Collect GIF's LZW data into larger batches

name                                    old speed     new speed     delta

wuffs_gif_decode_1k_bw/clang            224MB/s ± 1%  227MB/s ± 0%  +0.98%  (p=0.000 n=10+9)
wuffs_gif_decode_1k_color/clang         116MB/s ± 1%  118MB/s ± 1%  +2.26%  (p=0.000 n=10+10)
wuffs_gif_decode_10k_bgra/clang         541MB/s ± 0%  568MB/s ± 1%  +4.97%  (p=0.000 n=9+10)
wuffs_gif_decode_10k_indexed/clang      144MB/s ± 0%  152MB/s ± 0%  +5.61%  (p=0.000 n=9+10)
wuffs_gif_decode_100k_artificial/clang  438MB/s ± 1%  454MB/s ± 1%  +3.56%  (p=0.000 n=9+10)
wuffs_gif_decode_100k_realistic/clang   171MB/s ± 0%  179MB/s ± 0%  +4.64%  (p=0.000 n=8+10)
wuffs_gif_decode_1000k/clang            174MB/s ± 0%  180MB/s ± 0%  +3.90%  (p=0.000 n=9+10)
wuffs_gif_decode_anim_screencap/clang   897MB/s ± 1%  943MB/s ± 0%  +5.10%  (p=0.000 n=9+9)

wuffs_gif_decode_1k_bw/gcc              250MB/s ± 1%  241MB/s ± 1%  -3.26%  (p=0.000 n=10+10)
wuffs_gif_decode_1k_color/gcc           132MB/s ± 1%  130MB/s ± 1%  -2.15%  (p=0.000 n=10+10)
wuffs_gif_decode_10k_bgra/gcc           534MB/s ± 1%  535MB/s ± 1%    ~     (p=0.211 n=9+10)
wuffs_gif_decode_10k_indexed/gcc        162MB/s ± 1%  163MB/s ± 0%    ~     (p=0.105 n=10+10)
wuffs_gif_decode_100k_artificial/gcc    460MB/s ± 0%  456MB/s ± 0%  -0.80%  (p=0.000 n=10+10)
wuffs_gif_decode_100k_realistic/gcc     186MB/s ± 0%  186MB/s ± 1%  -0.33%  (p=0.001 n=9+10)
wuffs_gif_decode_1000k/gcc              189MB/s ± 1%  188MB/s ± 1%    ~     (p=0.053 n=9+10)
wuffs_gif_decode_anim_screencap/gcc     981MB/s ± 1%  971MB/s ± 1%  -0.95%  (p=0.000 n=10+10)
This commit is contained in:
Nigel Tao committed 2018-12-15 14:10:10 +11:00
1 parent e497b38da0
commit 33a4557d98
9 files changed
+325 -158

No files matched your search

+39 -7
View File
@@ -188,6 +188,26 @@ wuffs_base__io_writer__copy_n_from_slice(uint8_t** ptr_ioptr_w,
return n;
}
static inline wuffs_base__empty_struct //
wuffs_base__io_reader__set(wuffs_base__io_reader* o,
wuffs_base__io_buffer* b,
uint8_t** ptr_iop,
uint8_t** ptr_io1,
wuffs_base__slice_u8 data) {
b->data = data;
b->meta.wi = data.len;
b->meta.ri = 0;
b->meta.pos = 0;
b->meta.closed = false;
o->private_impl.buf = b;
o->private_impl.mark = data.ptr;
o->private_impl.limit = data.ptr + data.len;
*ptr_iop = data.ptr;
*ptr_io1 = data.ptr + data.len;
return ((wuffs_base__empty_struct){});
}
static inline wuffs_base__empty_struct //
wuffs_base__io_reader__set_limit(wuffs_base__io_reader* o,
uint8_t* ioptr_r,
@@ -204,24 +224,36 @@ wuffs_base__io_reader__set_mark(wuffs_base__io_reader* o, uint8_t* mark) {
return ((wuffs_base__empty_struct){});
}
static inline wuffs_base__slice_u8 //
wuffs_base__io_reader__take(uint8_t** ptr_iop, uint8_t* io1, uint64_t n) {
if (n <= (io1 - *ptr_iop)) {
uint8_t* p = *ptr_iop;
*ptr_iop += n;
return ((wuffs_base__slice_u8){
.ptr = p,
.len = n,
});
}
return ((wuffs_base__slice_u8){});
}
static inline wuffs_base__empty_struct //
wuffs_base__io_writer__set(wuffs_base__io_writer* o,
wuffs_base__io_buffer* b,
uint8_t** ioptr1_ptr,
uint8_t** ioptr2_ptr,
wuffs_base__slice_u8 s) {
b->data.ptr = s.ptr;
b->data.len = s.len;
wuffs_base__slice_u8 data) {
b->data = data;
b->meta.wi = 0;
b->meta.ri = 0;
b->meta.pos = 0;
b->meta.closed = false;
o->private_impl.buf = b;
o->private_impl.mark = s.ptr;
o->private_impl.limit = s.ptr + s.len;
*ioptr1_ptr = s.ptr;
*ioptr2_ptr = s.ptr + s.len;
o->private_impl.mark = data.ptr;
o->private_impl.limit = data.ptr + data.len;
*ioptr1_ptr = data.ptr;
*ioptr2_ptr = data.ptr + data.len;
return ((wuffs_base__empty_struct){});
}
+15 -7
View File
@@ -104,12 +104,11 @@ func (g *gen) writeBuiltinIO(b *buffer, recv *a.Expr, method t.ID, args []*a.Nod
return nil
case t.IDSet:
typ := "reader"
typ, v := "reader", "r"
if len(args) == 1 {
typ = "writer"
typ, v = "writer", "w"
}
// TODO: don't hard-code v_w and u_w.
b.printf("wuffs_base__io_%s__set(&v_w, &u_w, &iop_v_w, &io1_v_w,", typ)
b.printf("wuffs_base__io_%s__set(&v_%s, &u_%s, &iop_v_%s, &io1_v_%s,", typ, v, v, v, v)
return g.writeArgs(b, args, depth)
}
@@ -142,13 +141,18 @@ func (g *gen) writeBuiltinIOReader(b *buffer, recv *a.Expr, method t.ID, args []
return nil
case t.IDSinceMark, t.IDSinceMarkLength:
prefix, name := aPrefix, "src"
if recv.Operator() == 0 {
prefix, name = vPrefix, recv.Ident().Str(g.tm)
}
if method == t.IDSinceMark {
b.printf("((wuffs_base__slice_u8){ "+
".ptr = %ssrc.private_impl.mark, "+
".ptr = %s%s.private_impl.mark, "+
".len = (size_t)(",
aPrefix)
prefix, name)
}
b.printf("iop_a_src - %ssrc.private_impl.mark", aPrefix)
b.printf("iop_%s%s - %s%s.private_impl.mark", prefix, name, prefix, name)
if method == t.IDSinceMark {
b.writes("), })")
}
@@ -165,6 +169,10 @@ func (g *gen) writeBuiltinIOReader(b *buffer, recv *a.Expr, method t.ID, args []
}
b.writes(", wuffs_base__return_empty_struct())")
return nil
case t.IDTake:
b.printf("wuffs_base__io_reader__take(&iop_a_src, io1_a_src,")
return g.writeArgs(b, args, depth)
}
if method >= peekMethodsBase {
+4 -3
View File
@@ -151,9 +151,10 @@ const baseIOPrivateH = "" +
"t8_t** ptr_ptr,\n uint8_t* start,\n uint8_t* end,\n uint32_t length,\n uint32_t distance) {\n if (!distance) {\n return 0;\n }\n uint8_t* ptr = *ptr_ptr;\n if ((size_t)(ptr - start) < (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: 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 // copy_n_from_history Wuffs method should also take an unroll hint argument,\n // and the cgen can look if that argument is the constant expression '3'.\n //\n // See also wuffs_base__io_writer__copy_n_from_history_fast below.\n //\n // Alternatively, or additionally, ha" +
"ve a sloppy_copy_n_from_history method\n // that copies 8 bytes at a time, possibly writing more than length bytes?\n for (; n >= 3; n -= 3) {\n *ptr++ = *start++;\n *ptr++ = *start++;\n *ptr++ = *start++;\n }\n for (; n; n--) {\n *ptr++ = *start++;\n }\n *ptr_ptr = ptr;\n return length;\n}\n\n// wuffs_base__io_writer__copy_n_from_history_fast is like the\n// wuffs_base__io_writer__copy_n_from_history function above, but has stronger\n// pre-conditions. The caller needs to prove that:\n// - distance > 0\n// - distance <= (*ptr_ptr - start)\n// - length <= (end - *ptr_ptr)\nstatic inline uint32_t //\nwuffs_base__io_writer__copy_n_from_history_fast(uint8_t** ptr_ptr,\n uint8_t* start,\n uint8_t* end,\n uint32_t length,\n uint32_t distance) {\n uint8_t* ptr = *ptr_ptr;\n start = ptr - distance;\n uint32_t n = length;\n fo" +
"r (; n >= 3; n -= 3) {\n *ptr++ = *start++;\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 //\nwuffs_base__io_writer__copy_n_from_reader(uint8_t** ptr_ioptr_w,\n uint8_t* iobounds1_w,\n uint32_t length,\n uint8_t** ptr_ioptr_r,\n uint8_t* iobounds1_r) {\n uint8_t* ioptr_w = *ptr_ioptr_w;\n size_t n = length;\n if (n > ((size_t)(iobounds1_w - ioptr_w))) {\n n = iobounds1_w - ioptr_w;\n }\n uint8_t* ioptr_r = *ptr_ioptr_r;\n if (n > ((size_t)(iobounds1_r - ioptr_r))) {\n n = iobounds1_r - ioptr_r;\n }\n if (n > 0) {\n memmove(ioptr_w, ioptr_r, n);\n *ptr_ioptr_w += n;\n *ptr_ioptr_r += n;\n }\n return n;\n}\n\nstatic inline uint64_t //\nwuffs_base__io_writer__copy_from_slice(uint8_t** ptr_ioptr_w,\n u" +
"int8_t* iobounds1_w,\n wuffs_base__slice_u8 src) {\n uint8_t* ioptr_w = *ptr_ioptr_w;\n size_t n = src.len;\n if (n > ((size_t)(iobounds1_w - ioptr_w))) {\n n = iobounds1_w - ioptr_w;\n }\n if (n > 0) {\n memmove(ioptr_w, src.ptr, n);\n *ptr_ioptr_w += n;\n }\n return n;\n}\n\nstatic inline uint32_t //\nwuffs_base__io_writer__copy_n_from_slice(uint8_t** ptr_ioptr_w,\n uint8_t* iobounds1_w,\n uint32_t length,\n wuffs_base__slice_u8 src) {\n uint8_t* ioptr_w = *ptr_ioptr_w;\n size_t n = src.len;\n if (n > length) {\n n = length;\n }\n if (n > ((size_t)(iobounds1_w - ioptr_w))) {\n n = iobounds1_w - ioptr_w;\n }\n if (n > 0) {\n memmove(ioptr_w, src.ptr, n);\n *ptr_ioptr_w += n;\n }\n return n;\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_reader__set_limit(wuffs_base__io_reader* o,\n uint8_t* ioptr_r,\n " +
" uint64_t limit) {\n if (o && (((size_t)(o->private_impl.limit - ioptr_r)) > limit)) {\n o->private_impl.limit = ioptr_r + limit;\n }\n return ((wuffs_base__empty_struct){});\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_reader__set_mark(wuffs_base__io_reader* o, uint8_t* mark) {\n o->private_impl.mark = mark;\n return ((wuffs_base__empty_struct){});\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_writer__set(wuffs_base__io_writer* o,\n wuffs_base__io_buffer* b,\n uint8_t** ioptr1_ptr,\n uint8_t** ioptr2_ptr,\n wuffs_base__slice_u8 s) {\n b->data.ptr = s.ptr;\n b->data.len = s.len;\n b->meta.wi = 0;\n b->meta.ri = 0;\n b->meta.pos = 0;\n b->meta.closed = false;\n\n o->private_impl.buf = b;\n o->private_impl.mark = s.ptr;\n o->private_impl.limit = s.ptr + s.len;\n *ioptr1_ptr = s.ptr;\n *ioptr2_ptr = s.ptr + s.len;\n return ((wuffs_base__empty_struct){})" +
";\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_writer__set_mark(wuffs_base__io_writer* o, uint8_t* mark) {\n o->private_impl.mark = mark;\n return ((wuffs_base__empty_struct){});\n}\n" +
"int8_t* iobounds1_w,\n wuffs_base__slice_u8 src) {\n uint8_t* ioptr_w = *ptr_ioptr_w;\n size_t n = src.len;\n if (n > ((size_t)(iobounds1_w - ioptr_w))) {\n n = iobounds1_w - ioptr_w;\n }\n if (n > 0) {\n memmove(ioptr_w, src.ptr, n);\n *ptr_ioptr_w += n;\n }\n return n;\n}\n\nstatic inline uint32_t //\nwuffs_base__io_writer__copy_n_from_slice(uint8_t** ptr_ioptr_w,\n uint8_t* iobounds1_w,\n uint32_t length,\n wuffs_base__slice_u8 src) {\n uint8_t* ioptr_w = *ptr_ioptr_w;\n size_t n = src.len;\n if (n > length) {\n n = length;\n }\n if (n > ((size_t)(iobounds1_w - ioptr_w))) {\n n = iobounds1_w - ioptr_w;\n }\n if (n > 0) {\n memmove(ioptr_w, src.ptr, n);\n *ptr_ioptr_w += n;\n }\n return n;\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_reader__set(wuffs_base__io_reader* o,\n wuffs_base__io_buffer* b,\n " +
" uint8_t** ptr_iop,\n uint8_t** ptr_io1,\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 o->private_impl.buf = b;\n o->private_impl.mark = data.ptr;\n o->private_impl.limit = data.ptr + data.len;\n *ptr_iop = data.ptr;\n *ptr_io1 = data.ptr + data.len;\n return ((wuffs_base__empty_struct){});\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_reader__set_limit(wuffs_base__io_reader* o,\n uint8_t* ioptr_r,\n uint64_t limit) {\n if (o && (((size_t)(o->private_impl.limit - ioptr_r)) > limit)) {\n o->private_impl.limit = ioptr_r + limit;\n }\n return ((wuffs_base__empty_struct){});\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_reader__set_mark(wuffs_base__io_reader* o, uint8_t* mark) {\n o->private_impl.mark = mark;\n return ((wuffs_base__empty_struct){});\n}\n\nstatic " +
"inline wuffs_base__slice_u8 //\nwuffs_base__io_reader__take(uint8_t** ptr_iop, uint8_t* io1, uint64_t n) {\n if (n <= (io1 - *ptr_iop)) {\n uint8_t* p = *ptr_iop;\n *ptr_iop += n;\n return ((wuffs_base__slice_u8){\n .ptr = p,\n .len = n,\n });\n }\n return ((wuffs_base__slice_u8){});\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_writer__set(wuffs_base__io_writer* o,\n wuffs_base__io_buffer* b,\n uint8_t** ioptr1_ptr,\n uint8_t** ioptr2_ptr,\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 o->private_impl.buf = b;\n o->private_impl.mark = data.ptr;\n o->private_impl.limit = data.ptr + data.len;\n *ioptr1_ptr = data.ptr;\n *ioptr2_ptr = data.ptr + data.len;\n return ((wuffs_base__empty_struct){});\n}\n\nstatic inline wuffs_base__empty_struct //\nwuffs_base__io_writer__set_mark(wuffs_base__io_writer*" +
" o, uint8_t* mark) {\n o->private_impl.mark = mark;\n return ((wuffs_base__empty_struct){});\n}\n" +
""
const baseIOPublicH = "" +
+12 -6
View File
@@ -89,6 +89,9 @@ func (g *gen) writeLoadDerivedVar(b *buffer, hack string, prefix string, name t.
if hack == "w" {
b.printf("%s%sw = %sw.data.ptr + %sw.meta.wi;\n", iopPrefix, vPrefix, uPrefix, uPrefix)
return nil
} else if hack == "r" {
b.printf("%s%sr = %sr.data.ptr + %sr.meta.ri;\n", iopPrefix, vPrefix, uPrefix, uPrefix)
return nil
}
if !typ.IsIOType() {
@@ -151,6 +154,9 @@ func (g *gen) writeSaveDerivedVar(b *buffer, hack string, prefix string, name t.
if hack == "w" {
b.printf("%sw.meta.wi = %s%sw - %sw.data.ptr;\n", uPrefix, iopPrefix, vPrefix, uPrefix)
return nil
} else if hack == "r" {
b.printf("%sr.meta.ri = %s%sr - %sr.data.ptr;\n", uPrefix, iopPrefix, vPrefix, uPrefix)
return nil
}
if !typ.IsIOType() {
@@ -188,11 +194,11 @@ func (g *gen) writeLoadExprDerivedVars(b *buffer, n *a.Expr) error {
prefix := aPrefix
// TODO: don't hard-code these.
hack := ""
if s := o.Value().Str(g.tm); s != "args.dst" && s != "args.src" && s != "w" {
if s := o.Value().Str(g.tm); s != "args.dst" && s != "args.src" && s != "w" && s != "r" {
continue
} else if s == "w" {
} else if s == "w" || s == "r" {
prefix = vPrefix
hack = "w"
hack = s
}
if err := g.writeLoadDerivedVar(b, hack, prefix, o.Name(), o.Value().MType(), false); err != nil {
return err
@@ -213,11 +219,11 @@ func (g *gen) writeSaveExprDerivedVars(b *buffer, n *a.Expr) error {
prefix := aPrefix
// TODO: don't hard-code these.
hack := ""
if s := o.Value().Str(g.tm); s != "args.dst" && s != "args.src" && s != "w" {
if s := o.Value().Str(g.tm); s != "args.dst" && s != "args.src" && s != "w" && s != "r" {
continue
} else if s == "w" {
} else if s == "w" || s == "r" {
prefix = vPrefix
hack = "w"
hack = s
}
if err := g.writeSaveDerivedVar(b, hack, prefix, o.Name(), o.Value().MType()); err != nil {
return err