Drop obsolete etc_8_byte_chunks_fast comment

This commit is contained in:
Nigel Tao committed 2021-03-25 22:43:54 +11:00
1 parent ad40102c5a
commit 3f262b2121
3 files changed
+5 -12

No files matched your search

-3
View File
@@ -187,9 +187,6 @@ wuffs_base__io_writer__limited_copy_u32_from_history(uint8_t** ptr_iop_w,
// expression '3'.
//
// See also wuffs_base__io_writer__limited_copy_u32_from_history_fast below.
//
// Alternatively or additionally, have a sloppy_limited_copy_u32_from_history
// method that copies 8 bytes at a time, which can more than length bytes?
for (; n >= 3; n -= 3) {
*p++ = *q++;
*p++ = *q++;
+5 -6
View File
@@ -213,12 +213,11 @@ const BaseIOPrivateH = "" +
"r != ((uint8_t)(a))) {\n return 2;\n }\n iop_r++;\n a >>= 8;\n }\n return 0;\n}\n\nstatic inline wuffs_base__io_buffer* //\nwuffs_base__io_reader__set(wuffs_base__io_buffer* b,\n const uint8_t** ptr_iop_r,\n const uint8_t** ptr_io0_r,\n const uint8_t** ptr_io1_r,\n const uint8_t** ptr_io2_r,\n wuffs_base__slice_u8 data) {\n b->data = data;\n b->meta.wi = data.len;\n b->meta.ri = 0;\n b->meta.pos = 0;\n b->meta.closed = false;\n\n *ptr_iop_r = data.ptr;\n *ptr_io0_r = data.ptr;\n *ptr_io1_r = data.ptr;\n *ptr_io2_r = data.ptr + data.len;\n\n return b;\n}\n\n" +
"" +
"// --------\n\nstatic inline uint64_t //\nwuffs_base__io_writer__copy_from_slice(uint8_t** ptr_iop_w,\n uint8_t* io2_w,\n wuffs_base__slice_u8 src) {\n uint8_t* iop_w = *ptr_iop_w;\n size_t n = src.len;\n if (n > ((size_t)(io2_w - iop_w))) {\n n = (size_t)(io2_w - iop_w);\n }\n if (n > 0) {\n memmove(iop_w, src.ptr, n);\n *ptr_iop_w += n;\n }\n return (uint64_t)(n);\n}\n\nstatic inline void //\nwuffs_base__io_writer__limit(uint8_t** ptr_io2_w,\n uint8_t* iop_w,\n uint64_t limit) {\n if (((uint64_t)(*ptr_io2_w - iop_w)) > limit) {\n *ptr_io2_w = iop_w + limit;\n }\n}\n\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_history(uint8_t** ptr_iop_w,\n uint8_t* io1_w,\n uint8_t* io2_w,\n uint32_t length,\n " +
" uint32_t distance) {\n if (!distance) {\n return 0;\n }\n uint8_t* p = *ptr_iop_w;\n if ((size_t)(p - io1_w) < (size_t)(distance)) {\n return 0;\n }\n uint8_t* q = p - distance;\n size_t n = (size_t)(io2_w - p);\n if ((size_t)(length) > n) {\n length = (uint32_t)(n);\n } else {\n n = (size_t)(length);\n }\n // TODO: unrolling by 3 seems best for the std/deflate benchmarks, but that\n // is mostly because 3 is the minimum length for the deflate format. This\n // function implementation shouldn't overfit to that one format. Perhaps the\n // limited_copy_u32_from_history Wuffs method should also take an unroll hint\n // argument, and the cgen can look if that argument is the constant\n // expression '3'.\n //\n // See also wuffs_base__io_writer__limited_copy_u32_from_history_fast below.\n //\n // Alternatively or additionally, have a sloppy_limited_copy_u32_from_history\n // method that copies 8 bytes at a time, which can more than length bytes?\n for (; n >= 3; " +
"n -= 3) {\n *p++ = *q++;\n *p++ = *q++;\n *p++ = *q++;\n }\n for (; n; n--) {\n *p++ = *q++;\n }\n *ptr_iop_w = p;\n return length;\n}\n\n// wuffs_base__io_writer__limited_copy_u32_from_history_fast is like the\n// wuffs_base__io_writer__limited_copy_u32_from_history function above, but has\n// stronger pre-conditions.\n//\n// The caller needs to prove that:\n// - length <= (io2_w - *ptr_iop_w)\n// - distance >= 1\n// - distance <= (*ptr_iop_w - io1_w)\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_history_fast(uint8_t** ptr_iop_w,\n uint8_t* io1_w,\n uint8_t* io2_w,\n uint32_t length,\n uint32_t distance) {\n uint8_t* p = *ptr_iop_w;\n uint8_t* q = p - distance;\n uint32_t n = length;\n for (; n >= 3; n -= 3) {\n *p++ = *q++;\n *p++ = *q++;\n *p++ = " +
"*q++;\n }\n for (; n; n--) {\n *p++ = *q++;\n }\n *ptr_iop_w = p;\n return length;\n}\n\n// wuffs_base__io_writer__limited_copy_u32_from_history_8_byte_chunks_fast is\n// like the wuffs_base__io_writer__limited_copy_u32_from_history_fast function\n// above, but copies 8 byte chunks at a time.\n//\n// In terms of number of bytes copied, length is rounded up to a multiple of 8.\n// As a special case, a zero length rounds up to 8 (even though 0 is already a\n// multiple of 8), since there is always at least one 8 byte chunk copied.\n//\n// In terms of advancing *ptr_iop_w, length is not rounded up.\n//\n// The caller needs to prove that:\n// - (length + 8) <= (io2_w - *ptr_iop_w)\n// - distance >= 8\n// - distance <= (*ptr_iop_w - io1_w)\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_history_8_byte_chunks_fast(\n uint8_t** ptr_iop_w,\n uint8_t* io1_w,\n uint8_t* io2_w,\n uint32_t length,\n uint32_t distance) {\n uint8_t* p = *ptr_iop_w;\n uint8_t* q = p - distance;\n uint32_t" +
" n = length;\n while (1) {\n memcpy(p, q, 8);\n if (n <= 8) {\n p += n;\n break;\n }\n p += 8;\n q += 8;\n n -= 8;\n }\n *ptr_iop_w = p;\n return length;\n}\n\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_reader(uint8_t** ptr_iop_w,\n uint8_t* io2_w,\n uint32_t length,\n const uint8_t** ptr_iop_r,\n const uint8_t* io2_r) {\n uint8_t* iop_w = *ptr_iop_w;\n size_t n = length;\n if (n > ((size_t)(io2_w - iop_w))) {\n n = (size_t)(io2_w - iop_w);\n }\n const uint8_t* iop_r = *ptr_iop_r;\n if (n > ((size_t)(io2_r - iop_r))) {\n n = (size_t)(io2_r - iop_r);\n }\n if (n > 0) {\n memmove(iop_w, iop_r, n);\n *ptr_iop_w += n;\n *ptr_iop_r += n;\n }\n return (uint32_t)(n);\n}\n\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_slice(uint8_t** ptr_iop_" +
"w,\n uint8_t* io2_w,\n uint32_t length,\n wuffs_base__slice_u8 src) {\n uint8_t* iop_w = *ptr_iop_w;\n size_t n = src.len;\n if (n > length) {\n n = length;\n }\n if (n > ((size_t)(io2_w - iop_w))) {\n n = (size_t)(io2_w - iop_w);\n }\n if (n > 0) {\n memmove(iop_w, src.ptr, n);\n *ptr_iop_w += n;\n }\n return (uint32_t)(n);\n}\n\nstatic inline wuffs_base__io_buffer* //\nwuffs_base__io_writer__set(wuffs_base__io_buffer* b,\n uint8_t** ptr_iop_w,\n uint8_t** ptr_io0_w,\n uint8_t** ptr_io1_w,\n uint8_t** ptr_io2_w,\n wuffs_base__slice_u8 data) {\n b->data = data;\n b->meta.wi = 0;\n b->meta.ri = 0;\n b->meta.pos = 0;\n b->meta.closed = false;\n\n *ptr_iop_w = data.ptr;\n *ptr_io0_w = data.ptr;\n *ptr_io1_w = data.ptr;\n *ptr_io2_w = data.pt" +
"r + data.len;\n\n return b;\n}\n\n" +
" uint32_t distance) {\n if (!distance) {\n return 0;\n }\n uint8_t* p = *ptr_iop_w;\n if ((size_t)(p - io1_w) < (size_t)(distance)) {\n return 0;\n }\n uint8_t* q = p - distance;\n size_t n = (size_t)(io2_w - p);\n if ((size_t)(length) > n) {\n length = (uint32_t)(n);\n } else {\n n = (size_t)(length);\n }\n // TODO: unrolling by 3 seems best for the std/deflate benchmarks, but that\n // is mostly because 3 is the minimum length for the deflate format. This\n // function implementation shouldn't overfit to that one format. Perhaps the\n // limited_copy_u32_from_history Wuffs method should also take an unroll hint\n // argument, and the cgen can look if that argument is the constant\n // expression '3'.\n //\n // See also wuffs_base__io_writer__limited_copy_u32_from_history_fast below.\n for (; n >= 3; n -= 3) {\n *p++ = *q++;\n *p++ = *q++;\n *p++ = *q++;\n }\n for (; n; n--) {\n *p++ = *q++;\n }\n *ptr_iop_w = p;\n return length;\n}\n\n// wuffs_base__io_w" +
"riter__limited_copy_u32_from_history_fast is like the\n// wuffs_base__io_writer__limited_copy_u32_from_history function above, but has\n// stronger pre-conditions.\n//\n// The caller needs to prove that:\n// - length <= (io2_w - *ptr_iop_w)\n// - distance >= 1\n// - distance <= (*ptr_iop_w - io1_w)\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_history_fast(uint8_t** ptr_iop_w,\n uint8_t* io1_w,\n uint8_t* io2_w,\n uint32_t length,\n uint32_t distance) {\n uint8_t* p = *ptr_iop_w;\n uint8_t* q = p - distance;\n uint32_t n = length;\n for (; n >= 3; n -= 3) {\n *p++ = *q++;\n *p++ = *q++;\n *p++ = *q++;\n }\n for (; n; n--) {\n *p++ = *q++;\n }\n *ptr_iop_w = p;\n return length;\n}\n\n// wuffs_base__io_writer__limited_copy_u32_from_history_8_byte_chunks_fast" +
" is\n// like the wuffs_base__io_writer__limited_copy_u32_from_history_fast function\n// above, but copies 8 byte chunks at a time.\n//\n// In terms of number of bytes copied, length is rounded up to a multiple of 8.\n// As a special case, a zero length rounds up to 8 (even though 0 is already a\n// multiple of 8), since there is always at least one 8 byte chunk copied.\n//\n// In terms of advancing *ptr_iop_w, length is not rounded up.\n//\n// The caller needs to prove that:\n// - (length + 8) <= (io2_w - *ptr_iop_w)\n// - distance >= 8\n// - distance <= (*ptr_iop_w - io1_w)\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_history_8_byte_chunks_fast(\n uint8_t** ptr_iop_w,\n uint8_t* io1_w,\n uint8_t* io2_w,\n uint32_t length,\n uint32_t distance) {\n uint8_t* p = *ptr_iop_w;\n uint8_t* q = p - distance;\n uint32_t n = length;\n while (1) {\n memcpy(p, q, 8);\n if (n <= 8) {\n p += n;\n break;\n }\n p += 8;\n q += 8;\n n -= 8;\n }\n *ptr_iop_w = p;\n ret" +
"urn length;\n}\n\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_reader(uint8_t** ptr_iop_w,\n uint8_t* io2_w,\n uint32_t length,\n const uint8_t** ptr_iop_r,\n const uint8_t* io2_r) {\n uint8_t* iop_w = *ptr_iop_w;\n size_t n = length;\n if (n > ((size_t)(io2_w - iop_w))) {\n n = (size_t)(io2_w - iop_w);\n }\n const uint8_t* iop_r = *ptr_iop_r;\n if (n > ((size_t)(io2_r - iop_r))) {\n n = (size_t)(io2_r - iop_r);\n }\n if (n > 0) {\n memmove(iop_w, iop_r, n);\n *ptr_iop_w += n;\n *ptr_iop_r += n;\n }\n return (uint32_t)(n);\n}\n\nstatic inline uint32_t //\nwuffs_base__io_writer__limited_copy_u32_from_slice(uint8_t** ptr_iop_w,\n uint8_t* io2_w,\n uint32_t length,\n " +
" wuffs_base__slice_u8 src) {\n uint8_t* iop_w = *ptr_iop_w;\n size_t n = src.len;\n if (n > length) {\n n = length;\n }\n if (n > ((size_t)(io2_w - iop_w))) {\n n = (size_t)(io2_w - iop_w);\n }\n if (n > 0) {\n memmove(iop_w, src.ptr, n);\n *ptr_iop_w += n;\n }\n return (uint32_t)(n);\n}\n\nstatic inline wuffs_base__io_buffer* //\nwuffs_base__io_writer__set(wuffs_base__io_buffer* b,\n uint8_t** ptr_iop_w,\n uint8_t** ptr_io0_w,\n uint8_t** ptr_io1_w,\n uint8_t** ptr_io2_w,\n wuffs_base__slice_u8 data) {\n b->data = data;\n b->meta.wi = 0;\n b->meta.ri = 0;\n b->meta.pos = 0;\n b->meta.closed = false;\n\n *ptr_iop_w = data.ptr;\n *ptr_io0_w = data.ptr;\n *ptr_io1_w = data.ptr;\n *ptr_io2_w = data.ptr + data.len;\n\n return b;\n}\n\n" +
"" +
"// ---------------- I/O (Utility)\n\n#define wuffs_base__utility__empty_io_reader wuffs_base__empty_io_reader\n#define wuffs_base__utility__empty_io_writer wuffs_base__empty_io_writer\n" +
""
-3
View File
@@ -10209,9 +10209,6 @@ wuffs_base__io_writer__limited_copy_u32_from_history(uint8_t** ptr_iop_w,
// expression '3'.
//
// See also wuffs_base__io_writer__limited_copy_u32_from_history_fast below.
//
// Alternatively or additionally, have a sloppy_limited_copy_u32_from_history
// method that copies 8 bytes at a time, which can more than length bytes?
for (; n >= 3; n -= 3) {
*p++ = *q++;
*p++ = *q++;