Skip to content

tiffload: add 16-bit float support #3626

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Sep 21, 2023
Merged

tiffload: add 16-bit float support #3626

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
20c2bcf
tiffload: add 16-bit float support
DarthSim Aug 25, 2023
d7ccee5
Merge branch 'master' into feature/tiff-16-bit-float
kleisauke Sep 21, 2023
067131f
Appease clang-format
kleisauke Sep 21, 2023
b7e1094
Add ChangeLog note
kleisauke Sep 21, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions ChangeLog
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@
- remove libgsf dependency in favor of libarchive [kleisauke]
- better chunking for small shrinks [jcupitt]
- use alpha range of 0.0 - 1.0 for scRGB images [DarthSim]
- add support for 16-bit float TIFFs [DarthSim]

18/9/23 8.14.5

Expand Down
98 changes: 91 additions & 7 deletions libvips/foreign/tiff2vips.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -425,6 +425,27 @@ typedef struct _Rtiff {
int y_pos;
} Rtiff;

/* Convert IEEE 754-2008 16-bit float to 32-bit float
*/
static inline float
half_2_float(gushort h)
{
const float sign = (h >> 15) * -2 + 1;
const int exp = ((h & 0x7C00) >> 10) - 15;
const float prec = (h & 0x03FF);

switch (exp) {
case 16:
return INFINITY * sign;
case -15:
return sign / (float) (1 << 14) * (prec / 1024.0);
default:
return exp > 0
? sign * (float) (1 << exp) * (1.0 + prec / 1024.0)
: sign / (float) (1 << -exp) * (1.0 + prec / 1024.0);
}
}

/* Test for field exists.
*/
static int
Expand Down Expand Up @@ -869,6 +890,8 @@ rtiff_guess_format(Rtiff *rtiff)
return VIPS_FORMAT_SHORT;
if (sample_format == SAMPLEFORMAT_UINT)
return VIPS_FORMAT_USHORT;
if (sample_format == SAMPLEFORMAT_IEEEFP)
return VIPS_FORMAT_FLOAT;
break;

case 32:
Expand Down Expand Up @@ -1229,13 +1252,37 @@ rtiff_parse_fourbit(Rtiff *rtiff, VipsImage *out)
} \
}

/* GREY_LOOP implementation for 16-bit float
*/
#define GREY_LOOP_F16 \
{ \
gushort *p1; \
float *q1; \
\
p1 = (gushort *) p; \
q1 = (float *) q; \
for (x = 0; x < n; x++) { \
if (invert) \
q1[0] = 1.0 - half_2_float(p1[0]); \
else \
q1[0] = half_2_float(p1[0]); \
\
for (i = 1; i < samples_per_pixel; i++) \
q1[i] = half_2_float(p1[i]); \
\
q1 += samples_per_pixel; \
p1 += samples_per_pixel; \
} \
}

/* Per-scanline process function for greyscale images.
*/
static void
rtiff_greyscale_line(Rtiff *rtiff,
VipsPel *q, VipsPel *p, int n, void *client)
{
int samples_per_pixel = rtiff->header.samples_per_pixel;
int bits_per_sample = rtiff->header.bits_per_sample;
int photometric_interpretation =
rtiff->header.photometric_interpretation;
VipsBandFormat format = rtiff_guess_format(rtiff);
Expand Down Expand Up @@ -1274,7 +1321,12 @@ rtiff_greyscale_line(Rtiff *rtiff,
break;

case VIPS_FORMAT_FLOAT:
GREY_LOOP(float, 1.0);
if (bits_per_sample == 16) {
GREY_LOOP_F16;
}
else {
GREY_LOOP(float, 1.0);
}
break;

case VIPS_FORMAT_DOUBLE:
Expand Down Expand Up @@ -1565,6 +1617,27 @@ rtiff_memcpy_line(Rtiff *rtiff, VipsPel *q, VipsPel *p, int n, void *client)
memcpy(q, p, len);
}

/* Per-scanline process function when we just need to copy.
*/
static void
rtiff_memcpy_f16_line(Rtiff *rtiff, VipsPel *q, VipsPel *p, int n, void *client)
{
VipsImage *im = (VipsImage *) client;
size_t len = n * im->Bands;

if (im->BandFmt == VIPS_FORMAT_COMPLEX ||
im->BandFmt == VIPS_FORMAT_DPCOMPLEX)
len *= 2;

int i;

gushort *restrict hp = (gushort *) p;
float *restrict fq = (float *) q;

for (i = 0; i < len; i++)
fq[i] = half_2_float(hp[i]);
}

/* Read a regular multiband image where we can just copy pixels from the tiff
* buffer.
*/
Expand All @@ -1574,6 +1647,8 @@ rtiff_parse_copy(Rtiff *rtiff, VipsImage *out)
int samples_per_pixel = rtiff->header.samples_per_pixel;
int photometric_interpretation =
rtiff->header.photometric_interpretation;
int bits_per_sample = rtiff->header.bits_per_sample;
int sample_format = rtiff->header.sample_format;
int inkset = rtiff->header.inkset;

if (rtiff_non_fractional(rtiff))
Expand Down Expand Up @@ -1608,15 +1683,20 @@ rtiff_parse_copy(Rtiff *rtiff, VipsImage *out)
else
out->Type = VIPS_INTERPRETATION_MULTIBAND;

rtiff->sfn = rtiff_memcpy_line;
rtiff->client = out;

/* We expand YCBCR images to RGB using JPEGCOLORMODE_RGB, and this
* means we need a slightly larger read buffer for the edge pixels. In
* turn, this means we can't just memcpy to libvips regions.
*/
rtiff->memcpy = photometric_interpretation != PHOTOMETRIC_YCBCR;
if (bits_per_sample == 16 && sample_format == SAMPLEFORMAT_IEEEFP) {
rtiff->sfn = rtiff_memcpy_f16_line;
}
else {
rtiff->sfn = rtiff_memcpy_line;

/* We expand YCBCR images to RGB using JPEGCOLORMODE_RGB, and this
* means we need a slightly larger read buffer for the edge pixels. In
* turn, this means we can't just memcpy to libvips regions.
*/
rtiff->memcpy = photometric_interpretation != PHOTOMETRIC_YCBCR;
}
return 0;
}

Expand Down Expand Up @@ -2695,6 +2775,10 @@ rtiff_read_stripwise(Rtiff *rtiff, VipsImage *out)
else
vips_line_size = VIPS_IMAGE_SIZEOF_LINE(t[0]);

if (rtiff->header.bits_per_sample == 16 &&
rtiff->header.sample_format == SAMPLEFORMAT_IEEEFP)
vips_line_size /= 2;

if (vips_line_size != rtiff->header.scanline_size) {
vips_error("tiff2vips",
"%s", _("unsupported tiff image type"));
Expand Down