From 1ca79e948aa4bf923b92757ea68a98b7bcd0767d Mon Sep 17 00:00:00 2001
From: Lars-Dominik Braun <lars@6xq.net>
Date: Fri, 13 Feb 2015 12:56:33 +0100
Subject: Drop support for atomic ops and bucket formats
Atomic ops are slower than locks, probably going to switch buckets to
uint64_t later.
---
flam3.c | 231 +---------------------------------------------------------------
rect.c | 26 --------
wscript | 2 -
3 files changed, 2 insertions(+), 257 deletions(-)
diff --git a/flam3.c b/flam3.c
index 2b4976f..f1b72be 100644
--- a/flam3.c
+++ b/flam3.c
@@ -1832,11 +1832,7 @@ void flam3_print(FILE *f, flam3_genome *cp, char *extra_attributes, int print_ed
if (getenv("intpalette"))
fprintf(f, "<color index=\"%d\" rgb=\"%d %d %d\"/>", i, (int)rint(r), (int)rint(g), (int)rint(b));
else {
-#ifdef USE_FLOAT_INDICES
- fprintf(f, "<color index=\"%.10g\" rgb=\"%.6g %.6g %.6g\"/>", cp->palette[i].index, r, g, b);
-#else
fprintf(f, "<color index=\"%d\" rgb=\"%.6g %.6g %.6g\"/>", i, r, g, b);
-#endif
}
} else {
if (getenv("intpalette"))
@@ -3593,180 +3589,6 @@ typedef unsigned int abucket_int[4];
typedef float bucket_float[5];
typedef float abucket_float[4];
-#ifdef HAVE_GCC_64BIT_ATOMIC_OPS
-static inline void
-double_atomic_add(double *dest, double delta)
-{
- uint64_t *int_ptr = (uint64_t *)dest;
- union {
- double dblval;
- uint64_t intval;
- } old_val, new_val;
- int success;
-
- do {
- old_val.dblval = *dest;
- new_val.dblval = old_val.dblval + delta;
- success = __sync_bool_compare_and_swap(
- int_ptr, old_val.intval, new_val.intval);
- } while (!success);
-}
-#endif /* HAVE_GCC_64BIT_ATOMIC_OPS */
-
-#ifdef HAVE_GCC_ATOMIC_OPS
-static inline void
-float_atomic_add(float *dest, float delta)
-{
- uint32_t *int_ptr = (uint32_t *)dest;
- union {
- float fltval;
- uint32_t intval;
- } old_val, new_val;
- int success;
-
- do {
- old_val.fltval = *dest;
- new_val.fltval = old_val.fltval + delta;
- success = __sync_bool_compare_and_swap(
- int_ptr, old_val.intval, new_val.intval);
- } while (!success);
-}
-
-static inline void
-uint_atomic_add(unsigned int *dest, unsigned int delta)
-{
- unsigned int old_val, new_val;
- int success;
-
- do {
- old_val = *dest;
- if (UINT_MAX - old_val > delta)
- new_val = old_val + delta;
- else
- new_val = UINT_MAX;
- success = __sync_bool_compare_and_swap(
- dest, old_val, new_val);
- } while (!success);
-}
-
-static inline void
-ushort_atomic_add(unsigned short *dest, unsigned short delta)
-{
- unsigned short old_val, new_val;
- int success;
-
- do {
- old_val = *dest;
- if (USHRT_MAX - old_val > delta)
- new_val = old_val + delta;
- else
- new_val = USHRT_MAX;
- success = __sync_bool_compare_and_swap(
- dest, old_val, new_val);
- } while (!success);
-}
-#endif /* HAVE_GCC_ATOMIC_OPS */
-
-/* 64-bit datatypes */
-#define bucket bucket_double
-#define abucket abucket_double
-#define abump_no_overflow(dest, delta) do {dest += delta;} while (0)
-#define add_c_to_accum(acc,i,ii,j,jj,wid,hgt,c) do { \
- if ( (j) + (jj) >=0 && (j) + (jj) < (hgt) && (i) + (ii) >=0 && (i) + (ii) < (wid)) { \
- abucket *a = (acc) + ( (i) + (ii) ) + ( (j) + (jj) ) * (wid); \
- abump_no_overflow(a[0][0],(c)[0]); \
- abump_no_overflow(a[0][1],(c)[1]); \
- abump_no_overflow(a[0][2],(c)[2]); \
- abump_no_overflow(a[0][3],(c)[3]); \
- } \
-} while (0)
-/* single-threaded */
-#define USE_LOCKS
-#define bump_no_overflow(dest, delta) do {dest += delta;} while (0)
-#define render_rectangle render_rectangle_double
-#define iter_thread iter_thread_double
-#define de_thread_helper de_thread_helper_64
-#define de_thread de_thread_64
-#include "rect.c"
-#ifdef HAVE_GCC_64BIT_ATOMIC_OPS
- /* multi-threaded */
- #undef USE_LOCKS
- #undef bump_no_overflow
- #undef render_rectangle
- #undef iter_thread
- #undef de_thread_helper
- #undef de_thread
- #define bump_no_overflow(dest, delta) double_atomic_add(&dest, delta)
- #define render_rectangle render_rectangle_double_mt
- #define iter_thread iter_thread_double_mt
- #define de_thread_helper de_thread_helper_64_mt
- #define de_thread de_thread_64_mt
- #include "rect.c"
-#else /* !HAVE_GCC_64BIT_ATOMIC_OPS */
- #define render_rectangle_double_mt render_rectangle_double
-#endif /* HAVE_GCC_64BIT_ATOMIC_OPS */
-#undef render_rectangle
-#undef iter_thread
-#undef add_c_to_accum
-#undef bucket
-#undef abucket
-#undef bump_no_overflow
-#undef abump_no_overflow
-#undef de_thread_helper
-#undef de_thread
-
-/* 32-bit datatypes */
-#define bucket bucket_int
-#define abucket abucket_int
-#define abump_no_overflow(dest, delta) do { \
- if (UINT_MAX - dest > delta) dest += delta; else dest = UINT_MAX; \
-} while (0)
-#define add_c_to_accum(acc,i,ii,j,jj,wid,hgt,c) do { \
- if ( (j) + (jj) >=0 && (j) + (jj) < (hgt) && (i) + (ii) >=0 && (i) + (ii) < (wid)) { \
- abucket *a = (acc) + ( (i) + (ii) ) + ( (j) + (jj) ) * (wid); \
- abump_no_overflow(a[0][0],(c)[0]); \
- abump_no_overflow(a[0][1],(c)[1]); \
- abump_no_overflow(a[0][2],(c)[2]); \
- abump_no_overflow(a[0][3],(c)[3]); \
- } \
-} while (0)
-/* single-threaded */
-#define USE_LOCKS
-#define bump_no_overflow(dest, delta) do { \
- if (UINT_MAX - dest > delta) dest += delta; else dest = UINT_MAX; \
-} while (0)
-#define render_rectangle render_rectangle_int
-#define iter_thread iter_thread_int
-#define de_thread_helper de_thread_helper_32
-#define de_thread de_thread_32
-#include "rect.c"
-#ifdef HAVE_GCC_ATOMIC_OPS
- /* multi-threaded */
- #undef USE_LOCKS
- #undef bump_no_overflow
- #undef render_rectangle
- #undef iter_thread
- #undef de_thread_helper
- #undef de_thread
- #define bump_no_overflow(dest, delta) uint_atomic_add(&dest, delta)
- #define render_rectangle render_rectangle_int_mt
- #define iter_thread iter_thread_int_mt
- #define de_thread_helper de_thread_helper_32_mt
- #define de_thread de_thread_32_mt
- #include "rect.c"
-#else /* !HAVE_GCC_ATOMIC_OPS */
- #define render_rectangle_int_mt render_rectangle_int
-#endif /* HAVE_GCC_ATOMIC_OPS */
-#undef iter_thread
-#undef render_rectangle
-#undef add_c_to_accum
-#undef bucket
-#undef abucket
-#undef bump_no_overflow
-#undef abump_no_overflow
-#undef de_thread_helper
-#undef de_thread
-
/* experimental 32-bit datatypes (called 33) */
#define bucket bucket_int
#define abucket abucket_float
@@ -3785,28 +3607,10 @@ ushort_atomic_add(unsigned short *dest, unsigned short delta)
#define bump_no_overflow(dest, delta) do { \
if (UINT_MAX - dest > delta) dest += delta; else dest = UINT_MAX; \
} while (0)
-#define render_rectangle render_rectangle_float
#define iter_thread iter_thread_float
#define de_thread_helper de_thread_helper_33
#define de_thread de_thread_33
#include "rect.c"
-#ifdef HAVE_GCC_ATOMIC_OPS
- /* multi-threaded */
- #undef USE_LOCKS
- #undef bump_no_overflow
- #undef render_rectangle
- #undef iter_thread
- #undef de_thread_helper
- #undef de_thread
- #define bump_no_overflow(dest, delta) uint_atomic_add(&dest, delta)
- #define render_rectangle render_rectangle_float_mt
- #define iter_thread iter_thread_float_mt
- #define de_thread_helper de_thread_helper_33_mt
- #define de_thread de_thread_33_mt
- #include "rect.c"
-#else /* !HAVE_GCC_ATOMIC_OPS */
- #define render_rectangle_float_mt render_rectangle_float
-#endif /* HAVE_GCC_ATOMIC_OPS */
#undef iter_thread
#undef render_rectangle
#undef add_c_to_accum
@@ -3843,39 +3647,8 @@ int flam3_render(flam3_frame *spec, void *out,
int retval;
- if (spec->nthreads <= 2) {
- /* single-threaded or 2 threads without atomic operations */
- switch (spec->bits) {
- case 32:
- retval = render_rectangle_int(spec, out, field, nchan, trans, stats);
- return(retval);
- case 33:
- retval = render_rectangle_float(spec, out, field, nchan, trans, stats);
- return(retval);
- case 64:
- retval = render_rectangle_double(spec, out, field, nchan, trans, stats);
- return(retval);
- default:
- bits_error(spec);
- return(1);
- }
- } else {
- /* 3+ threads using atomic ops if available */
- switch (spec->bits) {
- case 32:
- retval = render_rectangle_int_mt(spec, out, field, nchan, trans, stats);
- return(retval);
- case 33:
- retval = render_rectangle_float_mt(spec, out, field, nchan, trans, stats);
- return(retval);
- case 64:
- retval = render_rectangle_double_mt(spec, out, field, nchan, trans, stats);
- return(retval);
- default:
- bits_error(spec);
- return(1);
- }
- }
+ retval = render_rectangle (spec, out, field, nchan, trans, stats);
+ return(retval);
}
diff --git a/rect.c b/rect.c
index 30d76fd..0c99e39 100644
--- a/rect.c
+++ b/rect.c
@@ -441,31 +441,6 @@ static void iter_thread(void *fth) {
b = buckets + (int)(ficp->ws0 * p0 - ficp->wb0s0) +
ficp->width * (int)(ficp->hs1 * p1 - ficp->hb1s1);
-#ifdef USE_FLOAT_INDICES
- color_index0 = 0;
-
- //fprintf(stdout,"%.16f\n",p[2]*256.0);
-
- while(color_index0 < cmap_size_m1) {
- if (ficp->dmap[color_index0+1].index > p[2])
- break;
- else
- color_index0++;
- }
-
- if (p[3]==1.0) {
- bump_no_overflow(b[0][0], ficp->dmap[color_index0].color[0]);
- bump_no_overflow(b[0][1], ficp->dmap[color_index0].color[1]);
- bump_no_overflow(b[0][2], ficp->dmap[color_index0].color[2]);
- bump_no_overflow(b[0][3], ficp->dmap[color_index0].color[3]);
- bump_no_overflow(b[0][4], 255.0);
- } else {
- bump_no_overflow(b[0][0], logvis*ficp->dmap[color_index0].color[0]);
- bump_no_overflow(b[0][1], logvis*ficp->dmap[color_index0].color[1]);
- bump_no_overflow(b[0][2], logvis*ficp->dmap[color_index0].color[2]);
- bump_no_overflow(b[0][3], logvis*ficp->dmap[color_index0].color[3]);
- bump_no_overflow(b[0][4], logvis*255.0);
-#else
dbl_index0 = p[2] * cmap_size;
color_index0 = (int) (dbl_index0);
@@ -511,7 +486,6 @@ static void iter_thread(void *fth) {
bump_no_overflow(b[0][3], logvis*interpcolor[3]);
bump_no_overflow(b[0][4], logvis*255.0);
}
-#endif
}
}
diff --git a/wscript b/wscript
index 07c9deb..1f97094 100644
--- a/wscript
+++ b/wscript
@@ -14,8 +14,6 @@ def configure(conf):
conf.check_cc (lib='jpeg', uselib_store='jpeg')
conf.check_cfg (package='libpng', uselib_store='png', args=['--cflags', '--libs'], msg='Checking for library png')
conf.check_cc (lib='amdlibm', header_name='amdlibm.h', mandatory=False, define_name='HAVE_AMDLIBM', uselib_store='amdlibm')
- conf.check_cc (fragment='#include <stdint.h>\nint main() { uint32_t a = 4; __sync_bool_compare_and_swap(&a, 4, 5); }', define_name='HAVE_GCC_ATOMIC_OPS', msg='Checking for atomic CAS')
- conf.check_cc (fragment='#include <stdint.h>\nint main() { uint64_t a = 4; __sync_bool_compare_and_swap(&a, 4, 5); }', define_name='HAVE_GCC_64BIT_ATOMIC_OPS', msg='Checking for 64 bit atomic CAS')
# does not work
#conf.check_cc (function_name='__builtin_ia32_rdrand64_step', define_name='HAVE_RDRAND64')
--
cgit v1.2.3