Clean up image-impl.c's loop_unroll_count

This commit is contained in:
Nigel Tao committed 2020-01-28 10:49:48 +11:00
1 parent 55aedb34f9
commit 7527103ec4
3 files changed
+120 -152

No files matched your search

+56 -72
View File
@@ -248,36 +248,35 @@ wuffs_base__pixel_swizzler__xx__index__src(wuffs_base__slice_u8 dst,
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
while (n >= N) {
while (n >= loop_unroll_count) {
wuffs_base__store_u16le(
d + (0 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[0] * 4)));
wuffs_base__store_u16le(
d + (1 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[1]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[1] * 4)));
wuffs_base__store_u16le(
d + (2 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[2] * 4)));
wuffs_base__store_u16le(
d + (3 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[3] * 4)));
s += 1 * N;
d += 2 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 2 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
wuffs_base__store_u16le(
d + (0 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[0] * 4)));
s += 1 * 1;
d += 2 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -296,8 +295,7 @@ wuffs_base__pixel_swizzler__xxx__index__src(wuffs_base__slice_u8 dst,
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
// The comparison in the while condition is ">", not ">=", because with ">=",
// the last 4-byte store could write past the end of the dst slice.
@@ -305,35 +303,34 @@ wuffs_base__pixel_swizzler__xxx__index__src(wuffs_base__slice_u8 dst,
// Each 4-byte store writes one too many bytes, but a subsequent store will
// overwrite that with the correct byte. There is always another store,
// whether a 4-byte store in this loop or a 1-byte store in the next loop.
while (n > N) {
while (n > loop_unroll_count) {
wuffs_base__store_u32le(
d + (0 * 3),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));
wuffs_base__store_u32le(
d + (1 * 3),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4)));
wuffs_base__store_u32le(
d + (2 * 3),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4)));
wuffs_base__store_u32le(
d + (3 * 3),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4)));
s += 1 * N;
d += 3 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 3 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
d[0] = (uint8_t)(s0 >> 0);
d[1] = (uint8_t)(s0 >> 8);
d[2] = (uint8_t)(s0 >> 16);
s += 1 * 1;
d += 3 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -353,46 +350,40 @@ wuffs_base__pixel_swizzler__xxx__index_binary_alpha__src_over(
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
while (n >= N) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
while (n >= loop_unroll_count) {
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
if (s0) {
wuffs_base__store_u24le(d + (0 * 4), s0);
}
uint32_t s1 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4));
uint32_t s1 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4));
if (s1) {
wuffs_base__store_u24le(d + (1 * 4), s1);
}
uint32_t s2 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4));
uint32_t s2 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4));
if (s2) {
wuffs_base__store_u24le(d + (2 * 4), s2);
}
uint32_t s3 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4));
uint32_t s3 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4));
if (s3) {
wuffs_base__store_u24le(d + (3 * 4), s3);
}
s += 1 * N;
d += 3 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 3 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
if (s0) {
wuffs_base__store_u24le(d + (0 * 4), s0);
}
s += 1 * 1;
d += 3 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -411,36 +402,35 @@ wuffs_base__pixel_swizzler__xxxx__index__src(wuffs_base__slice_u8 dst,
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
while (n >= N) {
while (n >= loop_unroll_count) {
wuffs_base__store_u32le(
d + (0 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));
wuffs_base__store_u32le(
d + (1 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4)));
wuffs_base__store_u32le(
d + (2 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4)));
wuffs_base__store_u32le(
d + (3 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4)));
s += 1 * N;
d += 4 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 4 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
wuffs_base__store_u32le(
d + (0 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));
s += 1 * 1;
d += 4 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -460,46 +450,40 @@ wuffs_base__pixel_swizzler__xxxx__index_binary_alpha__src_over(
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
while (n >= N) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
while (n >= loop_unroll_count) {
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
if (s0) {
wuffs_base__store_u32le(d + (0 * 4), s0);
}
uint32_t s1 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4));
uint32_t s1 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4));
if (s1) {
wuffs_base__store_u32le(d + (1 * 4), s1);
}
uint32_t s2 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4));
uint32_t s2 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4));
if (s2) {
wuffs_base__store_u32le(d + (2 * 4), s2);
}
uint32_t s3 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4));
uint32_t s3 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4));
if (s3) {
wuffs_base__store_u32le(d + (3 * 4), s3);
}
s += 1 * N;
d += 4 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 4 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
if (s0) {
wuffs_base__store_u32le(d + (0 * 4), s0);
}
s += 1 * 1;
d += 4 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -523,7 +507,7 @@ wuffs_base__pixel_swizzler__xxxx__y(wuffs_base__slice_u8 dst,
s += 1 * 1;
d += 4 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
+8 -8
View File
@@ -35,14 +35,14 @@ const baseImageImplC = "" +
" uint32_t r5 = 0x1F & (color >> (24 - 5));\n wuffs_base__store_u16le(row + (2 * ((size_t)x)),\n (b5 << 0) | (g6 << 5) | (r5 << 11));\n break;\n }\n case WUFFS_BASE__PIXEL_FORMAT__BGR:\n wuffs_base__store_u24le(row + (3 * ((size_t)x)), color);\n break;\n case WUFFS_BASE__PIXEL_FORMAT__BGRA_NONPREMUL:\n wuffs_base__store_u32le(row + (4 * ((size_t)x)),\n wuffs_base__nonpremul_u32_axxx(color));\n break;\n\n case WUFFS_BASE__PIXEL_FORMAT__RGB:\n wuffs_base__store_u24le(row + (3 * ((size_t)x)),\n wuffs_base__swap_u32_argb_abgr(color));\n break;\n case WUFFS_BASE__PIXEL_FORMAT__RGBA_NONPREMUL:\n wuffs_base__store_u32le(row + (4 * ((size_t)x)),\n wuffs_base__nonpremul_u32_axxx(\n wuffs_base__swap_u32_argb_abgr(color)));\n break;\n case WUFFS_BASE__PIXEL_FORMAT__RGBA_PREMUL:\n WUFFS_BASE__FALLTHROUGH;\n case WUFFS_BA" +
"SE__PIXEL_FORMAT__RGBX:\n wuffs_base__store_u32le(row + (4 * ((size_t)x)),\n wuffs_base__swap_u32_argb_abgr(color));\n break;\n\n default:\n // TODO: support more formats.\n return wuffs_base__make_status(wuffs_base__error__unsupported_option);\n }\n\n return wuffs_base__make_status(NULL);\n}\n\n" +
"" +
"// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__copy_1_1(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n return wuffs_base__slice_u8__copy_from_slice(dst, src);\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xx__index__src(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len2 = dst.len / 2;\n size_t len = dst_len2 < src.len ? dst_len2 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n // N is the loop unroll count.\n const int N = 4;\n\n while (n >= N) {\n wuffs_base__store_u16le(\n d + (0 * 2),\n wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));\n wuffs_base__store_u16le(\n d + (1 * 2),\n wuffs_base__load_u16le(dst_palet" +
"te.ptr + ((uint32_t)(s[1]) * 4)));\n wuffs_base__store_u16le(\n d + (2 * 2),\n wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));\n wuffs_base__store_u16le(\n d + (3 * 2),\n wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));\n\n s += 1 * N;\n d += 2 * N;\n n -= (size_t)(1 * N);\n }\n\n while (n >= 1) {\n wuffs_base__store_u16le(\n d + (0 * 2),\n wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));\n\n s += 1 * 1;\n d += 2 * 1;\n n -= (size_t)(1 * 1);\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxx__index__src(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len3 = dst.len / 3;\n size_t len = dst_len3 < src.len ? dst_len3 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n /" +
"/ N is the loop unroll count.\n const int N = 4;\n\n // The comparison in the while condition is \">\", not \">=\", because with \">=\",\n // the last 4-byte store could write past the end of the dst slice.\n //\n // Each 4-byte store writes one too many bytes, but a subsequent store will\n // overwrite that with the correct byte. There is always another store,\n // whether a 4-byte store in this loop or a 1-byte store in the next loop.\n while (n > N) {\n wuffs_base__store_u32le(\n d + (0 * 3),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));\n wuffs_base__store_u32le(\n d + (1 * 3),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4)));\n wuffs_base__store_u32le(\n d + (2 * 3),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));\n wuffs_base__store_u32le(\n d + (3 * 3),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));\n\n s += 1 * N;\n d += 3 * N;\n n -= (size_t)(1 * N);\n }\n\n while " +
"(n >= 1) {\n uint32_t s0 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));\n d[0] = (uint8_t)(s0 >> 0);\n d[1] = (uint8_t)(s0 >> 8);\n d[2] = (uint8_t)(s0 >> 16);\n\n s += 1 * 1;\n d += 3 * 1;\n n -= (size_t)(1 * 1);\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxx__index_binary_alpha__src_over(\n wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len3 = dst.len / 3;\n size_t len = dst_len3 < src.len ? dst_len3 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n // N is the loop unroll count.\n const int N = 4;\n\n while (n >= N) {\n uint32_t s0 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));\n if (s0) {\n wuffs_base__store_u24le(d + (0 * 4), s0);\n }\n uint32_t s1 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4));\n if (s1) {\n wuffs_base__" +
"store_u24le(d + (1 * 4), s1);\n }\n uint32_t s2 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4));\n if (s2) {\n wuffs_base__store_u24le(d + (2 * 4), s2);\n }\n uint32_t s3 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4));\n if (s3) {\n wuffs_base__store_u24le(d + (3 * 4), s3);\n }\n\n s += 1 * N;\n d += 3 * N;\n n -= (size_t)(1 * N);\n }\n\n while (n >= 1) {\n uint32_t s0 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));\n if (s0) {\n wuffs_base__store_u24le(d + (0 * 4), s0);\n }\n\n s += 1 * 1;\n d += 3 * 1;\n n -= (size_t)(1 * 1);\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__index__src(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len4 = dst.len / 4;\n size_t len = ds" +
"t_len4 < src.len ? dst_len4 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n // N is the loop unroll count.\n const int N = 4;\n\n while (n >= N) {\n wuffs_base__store_u32le(\n d + (0 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));\n wuffs_base__store_u32le(\n d + (1 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4)));\n wuffs_base__store_u32le(\n d + (2 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));\n wuffs_base__store_u32le(\n d + (3 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));\n\n s += 1 * N;\n d += 4 * N;\n n -= (size_t)(1 * N);\n }\n\n while (n >= 1) {\n wuffs_base__store_u32le(\n d + (0 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));\n\n s += 1 * 1;\n d += 4 * 1;\n n -= (size_t)(1 * 1);\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__ind" +
"ex_binary_alpha__src_over(\n wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len4 = dst.len / 4;\n size_t len = dst_len4 < src.len ? dst_len4 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n // N is the loop unroll count.\n const int N = 4;\n\n while (n >= N) {\n uint32_t s0 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));\n if (s0) {\n wuffs_base__store_u32le(d + (0 * 4), s0);\n }\n uint32_t s1 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4));\n if (s1) {\n wuffs_base__store_u32le(d + (1 * 4), s1);\n }\n uint32_t s2 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4));\n if (s2) {\n wuffs_base__store_u32le(d + (2 * 4), s2);\n }\n uint32_t s3 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4));\n if (s3) {\n wuffs_base__store_u32le(d " +
"+ (3 * 4), s3);\n }\n\n s += 1 * N;\n d += 4 * N;\n n -= (size_t)(1 * N);\n }\n\n while (n >= 1) {\n uint32_t s0 =\n wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));\n if (s0) {\n wuffs_base__store_u32le(d + (0 * 4), s0);\n }\n\n s += 1 * 1;\n d += 4 * 1;\n n -= (size_t)(1 * 1);\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__y(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n size_t dst_len4 = dst.len / 4;\n size_t len = dst_len4 < src.len ? dst_len4 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n // TODO: unroll.\n\n while (n >= 1) {\n wuffs_base__store_u32le(d + (0 * 4),\n 0xFF000000 | (0x010101 * (uint32_t)s[0]));\n\n s += 1 * 1;\n d += 4 * 1;\n n -= (size_t)(1 * 1);\n }\n\n return len;\n}\n\n" +
"// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__copy_1_1(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n return wuffs_base__slice_u8__copy_from_slice(dst, src);\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xx__index__src(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len2 = dst.len / 2;\n size_t len = dst_len2 < src.len ? dst_len2 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n const size_t loop_unroll_count = 4;\n\n while (n >= loop_unroll_count) {\n wuffs_base__store_u16le(\n d + (0 * 2),\n wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[0] * 4)));\n wuffs_base__store_u16le(\n d + (1 * 2),\n wuffs_base__load_u16le(dst_palette" +
".ptr + ((size_t)s[1] * 4)));\n wuffs_base__store_u16le(\n d + (2 * 2),\n wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[2] * 4)));\n wuffs_base__store_u16le(\n d + (3 * 2),\n wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[3] * 4)));\n\n s += 1 * loop_unroll_count;\n d += 2 * loop_unroll_count;\n n -= loop_unroll_count;\n }\n\n while (n >= 1) {\n wuffs_base__store_u16le(\n d + (0 * 2),\n wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[0] * 4)));\n\n s += 1 * 1;\n d += 2 * 1;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxx__index__src(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len3 = dst.len / 3;\n size_t len = dst_len3 < src.len ? dst_len3 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n " +
" const size_t loop_unroll_count = 4;\n\n // The comparison in the while condition is \">\", not \">=\", because with \">=\",\n // the last 4-byte store could write past the end of the dst slice.\n //\n // Each 4-byte store writes one too many bytes, but a subsequent store will\n // overwrite that with the correct byte. There is always another store,\n // whether a 4-byte store in this loop or a 1-byte store in the next loop.\n while (n > loop_unroll_count) {\n wuffs_base__store_u32le(\n d + (0 * 3),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));\n wuffs_base__store_u32le(\n d + (1 * 3),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4)));\n wuffs_base__store_u32le(\n d + (2 * 3),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4)));\n wuffs_base__store_u32le(\n d + (3 * 3),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4)));\n\n s += 1 * loop_unroll_count;\n d += 3 * loop_unroll_count;\n n -= loop_unrol" +
"l_count;\n }\n\n while (n >= 1) {\n uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));\n d[0] = (uint8_t)(s0 >> 0);\n d[1] = (uint8_t)(s0 >> 8);\n d[2] = (uint8_t)(s0 >> 16);\n\n s += 1 * 1;\n d += 3 * 1;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxx__index_binary_alpha__src_over(\n wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len3 = dst.len / 3;\n size_t len = dst_len3 < src.len ? dst_len3 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n const size_t loop_unroll_count = 4;\n\n while (n >= loop_unroll_count) {\n uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));\n if (s0) {\n wuffs_base__store_u24le(d + (0 * 4), s0);\n }\n uint32_t s1 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4));\n if (s1) {\n wuffs_base__store_u24le(d + (1 * 4), s" +
"1);\n }\n uint32_t s2 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4));\n if (s2) {\n wuffs_base__store_u24le(d + (2 * 4), s2);\n }\n uint32_t s3 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4));\n if (s3) {\n wuffs_base__store_u24le(d + (3 * 4), s3);\n }\n\n s += 1 * loop_unroll_count;\n d += 3 * loop_unroll_count;\n n -= loop_unroll_count;\n }\n\n while (n >= 1) {\n uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));\n if (s0) {\n wuffs_base__store_u24le(d + (0 * 4), s0);\n }\n\n s += 1 * 1;\n d += 3 * 1;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__index__src(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len4 = dst.len / 4;\n size_t len = dst_len4 < src.len ? dst_len4 : src.len;\n u" +
"int8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n const size_t loop_unroll_count = 4;\n\n while (n >= loop_unroll_count) {\n wuffs_base__store_u32le(\n d + (0 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));\n wuffs_base__store_u32le(\n d + (1 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4)));\n wuffs_base__store_u32le(\n d + (2 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4)));\n wuffs_base__store_u32le(\n d + (3 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4)));\n\n s += 1 * loop_unroll_count;\n d += 4 * loop_unroll_count;\n n -= loop_unroll_count;\n }\n\n while (n >= 1) {\n wuffs_base__store_u32le(\n d + (0 * 4),\n wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));\n\n s += 1 * 1;\n d += 4 * 1;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__index_binary_alpha__src_over(\n wuffs_bas" +
"e__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n if (dst_palette.len != 1024) {\n return 0;\n }\n size_t dst_len4 = dst.len / 4;\n size_t len = dst_len4 < src.len ? dst_len4 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n const size_t loop_unroll_count = 4;\n\n while (n >= loop_unroll_count) {\n uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));\n if (s0) {\n wuffs_base__store_u32le(d + (0 * 4), s0);\n }\n uint32_t s1 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4));\n if (s1) {\n wuffs_base__store_u32le(d + (1 * 4), s1);\n }\n uint32_t s2 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4));\n if (s2) {\n wuffs_base__store_u32le(d + (2 * 4), s2);\n }\n uint32_t s3 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4));\n if (s3) {\n wuffs_base__store_u32le(d + (3 * 4), s3);\n }\n\n s += 1 * loop_unroll_count;\n d += 4 * loop_unroll_count;" +
"\n n -= loop_unroll_count;\n }\n\n while (n >= 1) {\n uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));\n if (s0) {\n wuffs_base__store_u32le(d + (0 * 4), s0);\n }\n\n s += 1 * 1;\n d += 4 * 1;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__xxxx__y(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n size_t dst_len4 = dst.len / 4;\n size_t len = dst_len4 < src.len ? dst_len4 : src.len;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n // TODO: unroll.\n\n while (n >= 1) {\n wuffs_base__store_u32le(d + (0 * 4),\n 0xFF000000 | (0x010101 * (uint32_t)s[0]));\n\n s += 1 * 1;\n d += 4 * 1;\n n -= 1;\n }\n\n return len;\n}\n\n" +
"" +
"// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__squash_bgr_565_888(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 src) {\n size_t len4 = (dst.len < src.len ? dst.len : src.len) / 4;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n\n size_t n = len4;\n while (n--) {\n uint32_t argb = wuffs_base__load_u32le(s);\n uint32_t b5 = 0x1F & (argb >> (8 - 5));\n uint32_t g6 = 0x3F & (argb >> (16 - 6));\n uint32_t r5 = 0x1F & (argb >> (24 - 5));\n wuffs_base__store_u32le(d, (b5 << 0) | (g6 << 5) | (r5 << 11));\n s += 4;\n d += 4;\n }\n return len4 * 4;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__swap_rgbx_bgrx(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 src) {\n size_t len4 = (dst.len < src.len ? dst.len : src.len) / 4;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n\n size_t n = len4;\n while (n--) {\n uint8_t b0 = s[0];\n uint8_t b1 = s[1];\n uint8_t b2 = s[2];\n uint8_t b3 = s[3];\n " +
" d[0] = b2;\n d[1] = b1;\n d[2] = b0;\n d[3] = b3;\n s += 4;\n d += 4;\n }\n return len4 * 4;\n}\n\n" +
+56 -72
View File
@@ -6076,36 +6076,35 @@ wuffs_base__pixel_swizzler__xx__index__src(wuffs_base__slice_u8 dst,
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
while (n >= N) {
while (n >= loop_unroll_count) {
wuffs_base__store_u16le(
d + (0 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[0] * 4)));
wuffs_base__store_u16le(
d + (1 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[1]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[1] * 4)));
wuffs_base__store_u16le(
d + (2 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[2] * 4)));
wuffs_base__store_u16le(
d + (3 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[3] * 4)));
s += 1 * N;
d += 2 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 2 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
wuffs_base__store_u16le(
d + (0 * 2),
wuffs_base__load_u16le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u16le(dst_palette.ptr + ((size_t)s[0] * 4)));
s += 1 * 1;
d += 2 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -6124,8 +6123,7 @@ wuffs_base__pixel_swizzler__xxx__index__src(wuffs_base__slice_u8 dst,
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
// The comparison in the while condition is ">", not ">=", because with ">=",
// the last 4-byte store could write past the end of the dst slice.
@@ -6133,35 +6131,34 @@ wuffs_base__pixel_swizzler__xxx__index__src(wuffs_base__slice_u8 dst,
// Each 4-byte store writes one too many bytes, but a subsequent store will
// overwrite that with the correct byte. There is always another store,
// whether a 4-byte store in this loop or a 1-byte store in the next loop.
while (n > N) {
while (n > loop_unroll_count) {
wuffs_base__store_u32le(
d + (0 * 3),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));
wuffs_base__store_u32le(
d + (1 * 3),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4)));
wuffs_base__store_u32le(
d + (2 * 3),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4)));
wuffs_base__store_u32le(
d + (3 * 3),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4)));
s += 1 * N;
d += 3 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 3 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
d[0] = (uint8_t)(s0 >> 0);
d[1] = (uint8_t)(s0 >> 8);
d[2] = (uint8_t)(s0 >> 16);
s += 1 * 1;
d += 3 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -6181,46 +6178,40 @@ wuffs_base__pixel_swizzler__xxx__index_binary_alpha__src_over(
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
while (n >= N) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
while (n >= loop_unroll_count) {
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
if (s0) {
wuffs_base__store_u24le(d + (0 * 4), s0);
}
uint32_t s1 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4));
uint32_t s1 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4));
if (s1) {
wuffs_base__store_u24le(d + (1 * 4), s1);
}
uint32_t s2 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4));
uint32_t s2 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4));
if (s2) {
wuffs_base__store_u24le(d + (2 * 4), s2);
}
uint32_t s3 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4));
uint32_t s3 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4));
if (s3) {
wuffs_base__store_u24le(d + (3 * 4), s3);
}
s += 1 * N;
d += 3 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 3 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
if (s0) {
wuffs_base__store_u24le(d + (0 * 4), s0);
}
s += 1 * 1;
d += 3 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -6239,36 +6230,35 @@ wuffs_base__pixel_swizzler__xxxx__index__src(wuffs_base__slice_u8 dst,
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
while (n >= N) {
while (n >= loop_unroll_count) {
wuffs_base__store_u32le(
d + (0 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));
wuffs_base__store_u32le(
d + (1 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4)));
wuffs_base__store_u32le(
d + (2 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4)));
wuffs_base__store_u32le(
d + (3 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4)));
s += 1 * N;
d += 4 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 4 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
wuffs_base__store_u32le(
d + (0 * 4),
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4)));
wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4)));
s += 1 * 1;
d += 4 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -6288,46 +6278,40 @@ wuffs_base__pixel_swizzler__xxxx__index_binary_alpha__src_over(
uint8_t* s = src.ptr;
size_t n = len;
// N is the loop unroll count.
const int N = 4;
const size_t loop_unroll_count = 4;
while (n >= N) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
while (n >= loop_unroll_count) {
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
if (s0) {
wuffs_base__store_u32le(d + (0 * 4), s0);
}
uint32_t s1 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[1]) * 4));
uint32_t s1 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[1] * 4));
if (s1) {
wuffs_base__store_u32le(d + (1 * 4), s1);
}
uint32_t s2 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[2]) * 4));
uint32_t s2 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[2] * 4));
if (s2) {
wuffs_base__store_u32le(d + (2 * 4), s2);
}
uint32_t s3 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[3]) * 4));
uint32_t s3 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[3] * 4));
if (s3) {
wuffs_base__store_u32le(d + (3 * 4), s3);
}
s += 1 * N;
d += 4 * N;
n -= (size_t)(1 * N);
s += 1 * loop_unroll_count;
d += 4 * loop_unroll_count;
n -= loop_unroll_count;
}
while (n >= 1) {
uint32_t s0 =
wuffs_base__load_u32le(dst_palette.ptr + ((uint32_t)(s[0]) * 4));
uint32_t s0 = wuffs_base__load_u32le(dst_palette.ptr + ((size_t)s[0] * 4));
if (s0) {
wuffs_base__store_u32le(d + (0 * 4), s0);
}
s += 1 * 1;
d += 4 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;
@@ -6351,7 +6335,7 @@ wuffs_base__pixel_swizzler__xxxx__y(wuffs_base__slice_u8 dst,
s += 1 * 1;
d += 4 * 1;
n -= (size_t)(1 * 1);
n -= 1;
}
return len;