_struct){});\n}\n\nstatic inline puffs_base__empty_struct puffs_base__writer1__mark(\n puffs_base__writer1* o,\n uint8_t* mark) {\n o->private_impl.mark = mark;\n return ((puffs_base__empty_struct){});\n}\n\n#endif // PUFFS_BASE_IMPL_H\n" +
" puffs_base__slice_u8 src,\n uint32_t length) {\n uint8_t* wptr = *ptr_wptr;\n size_t n = src.len;\n if (n > length) {\n n = length;\n }\n if (n > wend - wptr) {\n n = wend - wptr;\n }\n if (n > 0) {\n memmove(wptr, src.ptr, n);\n *ptr_wptr += n;\n }\n return n;\n}\n\n// Note that the *__limit and *__mark methods are private (in base-impl.h) not\n// public (in base-header.h). We assume that, at the boundary between user code\n// and Puffs code, the reader1 and writer1's private_impl fields (including\n// limit and mark) are NULL. Otherwise, some internal assumptions break down.\n// For example, limits could be represented as pointers, even though\n// conceptually they are counts, but that pointer-to-count correspondence\n// becomes invalid if a buffer is re-used (e.g. on resuming a coroutine).\n//\n// Admittedly, some of the Puffs test code calls these methods, but that test\n// code is still Puffs code, not user code. Other Puffs test code modifies\n// private_impl fields directly.\n\nstatic inline puffs_base__re" +
"ader1 puffs_base__reader1__limit(\n puffs_base__reader1* o,\n uint64_t* ptr_to_len) {\n puffs_base__reader1 ret = *o;\n ret.private_impl.limit.ptr_to_len = ptr_to_len;\n ret.private_impl.limit.next = &o->private_impl.limit;\n return ret;\n}\n\nstatic inline puffs_base__empty_struct puffs_base__reader1__mark(\n puffs_base__reader1* o,\n uint8_t* mark) {\n o->private_impl.mark = mark;\n return ((puffs_base__empty_struct){});\n}\n\n// TODO: static inline puffs_base__writer1 puffs_base__writer1__limit()\n\nstatic inline puffs_base__empty_struct puffs_base__writer1__mark(\n puffs_base__writer1* o,\n uint8_t* mark) {\n o->private_impl.mark = mark;\n return ((puffs_base__empty_struct){});\n}\n\n#endif // PUFFS_BASE_IMPL_H\n" +
""
type template_args_short_read struct {
@@ -44,7 +43,7 @@ type template_args_short_read struct {
}
func template_short_read(b *buffer, args template_args_short_read) error {
b.printf("short_read_%s:\nif (a_%s.buf && a_%s.buf->closed &&\n!a_%s.private_impl.use_limit) {\nstatus = %sERROR_UNEXPECTED_EOF;\ngoto exit;\n}\nstatus = %sSUSPENSION_SHORT_READ;\ngoto suspend;\n",
b.printf("short_read_%s:\nif (a_%s.buf && a_%s.buf->closed &&\n!a_%s.private_impl.limit.ptr_to_len) {\nstatus = %sERROR_UNEXPECTED_EOF;\ngoto exit;\n}\nstatus = %sSUSPENSION_SHORT_READ;\ngoto suspend;\n",
args.name,
args.name,
args.name,
+3 -9
View File
@@ -172,7 +172,7 @@ func (g *gen) writeExprOther(b *buffer, n *a.Expr, rp replacementPolicy, pp pare
return nil
}
if isInSrc(g.tm, n, t.KeyLimit, 1) {
return fmt.Errorf("TODO: in.src.limit")
return fmt.Errorf(`TODO: cgen an "in.src.limit" expression`)
}
if isInSrc(g.tm, n, t.KeyMark, 0) {
b.printf("puffs_base__reader1__mark(&%ssrc, %srptr_src)", aPrefix, bPrefix)
@@ -186,7 +186,7 @@ func (g *gen) writeExprOther(b *buffer, n *a.Expr, rp replacementPolicy, pp pare
return nil
}
if isInDst(g.tm, n, t.KeyLimit, 1) {
return fmt.Errorf("TODO: in.dst.limit")
return fmt.Errorf(`TODO: cgen an "in.dst.limit" expression`)
}
if isInDst(g.tm, n, t.KeyMark, 0) {
// TODO: is a private_impl.mark the right representation? What if
@@ -307,13 +307,7 @@ func (g *gen) writeExprOther(b *buffer, n *a.Expr, rp replacementPolicy, pp pare
return nil
}
if isThatMethod(g.tm, n, t.KeyLimit, 1) {
// TODO: don't hard-code v_r or b_rptr_src.
b.printf("puffs_base__reader1__limit(&v_r, b_rptr_src,")
if err := g.writeExpr(b, n.Args()[0].Arg().Value(), rp, parenthesesOptional, depth); err != nil {
return err
}
b.writes(")")
return nil
return fmt.Errorf(`TODO: cgen a "foo.limit" expression`)
}
if isThatMethod(g.tm, n, t.KeyMark, 0) {
// TODO: don't hard-code v_r or b_rptr_src.
+4 -2
View File
@@ -550,8 +550,10 @@ func (g *gen) writeCallSuspendibles(b *buffer, n *a.Expr, depth uint32) error {
temp := g.currFunk.tempW
g.currFunk.tempW++
// TODO: don't hard-code a_dst or v_r.
b.printf("%sstatus %s%d = %slzw_decoder__decode(&self->private_impl.f_lzw, %sdst, %sr);\n",
// TODO: don't hard-code a_dst or v_r or l_rlimit or v_block_size.
b.writes("uint64_t l_rlimit0 = v_block_size;\n")
b.printf("%sstatus %s%d = %slzw_decoder__decode(&self->private_impl.f_lzw, %sdst,"+
"puffs_base__reader1__limit(&%sr, &l_rlimit0));\n",
g.pkgPrefix, tPrefix, temp,
g.pkgPrefix, aPrefix, vPrefix)
if err := g.writeLoadExprDerivedVars(b, n); err != nil {
+1 -1
View File
@@ -24,7 +24,7 @@
template short_read(string qPKGPREFIXq, string qnameq) {
short_read_qnameq:
if (a_qnameq.buf && a_qnameq.buf->closed &&
!a_qnameq.private_impl.use_limit) {
!a_qnameq.private_impl.limit.ptr_to_len) {
status = qPKGPREFIXqERROR_UNEXPECTED_EOF;
goto exit;
}
+33 -6
View File
@@ -100,12 +100,18 @@ func (g *gen) writeLoadDerivedVar(b *buffer, name t.ID, typ *a.TypeExpr, header
b.printf("%srptr_%s = %s%s.buf->ptr + %s%s.buf->ri;",
bPrefix, nameStr, aPrefix, nameStr, aPrefix, nameStr)
if header {
b.printf("uint64_t len = %s%s.buf->wi - %s%s.buf->ri;", aPrefix, nameStr, aPrefix, nameStr)
b.printf("puffs_base__limit1* lim;\n")
b.printf("for (lim = &%s%s.private_impl.limit; lim; lim = lim->next) {\n", aPrefix, nameStr)
b.printf("if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {\n")
b.printf("len = *lim->ptr_to_len;\n")
b.printf("}\n")
b.printf("}\n")
b.printf("%srend_%s = %srptr_%s + len;", bPrefix, nameStr, bPrefix, nameStr)
b.printf("if (%s%s.private_impl.limit && (%srend_%s > %s%s.private_impl.limit)) {"+
" %srend_%s = %s%s.private_impl.limit; }",
aPrefix, nameStr, bPrefix, nameStr, aPrefix, nameStr, bPrefix, nameStr, aPrefix, nameStr)
}
b.printf("}\n")
@@ -119,14 +125,20 @@ func (g *gen) writeLoadDerivedVar(b *buffer, name t.ID, typ *a.TypeExpr, header
b.printf("%swptr_%s = %s%s.buf->ptr + %s%s.buf->wi;",
bPrefix, nameStr, aPrefix, nameStr, aPrefix, nameStr)
if header {
b.printf("%swend_%s = %swptr_%s;", bPrefix, nameStr, bPrefix, nameStr)
b.printf("if (!%s%s.buf->closed) {", aPrefix, nameStr)
b.printf("uint64_t len = %s%s.buf->len - %s%s.buf->wi;", aPrefix, nameStr, aPrefix, nameStr)
b.printf("puffs_base__limit1* lim;\n")
b.printf("for (lim = &%s%s.private_impl.limit; lim; lim = lim->next) {\n", aPrefix, nameStr)
b.printf("if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {\n")
b.printf("len = *lim->ptr_to_len;\n")
b.printf("}\n")
b.printf("}\n")
b.printf("%swend_%s += len;", bPrefix, nameStr)
b.printf("if (%s%s.private_impl.limit && (%swend_%s > %s%s.private_impl.limit)) {"+
" %swend_%s = %s%s.private_impl.limit; }",
aPrefix, nameStr, bPrefix, nameStr, aPrefix, nameStr, bPrefix, nameStr, aPrefix, nameStr)
b.printf("}\n")
}
@@ -156,6 +168,14 @@ func (g *gen) writeSaveDerivedVar(b *buffer, name t.ID, typ *a.TypeExpr, footer
b.printf("size_t n = %srptr_%s - (%s%s.buf->ptr + %s%s.buf->ri);",
bPrefix, nameStr, aPrefix, nameStr, aPrefix, nameStr)
b.printf("%s%s.buf->ri += n;", aPrefix, nameStr)
b.printf("puffs_base__limit1* lim;\n")
b.printf("for (lim = &%s%s.private_impl.limit; lim; lim = lim->next) {\n", aPrefix, nameStr)
b.printf("if (lim->ptr_to_len) {\n")
b.printf("*lim->ptr_to_len -= n;\n")
b.printf("}\n")
b.printf("}\n")
if footer {
b.printf("PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(%srend_%s);", bPrefix, nameStr)
}
@@ -169,6 +189,13 @@ func (g *gen) writeSaveDerivedVar(b *buffer, name t.ID, typ *a.TypeExpr, footer
bPrefix, nameStr, aPrefix, nameStr, aPrefix, nameStr)
b.printf("%s%s.buf->wi += n;", aPrefix, nameStr)
b.printf("puffs_base__limit1* lim;\n")
b.printf("for (lim = &%s%s.private_impl.limit; lim; lim = lim->next) {\n", aPrefix, nameStr)
b.printf("if (lim->ptr_to_len) {\n")
b.printf("*lim->ptr_to_len -= n;\n")
b.printf("}\n")
b.printf("}\n")
if footer {
b.printf("PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(%swend_%s);", bPrefix, nameStr)
}
+172 -79
View File
@@ -79,6 +79,17 @@ typedef struct {
bool closed; // No further writes are expected.
} puffs_base__buf1;
// puffs_base__limit1 provides a limited view of a 1-dimensional byte stream:
// its first N bytes. That N can be greater than a buffer's current read or
// write capacity. N decreases naturally over time as bytes are read from or
// written to the stream.
//
// A value with all fields NULL or zero is a valid, unlimited view.
typedef struct puffs_base__limit1 {
uint64_t* ptr_to_len; // Pointer to N.
struct puffs_base__limit1* next; // Linked list of limits.
} puffs_base__limit1;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
typedef struct {
void* always_null0;
@@ -94,12 +105,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
struct {
puffs_base__paired_nulls* noph0;
@@ -117,12 +124,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
} private_impl;
} puffs_base__writer1;
@@ -632,31 +635,21 @@ static inline uint32_t puffs_base__writer1__copy_from_slice32(
// public (in base-header.h). We assume that, at the boundary between user code
// and Puffs code, the reader1 and writer1's private_impl fields (including
// limit and mark) are NULL. Otherwise, some internal assumptions break down.
// For example, the limit field is represented as a pointer, even though
// conceptually it is a count, but that pointer-to-count correspondence becomes
// invalid if a buffer is re-used (e.g. on resuming a coroutine).
// For example, limits could be represented as pointers, even though
// conceptually they are counts, but that pointer-to-count correspondence
// becomes invalid if a buffer is re-used (e.g. on resuming a coroutine).
//
// Admittedly, some of the Puffs test code calls these methods, but that test
// code is still Puffs code, not user code. Other Puffs test code modifies
// private_impl fields directly.
static inline puffs_base__empty_struct puffs_base__reader1__limit(
static inline puffs_base__reader1 puffs_base__reader1__limit(
puffs_base__reader1* o,
uint8_t* ptr,
uint64_t limit) {
if (!o->buf) {
o->private_impl.limit = NULL;
} else {
uint64_t len = o->buf->wi - o->buf->ri;
if (limit > len) {
limit = len;
}
if (!o->private_impl.limit || (limit < (o->private_impl.limit - ptr))) {
o->private_impl.limit = ptr + limit;
}
}
o->private_impl.use_limit = true;
return ((puffs_base__empty_struct){});
uint64_t* ptr_to_len) {
puffs_base__reader1 ret = *o;
ret.private_impl.limit.ptr_to_len = ptr_to_len;
ret.private_impl.limit.next = &o->private_impl.limit;
return ret;
}
static inline puffs_base__empty_struct puffs_base__reader1__mark(
@@ -666,24 +659,7 @@ static inline puffs_base__empty_struct puffs_base__reader1__mark(
return ((puffs_base__empty_struct){});
}
static inline puffs_base__empty_struct puffs_base__writer1__limit(
puffs_base__writer1* o,
uint8_t* ptr,
uint64_t limit) {
if (!o->buf) {
o->private_impl.limit = NULL;
} else {
uint64_t len = o->buf->len - o->buf->wi;
if (limit > len) {
limit = len;
}
if (!o->private_impl.limit || (limit < (o->private_impl.limit - ptr))) {
o->private_impl.limit = ptr + limit;
}
}
o->private_impl.use_limit = true;
return ((puffs_base__empty_struct){});
}
// TODO: static inline puffs_base__writer1 puffs_base__writer1__limit()
static inline puffs_base__empty_struct puffs_base__writer1__mark(
puffs_base__writer1* o,
@@ -919,10 +895,13 @@ puffs_flate__status puffs_flate__flate_decoder__decode(
b_wend_dst = b_wptr_dst;
if (!a_dst.buf->closed) {
uint64_t len = a_dst.buf->len - a_dst.buf->wi;
b_wend_dst += len;
if (a_dst.private_impl.limit && (b_wend_dst > a_dst.private_impl.limit)) {
b_wend_dst = a_dst.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_wend_dst += len;
}
}
@@ -942,6 +921,12 @@ puffs_flate__status puffs_flate__flate_decoder__decode(
if (a_dst.buf) {
size_t n = b_wptr_dst - (a_dst.buf->ptr + a_dst.buf->wi);
a_dst.buf->wi += n;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
puffs_flate__status t_0 =
puffs_flate__flate_decoder__decode_blocks(self, a_dst, a_src);
@@ -1011,6 +996,12 @@ exit:
if (a_dst.buf) {
size_t n = b_wptr_dst - (a_dst.buf->ptr + a_dst.buf->wi);
a_dst.buf->wi += n;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_wend_dst);
}
@@ -1032,10 +1023,13 @@ puffs_flate__status puffs_flate__flate_decoder__decode_blocks(
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point =
@@ -1069,6 +1063,12 @@ puffs_flate__status puffs_flate__flate_decoder__decode_blocks(
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
status =
puffs_flate__flate_decoder__decode_uncompressed(self, a_dst, a_src);
@@ -1090,6 +1090,12 @@ puffs_flate__status puffs_flate__flate_decoder__decode_blocks(
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
status = puffs_flate__flate_decoder__init_dynamic_huffman(self, a_src);
if (a_src.buf) {
@@ -1106,6 +1112,12 @@ puffs_flate__status puffs_flate__flate_decoder__decode_blocks(
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
status = puffs_flate__flate_decoder__decode_huffman(self, a_dst, a_src);
if (a_src.buf) {
@@ -1132,13 +1144,19 @@ exit:
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_FLATE__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -1162,10 +1180,13 @@ puffs_flate__status puffs_flate__flate_decoder__decode_uncompressed(
b_wend_dst = b_wptr_dst;
if (!a_dst.buf->closed) {
uint64_t len = a_dst.buf->len - a_dst.buf->wi;
b_wend_dst += len;
if (a_dst.private_impl.limit && (b_wend_dst > a_dst.private_impl.limit)) {
b_wend_dst = a_dst.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_wend_dst += len;
}
}
uint8_t* b_rptr_src = NULL;
@@ -1173,10 +1194,13 @@ puffs_flate__status puffs_flate__flate_decoder__decode_uncompressed(
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point =
@@ -1261,18 +1285,30 @@ exit:
if (a_dst.buf) {
size_t n = b_wptr_dst - (a_dst.buf->ptr + a_dst.buf->wi);
a_dst.buf->wi += n;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_wend_dst);
}
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_FLATE__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -1307,10 +1343,13 @@ puffs_flate__status puffs_flate__flate_decoder__decode_huffman(
b_wend_dst = b_wptr_dst;
if (!a_dst.buf->closed) {
uint64_t len = a_dst.buf->len - a_dst.buf->wi;
b_wend_dst += len;
if (a_dst.private_impl.limit && (b_wend_dst > a_dst.private_impl.limit)) {
b_wend_dst = a_dst.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_wend_dst += len;
}
}
uint8_t* b_rptr_src = NULL;
@@ -1318,10 +1357,13 @@ puffs_flate__status puffs_flate__flate_decoder__decode_huffman(
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point =
@@ -1654,18 +1696,30 @@ exit:
if (a_dst.buf) {
size_t n = b_wptr_dst - (a_dst.buf->ptr + a_dst.buf->wi);
a_dst.buf->wi += n;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_wend_dst);
}
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_FLATE__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -1757,10 +1811,13 @@ puffs_flate__status puffs_flate__flate_decoder__init_dynamic_huffman(
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point =
@@ -1965,13 +2022,19 @@ exit:
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_FLATE__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -2285,10 +2348,13 @@ puffs_flate__status puffs_flate__zlib_decoder__decode(
b_wend_dst = b_wptr_dst;
if (!a_dst.buf->closed) {
uint64_t len = a_dst.buf->len - a_dst.buf->wi;
b_wend_dst += len;
if (a_dst.private_impl.limit && (b_wend_dst > a_dst.private_impl.limit)) {
b_wend_dst = a_dst.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_wend_dst += len;
}
}
uint8_t* b_rptr_src = NULL;
@@ -2296,10 +2362,13 @@ puffs_flate__status puffs_flate__zlib_decoder__decode(
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point = self->private_impl.c_decode[0].coro_susp_point;
@@ -2360,10 +2429,22 @@ puffs_flate__status puffs_flate__zlib_decoder__decode(
if (a_dst.buf) {
size_t n = b_wptr_dst - (a_dst.buf->ptr + a_dst.buf->wi);
a_dst.buf->wi += n;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
puffs_flate__status t_2 = puffs_flate__flate_decoder__decode(
&self->private_impl.f_flate, a_dst, a_src);
@@ -2436,11 +2517,23 @@ exit:
if (a_dst.buf) {
size_t n = b_wptr_dst - (a_dst.buf->ptr + a_dst.buf->wi);
a_dst.buf->wi += n;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_wend_dst);
}
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
@@ -2448,7 +2541,7 @@ exit:
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_FLATE__ERROR_UNEXPECTED_EOF;
goto exit;
}
+146 -76
View File
@@ -79,6 +79,17 @@ typedef struct {
bool closed; // No further writes are expected.
} puffs_base__buf1;
// puffs_base__limit1 provides a limited view of a 1-dimensional byte stream:
// its first N bytes. That N can be greater than a buffer's current read or
// write capacity. N decreases naturally over time as bytes are read from or
// written to the stream.
//
// A value with all fields NULL or zero is a valid, unlimited view.
typedef struct puffs_base__limit1 {
uint64_t* ptr_to_len; // Pointer to N.
struct puffs_base__limit1* next; // Linked list of limits.
} puffs_base__limit1;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
typedef struct {
void* always_null0;
@@ -94,12 +105,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
struct {
puffs_base__paired_nulls* noph0;
@@ -117,12 +124,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
} private_impl;
} puffs_base__writer1;
@@ -579,31 +582,21 @@ static inline uint32_t puffs_base__writer1__copy_from_slice32(
// public (in base-header.h). We assume that, at the boundary between user code
// and Puffs code, the reader1 and writer1's private_impl fields (including
// limit and mark) are NULL. Otherwise, some internal assumptions break down.
// For example, the limit field is represented as a pointer, even though
// conceptually it is a count, but that pointer-to-count correspondence becomes
// invalid if a buffer is re-used (e.g. on resuming a coroutine).
// For example, limits could be represented as pointers, even though
// conceptually they are counts, but that pointer-to-count correspondence
// becomes invalid if a buffer is re-used (e.g. on resuming a coroutine).
//
// Admittedly, some of the Puffs test code calls these methods, but that test
// code is still Puffs code, not user code. Other Puffs test code modifies
// private_impl fields directly.
static inline puffs_base__empty_struct puffs_base__reader1__limit(
static inline puffs_base__reader1 puffs_base__reader1__limit(
puffs_base__reader1* o,
uint8_t* ptr,
uint64_t limit) {
if (!o->buf) {
o->private_impl.limit = NULL;
} else {
uint64_t len = o->buf->wi - o->buf->ri;
if (limit > len) {
limit = len;
}
if (!o->private_impl.limit || (limit < (o->private_impl.limit - ptr))) {
o->private_impl.limit = ptr + limit;
}
}
o->private_impl.use_limit = true;
return ((puffs_base__empty_struct){});
uint64_t* ptr_to_len) {
puffs_base__reader1 ret = *o;
ret.private_impl.limit.ptr_to_len = ptr_to_len;
ret.private_impl.limit.next = &o->private_impl.limit;
return ret;
}
static inline puffs_base__empty_struct puffs_base__reader1__mark(
@@ -613,24 +606,7 @@ static inline puffs_base__empty_struct puffs_base__reader1__mark(
return ((puffs_base__empty_struct){});
}
static inline puffs_base__empty_struct puffs_base__writer1__limit(
puffs_base__writer1* o,
uint8_t* ptr,
uint64_t limit) {
if (!o->buf) {
o->private_impl.limit = NULL;
} else {
uint64_t len = o->buf->len - o->buf->wi;
if (limit > len) {
limit = len;
}
if (!o->private_impl.limit || (limit < (o->private_impl.limit - ptr))) {
o->private_impl.limit = ptr + limit;
}
}
o->private_impl.use_limit = true;
return ((puffs_base__empty_struct){});
}
// TODO: static inline puffs_base__writer1 puffs_base__writer1__limit()
static inline puffs_base__empty_struct puffs_base__writer1__mark(
puffs_base__writer1* o,
@@ -767,10 +743,13 @@ puffs_gif__status puffs_gif__decoder__decode(puffs_gif__decoder* self,
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point = self->private_impl.c_decode[0].coro_susp_point;
@@ -784,6 +763,12 @@ puffs_gif__status puffs_gif__decoder__decode(puffs_gif__decoder* self,
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
status = puffs_gif__decoder__decode_header(self, a_src);
if (a_src.buf) {
@@ -796,6 +781,12 @@ puffs_gif__status puffs_gif__decoder__decode(puffs_gif__decoder* self,
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
status = puffs_gif__decoder__decode_lsd(self, a_src);
if (a_src.buf) {
@@ -816,6 +807,12 @@ puffs_gif__status puffs_gif__decoder__decode(puffs_gif__decoder* self,
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
status = puffs_gif__decoder__decode_extension(self, a_src);
if (a_src.buf) {
@@ -829,6 +826,12 @@ puffs_gif__status puffs_gif__decoder__decode(puffs_gif__decoder* self,
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
status = puffs_gif__decoder__decode_id(self, a_dst, a_src);
if (a_src.buf) {
@@ -861,6 +864,12 @@ exit:
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
@@ -868,7 +877,7 @@ exit:
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_GIF__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -888,10 +897,13 @@ puffs_gif__status puffs_gif__decoder__decode_header(puffs_gif__decoder* self,
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point =
@@ -936,13 +948,19 @@ exit:
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_GIF__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -963,10 +981,13 @@ puffs_gif__status puffs_gif__decoder__decode_lsd(puffs_gif__decoder* self,
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point = self->private_impl.c_decode_lsd[0].coro_susp_point;
@@ -1037,13 +1058,19 @@ exit:
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_GIF__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -1064,10 +1091,13 @@ puffs_gif__status puffs_gif__decoder__decode_extension(
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point =
@@ -1134,13 +1164,19 @@ exit:
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_GIF__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -1166,10 +1202,13 @@ puffs_gif__status puffs_gif__decoder__decode_id(puffs_gif__decoder* self,
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point = self->private_impl.c_decode_id[0].coro_susp_point;
@@ -1228,14 +1267,21 @@ puffs_gif__status puffs_gif__decoder__decode_id(puffs_gif__decoder* self,
while (true) {
v_r = a_src;
puffs_base__reader1__mark(&v_r, b_rptr_src);
puffs_base__reader1__limit(&v_r, b_rptr_src, v_block_size);
PUFFS_BASE__COROUTINE_SUSPENSION_POINT(4);
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
}
uint64_t l_rlimit0 = v_block_size;
puffs_gif__status t_3 = puffs_gif__lzw_decoder__decode(
&self->private_impl.f_lzw, a_dst, v_r);
&self->private_impl.f_lzw, a_dst,
puffs_base__reader1__limit(&v_r, &l_rlimit0));
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
}
@@ -1292,13 +1338,19 @@ exit:
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_GIF__ERROR_UNEXPECTED_EOF;
goto exit;
}
@@ -1359,10 +1411,13 @@ puffs_gif__status puffs_gif__lzw_decoder__decode(puffs_gif__lzw_decoder* self,
b_wend_dst = b_wptr_dst;
if (!a_dst.buf->closed) {
uint64_t len = a_dst.buf->len - a_dst.buf->wi;
b_wend_dst += len;
if (a_dst.private_impl.limit && (b_wend_dst > a_dst.private_impl.limit)) {
b_wend_dst = a_dst.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_wend_dst += len;
}
}
uint8_t* b_rptr_src = NULL;
@@ -1370,10 +1425,13 @@ puffs_gif__status puffs_gif__lzw_decoder__decode(puffs_gif__lzw_decoder* self,
if (a_src.buf) {
b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
uint64_t len = a_src.buf->wi - a_src.buf->ri;
b_rend_src = b_rptr_src + len;
if (a_src.private_impl.limit && (b_rend_src > a_src.private_impl.limit)) {
b_rend_src = a_src.private_impl.limit;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len && (len > *lim->ptr_to_len)) {
len = *lim->ptr_to_len;
}
}
b_rend_src = b_rptr_src + len;
}
uint32_t coro_susp_point = self->private_impl.c_decode[0].coro_susp_point;
@@ -1513,11 +1571,23 @@ exit:
if (a_dst.buf) {
size_t n = b_wptr_dst - (a_dst.buf->ptr + a_dst.buf->wi);
a_dst.buf->wi += n;
puffs_base__limit1* lim;
for (lim = &a_dst.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_wend_dst);
}
if (a_src.buf) {
size_t n = b_rptr_src - (a_src.buf->ptr + a_src.buf->ri);
a_src.buf->ri += n;
puffs_base__limit1* lim;
for (lim = &a_src.private_impl.limit; lim; lim = lim->next) {
if (lim->ptr_to_len) {
*lim->ptr_to_len -= n;
}
}
PUFFS_BASE__IGNORE_POTENTIALLY_UNUSED_VARIABLE(b_rend_src);
}
@@ -1525,7 +1595,7 @@ exit:
return status;
short_read_src:
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.use_limit) {
if (a_src.buf && a_src.buf->closed && !a_src.private_impl.limit.ptr_to_len) {
status = PUFFS_GIF__ERROR_UNEXPECTED_EOF;
goto exit;
}
+13 -10
View File
@@ -79,6 +79,17 @@ typedef struct {
bool closed; // No further writes are expected.
} puffs_base__buf1;
// puffs_base__limit1 provides a limited view of a 1-dimensional byte stream:
// its first N bytes. That N can be greater than a buffer's current read or
// write capacity. N decreases naturally over time as bytes are read from or
// written to the stream.
//
// A value with all fields NULL or zero is a valid, unlimited view.
typedef struct puffs_base__limit1 {
uint64_t* ptr_to_len; // Pointer to N.
struct puffs_base__limit1* next; // Linked list of limits.
} puffs_base__limit1;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
typedef struct {
void* always_null0;
@@ -94,12 +105,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
struct {
puffs_base__paired_nulls* noph0;
@@ -117,12 +124,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
} private_impl;
} puffs_base__writer1;
+13 -10
View File
@@ -79,6 +79,17 @@ typedef struct {
bool closed; // No further writes are expected.
} puffs_base__buf1;
// puffs_base__limit1 provides a limited view of a 1-dimensional byte stream:
// its first N bytes. That N can be greater than a buffer's current read or
// write capacity. N decreases naturally over time as bytes are read from or
// written to the stream.
//
// A value with all fields NULL or zero is a valid, unlimited view.
typedef struct puffs_base__limit1 {
uint64_t* ptr_to_len; // Pointer to N.
struct puffs_base__limit1* next; // Linked list of limits.
} puffs_base__limit1;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
typedef struct {
void* always_null0;
@@ -94,12 +105,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
#ifdef PUFFS_USE_NO_OP_PERFORMANCE_HACKS
struct {
puffs_base__paired_nulls* noph0;
@@ -117,12 +124,8 @@ typedef struct {
// Do not access the private_impl's fields directly. There is no API/ABI
// compatibility or safety guarantee if you do so.
struct {
uint8_t* limit;
puffs_base__limit1 limit;
uint8_t* mark;
// use_limit is redundant, in that it always equals (limit != NULL), but
// having a separate bool can have a significant performance effect with
// gcc 4.8.4 (e.g. 1.1x on some benchmarks).
bool use_limit;
} private_impl;
} puffs_base__writer1;
+5 -2
View File
@@ -163,12 +163,15 @@ pri func decoder.decode_id?(dst writer1, src reader1)() {
}
while true {
var r reader1 = in.src
// TODO: should "mark" be "set_mark"? Unlike "limit", "mark" does
// not return a different reader1.
r.mark()
r.limit(l:block_size)
// TODO: remove the dummy param. It's needed for now so that
// writeSaveExprDerivedVars updates e.g. the b_rptr_src derived
// variables.
var z status = try this.lzw.decode?(dst:in.dst, src:r, dummy:in.src)
//
// TODO: enforce that limit can only be called in a "foo?" call?
var z status = try this.lzw.decode?(dst:in.dst, src:r.limit(l:block_size), dummy:in.src)
if z.is_ok() {
break
}
+13 -5
View File
@@ -189,14 +189,18 @@ const char* puffs_flate_decode(puffs_base__buf1* dst,
uint64_t rlimit) {
puffs_flate__flate_decoder dec;
puffs_flate__flate_decoder__initialize(&dec, PUFFS_VERSION, 0);
uint64_t wlim = 0;
uint64_t rlim = 0;
while (true) {
puffs_base__writer1 dst_writer = {.buf = dst};
if (wlimit) {
puffs_base__writer1__limit(&dst_writer, dst->ptr + dst->wi, wlimit);
wlim = wlimit;
dst_writer.private_impl.limit.ptr_to_len = &wlim;
}
puffs_base__reader1 src_reader = {.buf = src};
if (rlimit) {
puffs_base__reader1__limit(&src_reader, src->ptr + src->ri, rlimit);
rlim = rlimit;
src_reader.private_impl.limit.ptr_to_len = &rlim;
}
puffs_flate__status s =
@@ -220,14 +224,18 @@ const char* puffs_zlib_decode(puffs_base__buf1* dst,
puffs_flate__zlib_decoder dec;
puffs_flate__zlib_decoder__initialize(&dec, PUFFS_VERSION, 0);
uint64_t wlim = 0;
uint64_t rlim = 0;
while (true) {
puffs_base__writer1 dst_writer = {.buf = dst};
if (wlimit) {
puffs_base__writer1__limit(&dst_writer, dst->ptr + dst->wi, wlimit);
wlim = wlimit;
dst_writer.private_impl.limit.ptr_to_len = &wlim;
}
puffs_base__reader1 src_reader = {.buf = src};
if (rlimit) {
puffs_base__reader1__limit(&src_reader, src->ptr + src->ri, rlimit);
rlim = rlimit;
src_reader.private_impl.limit.ptr_to_len = &rlim;
}
puffs_flate__status s =
@@ -373,7 +381,7 @@ bool do_test_puffs_flate_history(int i,
dec->private_impl.f_history_index = starting_history_index;
puffs_base__writer1__limit(&dst_writer, got->ptr + got->wi, limit);
dst_writer.private_impl.limit.ptr_to_len = &limit;
puffs_flate__status got_s =
puffs_flate__flate_decoder__decode(dec, dst_writer, src_reader);
+12 -4
View File
@@ -230,15 +230,19 @@ bool do_test_puffs_gif_lzw_decode(const char* src_filename,
puffs_gif__lzw_decoder__initialize(&dec, PUFFS_VERSION, 0);
puffs_gif__lzw_decoder__set_literal_width(&dec, literal_width);
int num_iters = 0;
uint64_t wlim = 0;
uint64_t rlim = 0;
while (true) {
num_iters++;
puffs_base__writer1 got_writer = {.buf = &got};
if (wlimit) {
puffs_base__writer1__limit(&got_writer, got.ptr + got.wi, wlimit);
wlim = wlimit;
got_writer.private_impl.limit.ptr_to_len = &wlim;
}
puffs_base__reader1 src_reader = {.buf = &src};
if (rlimit) {
puffs_base__reader1__limit(&src_reader, src.ptr + src.ri, rlimit);
rlim = rlimit;
src_reader.private_impl.limit.ptr_to_len = &rlim;
}
size_t old_wi = got.wi;
size_t old_ri = src.ri;
@@ -409,15 +413,19 @@ bool do_test_puffs_gif_decode(const char* filename,
puffs_gif__decoder dec;
puffs_gif__decoder__initialize(&dec, PUFFS_VERSION, 0);
int num_iters = 0;
uint64_t wlim = 0;
uint64_t rlim = 0;
while (true) {
num_iters++;
puffs_base__writer1 got_writer = {.buf = &got};
if (wlimit) {
puffs_base__writer1__limit(&got_writer, got.ptr + got.wi, wlimit);
wlim = wlimit;
got_writer.private_impl.limit.ptr_to_len = &wlim;
}
puffs_base__reader1 src_reader = {.buf = &src};
if (rlimit) {
puffs_base__reader1__limit(&src_reader, src.ptr + src.ri, rlimit);
rlim = rlimit;
src_reader.private_impl.limit.ptr_to_len = &rlim;
}
size_t old_wi = got.wi;
size_t old_ri = src.ri;