Skip to content

Commit

Permalink
workaround "Too many args to microtask" error on freebsd-arm64
Browse files Browse the repository at this point in the history 
http://www.cpantesters.org/cpan/report/2ec61c60-b205-11ee-b3df-e9156e8775ea

it also doesn't seem to respect #omp default(shared), so reduce
the number of arguments passed
  • Loading branch information
@dk
dk committed Jan 14, 2024
1 parent da36741 commit 1f5a328
Showing 1 changed file with 77 additions and 44 deletions.
121 changes: 77 additions & 44 deletions img/rotate.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -432,39 +432,47 @@ create_tmp_image( PImage template, int channels, PImage target, Point size, Colo
return true;
}

typedef struct {
int channels;
int src_w;
int dst_w;
float *fill;
Bool reverse;
} ShearXCommonRec;

#define SHEAR_X_FUNCTION(type,pixel_interim_type) \
static void \
shear_x_scanline_ ## type( \
void * _src, int channels, int src_w, \
void * _dst, int dst_w, \
int delta, float sf, float * fill, Bool reverse \
ShearXCommonRec *ctx, \
void * _src, void * _dst, int delta, float sf \
) { \
int x, c, nx, dsrc; \
float leftover[3]; \
const int channels = (const int)ctx->channels; \
type *src = (type*)_src, *dst = (type*) _dst; \
pixel_interim_type new_pixel; \
\
if ( reverse ) { \
if ( ctx->reverse ) { \
dsrc = channels * 2; \
src += (src_w - 1) * channels; \
src += (ctx->src_w - 1) * channels; \
} else \
dsrc = 0; \
\
for ( c = 0; c < channels; c++) \
leftover[c] = fill[c] * (1.0 - sf); \
leftover[c] = ctx->fill[c] * (1.0 - sf); \
\
dst += delta * channels; \
for ( \
x = 0, nx = delta; \
x < src_w; \
x < ctx->src_w; \
x++, nx++, src -= dsrc \
) { \
for ( c = 0; c < channels; c++, src++) { \
float to_transfer = ((float)*src) * sf;

#define SHEAR_X_LOOP \
float to_leave = *src - new_pixel + leftover[c]; \
if ( nx >= dst_w ) return;
if ( nx >= ctx->dst_w ) return;

#define CLAMP(minval,maxval) \
if ( new_pixel < minval ) new_pixel = minval; \
Expand All @@ -479,7 +487,7 @@ shear_x_scanline_ ## type( \
} \
}
#define SHEAR_X_TAIL \
if ( nx >= 0 && nx < dst_w) \
if ( nx >= 0 && nx < ctx->dst_w) \
for ( c = 0; c < channels; c++) {

#define SHEAR_X_FUNCTION_END \
Expand All @@ -493,7 +501,7 @@ SHEAR_X_FUNCTION(Byte,short int)
CLAMP(0,255)
SHEAR_X_LOOP_END
SHEAR_X_TAIL
new_pixel = leftover[c] + fill[c] * sf + 0.5;
new_pixel = leftover[c] + ctx->fill[c] * sf + 0.5;
CLAMP(0,255)
SHEAR_X_FUNCTION_END

Expand All @@ -503,7 +511,7 @@ SHEAR_X_FUNCTION(Short,int)
CLAMP(INT16_MIN,INT16_MAX)
SHEAR_X_LOOP_END
SHEAR_X_TAIL
new_pixel = leftover[c] + fill[c] * sf + 0.5;
new_pixel = leftover[c] + ctx->fill[c] * sf + 0.5;
CLAMP(INT16_MIN,INT16_MAX)
SHEAR_X_FUNCTION_END

Expand All @@ -513,7 +521,7 @@ SHEAR_X_FUNCTION(Long,int64_t)
CLAMP(INT32_MIN,INT32_MAX)
SHEAR_X_LOOP_END
SHEAR_X_TAIL
new_pixel = leftover[c] + fill[c] * sf + 0.5;
new_pixel = leftover[c] + ctx->fill[c] * sf + 0.5;
CLAMP(INT32_MIN,INT32_MAX)
SHEAR_X_FUNCTION_END

Expand All @@ -522,53 +530,64 @@ SHEAR_X_FUNCTION(float,float)
SHEAR_X_LOOP
SHEAR_X_LOOP_END
SHEAR_X_TAIL
new_pixel = leftover[c] + fill[c] * sf ;
new_pixel = leftover[c] + ctx->fill[c] * sf ;
SHEAR_X_FUNCTION_END

SHEAR_X_FUNCTION(double,double)
new_pixel = leftover[c] + to_transfer;
SHEAR_X_LOOP
SHEAR_X_LOOP_END
SHEAR_X_TAIL
new_pixel = leftover[c] + fill[c] * sf;
new_pixel = leftover[c] + ctx->fill[c] * sf;
SHEAR_X_FUNCTION_END

typedef struct {
int channels;
int src_w;
int src_h;
int src_stride;
int dst_w;
int dst_h;
int dst_stride;
float *fill;
} ShearYCommonRec;

#define SHEAR_Y_FUNCTION(type,pixel_interim_type) \
static void \
shear_y_scanline_ ## type( \
void * _src, int channels, int src_w, int src_h, int src_stride, \
void * _dst, int dst_w, int dst_h, int dst_stride, \
int delta, float sf, float * fill \
ShearYCommonRec *ctx, \
void * _src, void * _dst, int delta, float sf \
) { \
int y, c, ny; \
float leftover[3]; \
const int channels = (const int)ctx->channels; \
type * src = (type*)_src, * dst = ( type* ) _dst; \
pixel_interim_type new_pixel; \
\
for ( c = 0; c < channels; c++) \
leftover[c] = fill[c] * ( 1 - sf ); \
leftover[c] = ctx->fill[c] * ( 1 - sf ); \
\
dst = (type*) (((Byte*) dst) + delta * dst_stride); \
dst = (type*) (((Byte*) dst) + delta * ctx->dst_stride); \
\
for ( y = 0, ny = delta; y < src_h; y++, ny++) { \
for ( y = 0, ny = delta; y < ctx->src_h; y++, ny++) { \
for ( c = 0; c < channels; c++) { \
type pixel = src[c]; \
float to_transfer = ((float)pixel) * sf;

#define SHEAR_Y_LOOP \
float to_leave = pixel - new_pixel + leftover[c];\
if ( ny >= dst_h) return;
if ( ny >= ctx->dst_h) return;

#define SHEAR_Y_LOOP_END(type) \
if ( ny >= 0 ) dst[c] = new_pixel; \
leftover[c] = to_leave; \
} \
src = (type*)(((Byte*)src) + src_stride); \
dst = (type*)(((Byte*)dst) + dst_stride); \
src = (type*)(((Byte*)src) + ctx->src_stride); \
dst = (type*)(((Byte*)dst) + ctx->dst_stride); \
}

#define SHEAR_Y_TAIL \
if ( ny >= 0 && ny < dst_h) \
if ( ny >= 0 && ny < ctx->dst_h) \
for ( c = 0; c < channels; c++) {

#define SHEAR_Y_FUNCTION_END \
Expand All @@ -582,7 +601,7 @@ SHEAR_Y_FUNCTION(Byte,short int)
CLAMP(0,255)
SHEAR_Y_LOOP_END(Byte)
SHEAR_Y_TAIL
new_pixel = leftover[c] + fill[c] * sf + 0.5;
new_pixel = leftover[c] + ctx->fill[c] * sf + 0.5;
CLAMP(0,255)
SHEAR_Y_FUNCTION_END

Expand All @@ -592,7 +611,7 @@ SHEAR_Y_FUNCTION(Short,int)
CLAMP(INT16_MIN,INT16_MAX)
SHEAR_Y_LOOP_END(Short)
SHEAR_Y_TAIL
new_pixel = leftover[c] + fill[c] * sf + 0.5;
new_pixel = leftover[c] + ctx->fill[c] * sf + 0.5;
CLAMP(INT16_MIN,INT16_MAX)
SHEAR_Y_FUNCTION_END

Expand All @@ -602,7 +621,7 @@ SHEAR_Y_FUNCTION(Long,int64_t)
CLAMP(INT32_MIN,INT32_MAX)
SHEAR_Y_LOOP_END(Long)
SHEAR_Y_TAIL
new_pixel = leftover[c] + fill[c] * sf + 0.5;
new_pixel = leftover[c] + ctx->fill[c] * sf + 0.5;
CLAMP(INT32_MIN,INT32_MAX)
SHEAR_Y_FUNCTION_END

Expand All @@ -611,27 +630,25 @@ SHEAR_Y_FUNCTION(float,float)
SHEAR_Y_LOOP
SHEAR_Y_LOOP_END(float)
SHEAR_Y_TAIL
new_pixel = leftover[c] + fill[c] * sf ;
new_pixel = leftover[c] + ctx->fill[c] * sf ;
SHEAR_Y_FUNCTION_END

SHEAR_Y_FUNCTION(double,double)
new_pixel = leftover[c] + to_transfer;
SHEAR_Y_LOOP
SHEAR_Y_LOOP_END(double)
SHEAR_Y_TAIL
new_pixel = leftover[c] + fill[c] * sf;
new_pixel = leftover[c] + ctx->fill[c] * sf;
SHEAR_Y_FUNCTION_END

typedef void ShearXFunc(
void * src, int channels, int src_w,
void * dst, int dst_w,
int delta, float sf, float * fill, Bool reverse);
ShearXCommonRec *ctx,
void * src, void * dst,
int delta, float sf);

typedef void ShearYFunc(
void * _src, int channels, int src_w, int src_h, int src_stride,
void * _dst, int dst_w, int dst_h, int dst_stride,
int delta, float sf, float * fill
);
ShearYCommonRec *ctx,
void * _src, void * _dst, int delta, float sf);

#define FIND_SHEAR_FUNC(letter) \
case imByte : shear_func = shear_## letter ##_scanline_Byte; break;\
Expand All @@ -646,6 +663,7 @@ shear_x( PImage src, int channels, PImage dst, float func_mul, float func_add, F
int w, dw, h, y, lim_y, dsrc, ddst;
Byte * src_scanline, * dst_scanline;
ShearXFunc * shear_func;
ShearXCommonRec ctx;

w = src->w / channels;
dw = dst->w / channels;
Expand All @@ -668,16 +686,21 @@ shear_x( PImage src, int channels, PImage dst, float func_mul, float func_add, F
}

lim_y = ( h < dst->h ) ? h : dst->h;
ctx.channels = channels;
ctx.src_w = w;
ctx.dst_w = dw;
ctx.fill = fill;
ctx.reverse = apply_180;
#ifdef HAVE_OPENMP
#pragma omp parallel for
#endif
for ( y = 0; y < lim_y; y++) {
float sk = func_add + (( func_mul > 0 ) ? func_mul * y : (-func_mul) * (h - y - 1));
int si = (int) floorf(sk);
shear_func(
src_scanline + dsrc * y, channels, w,
dst_scanline + ddst * y, dw,
si + dx, filter(sk - si), fill, apply_180
shear_func( &ctx,
src_scanline + dsrc * y,
dst_scanline + ddst * y,
si + dx, filter(sk - si)
);
}
}
Expand All @@ -688,6 +711,7 @@ shear_y( PImage src, int channels, PImage dst, float func_mul, float func_add, F
int x, w, h, dw, lim_x, dscanline;
Byte * src_scanline, * dst_scanline;
ShearYFunc * shear_func;
ShearYCommonRec ctx;

w = src->w / channels;
dw = dst->w / channels;
Expand All @@ -704,16 +728,25 @@ shear_y( PImage src, int channels, PImage dst, float func_mul, float func_add, F
}

lim_x = ( w < dst-> w ) ? w : dst->w;
ctx.channels = channels;
ctx.src_w = w;
ctx.src_h = h;
ctx.src_stride = src->lineSize;
ctx.dst_w = dw;
ctx.dst_h = dst->h;
ctx.dst_stride = dst->lineSize;
ctx.fill = fill;
#ifdef HAVE_OPENMP
#pragma omp parallel for
#endif
for ( x = 0; x < lim_x; x++) {
float sk = func_add + (( func_mul > 0 ) ? func_mul * x : (-func_mul) * (w - x - 1));
int si = (int) floorf(sk);
shear_func(
src_scanline + dscanline * x, channels, w, h, src->lineSize,
dst_scanline + dscanline * x, dw, dst->h, dst->lineSize,
si + dy, filter(sk - si), fill);
shear_func( &ctx,
src_scanline + dscanline * x,
dst_scanline + dscanline * x,
si + dy, filter(sk - si)
);
}
}

Expand Down

0 comments on commit 1f5a328

Please sign in to comment.