[PATCH libdrm 04/10] util: add fp16 format support
Ilia Mirkin
imirkin at alum.mit.edu
Mon Jun 3 00:40:12 UTC 2019
This change adds support for all current patterns.
Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
---
tests/util/format.c | 5 ++
tests/util/pattern.c | 207 ++++++++++++++++++++++++++++++++++++++++++-
2 files changed, 209 insertions(+), 3 deletions(-)
diff --git a/tests/util/format.c b/tests/util/format.c
index e52213bb..1ca1b82c 100644
--- a/tests/util/format.c
+++ b/tests/util/format.c
@@ -93,6 +93,11 @@ static const struct util_format_info format_info[] = {
{ DRM_FORMAT_RGBX1010102, "RX30", MAKE_RGB_INFO(10, 22, 10, 12, 10, 2, 0, 0) },
{ DRM_FORMAT_BGRA1010102, "BA30", MAKE_RGB_INFO(10, 2, 10, 12, 10, 22, 2, 0) },
{ DRM_FORMAT_BGRX1010102, "BX30", MAKE_RGB_INFO(10, 2, 10, 12, 10, 22, 0, 0) },
+ { DRM_FORMAT_XRGB16161616F, "XR4H", MAKE_RGB_INFO(16, 32, 16, 16, 16, 0, 0, 0) },
+ { DRM_FORMAT_XBGR16161616F, "XB4H", MAKE_RGB_INFO(16, 0, 16, 16, 16, 32, 0, 0) },
+ { DRM_FORMAT_ARGB16161616F, "AR4H", MAKE_RGB_INFO(16, 32, 16, 16, 16, 0, 16, 48) },
+ { DRM_FORMAT_ABGR16161616F, "AB4H", MAKE_RGB_INFO(16, 0, 16, 16, 16, 32, 16, 48) },
+
};
uint32_t util_format_fourcc(const char *name)
diff --git a/tests/util/pattern.c b/tests/util/pattern.c
index ff110fd7..37796dbf 100644
--- a/tests/util/pattern.c
+++ b/tests/util/pattern.c
@@ -86,6 +86,17 @@ static inline uint32_t shiftcolor10(const struct util_color_component *comp,
return value << comp->offset;
}
+/* This function takes 16-bit color values */
+static inline uint64_t shiftcolor16(const struct util_color_component *comp,
+ uint64_t value)
+{
+ value &= 0xffff;
+ /* Shift down to remove unwanted low bits */
+ value = value >> (16 - comp->length);
+ /* Shift back up to where the value should be */
+ return value << comp->offset;
+}
+
#define MAKE_RGBA10(rgb, r, g, b, a) \
(shiftcolor10(&(rgb)->red, (r)) | \
shiftcolor10(&(rgb)->green, (g)) | \
@@ -101,6 +112,49 @@ static inline uint32_t shiftcolor10(const struct util_color_component *comp,
#define MAKE_RGB24(rgb, r, g, b) \
{ .value = MAKE_RGBA(rgb, r, g, b, 0) }
+
+/**
+ * Takes a uint16_t, divides by 65536, converts the infinite-precision
+ * result to fp16 with round-to-zero.
+ *
+ * Copied from mesa:src/util/half_float.c
+ */
+static uint16_t uint16_div_64k_to_half(uint16_t v)
+{
+ /* Zero or subnormal. Set the mantissa to (v << 8) and return. */
+ if (v < 4)
+ return v << 8;
+
+ /* Count the leading 0s in the uint16_t */
+ int n = __builtin_clz(v) - 16;
+
+ /* Shift the mantissa up so bit 16 is the hidden 1 bit,
+ * mask it off, then shift back down to 10 bits
+ */
+ int m = ( ((uint32_t)v << (n + 1)) & 0xffff ) >> 6;
+
+ /* (0{n} 1 X{15-n}) * 2^-16
+ * = 1.X * 2^(15-n-16)
+ * = 1.X * 2^(14-n - 15)
+ * which is the FP16 form with e = 14 - n
+ */
+ int e = 14 - n;
+
+ return (e << 10) | m;
+}
+
+#define MAKE_RGBA8FP16(rgb, r, g, b, a) \
+ (shiftcolor16(&(rgb)->red, uint16_div_64k_to_half((r) << 8)) | \
+ shiftcolor16(&(rgb)->green, uint16_div_64k_to_half((g) << 8)) | \
+ shiftcolor16(&(rgb)->blue, uint16_div_64k_to_half((b) << 8)) | \
+ shiftcolor16(&(rgb)->alpha, uint16_div_64k_to_half((a) << 8)))
+
+#define MAKE_RGBA10FP16(rgb, r, g, b, a) \
+ (shiftcolor16(&(rgb)->red, uint16_div_64k_to_half((r) << 6)) | \
+ shiftcolor16(&(rgb)->green, uint16_div_64k_to_half((g) << 6)) | \
+ shiftcolor16(&(rgb)->blue, uint16_div_64k_to_half((b) << 6)) | \
+ shiftcolor16(&(rgb)->alpha, uint16_div_64k_to_half((a) << 6)))
+
static void fill_smpte_yuv_planar(const struct util_yuv_info *yuv,
unsigned char *y_mem, unsigned char *u_mem,
unsigned char *v_mem, unsigned int width,
@@ -489,6 +543,67 @@ static void fill_smpte_rgb32(const struct util_rgb_info *rgb, void *mem,
}
}
+static void fill_smpte_rgb16fp(const struct util_rgb_info *rgb, void *mem,
+ unsigned int width, unsigned int height,
+ unsigned int stride)
+{
+ const uint64_t colors_top[] = {
+ MAKE_RGBA8FP16(rgb, 192, 192, 192, 255),/* grey */
+ MAKE_RGBA8FP16(rgb, 192, 192, 0, 255), /* yellow */
+ MAKE_RGBA8FP16(rgb, 0, 192, 192, 255), /* cyan */
+ MAKE_RGBA8FP16(rgb, 0, 192, 0, 255), /* green */
+ MAKE_RGBA8FP16(rgb, 192, 0, 192, 255), /* magenta */
+ MAKE_RGBA8FP16(rgb, 192, 0, 0, 255), /* red */
+ MAKE_RGBA8FP16(rgb, 0, 0, 192, 255), /* blue */
+ };
+ const uint64_t colors_middle[] = {
+ MAKE_RGBA8FP16(rgb, 0, 0, 192, 127), /* blue */
+ MAKE_RGBA8FP16(rgb, 19, 19, 19, 127), /* black */
+ MAKE_RGBA8FP16(rgb, 192, 0, 192, 127), /* magenta */
+ MAKE_RGBA8FP16(rgb, 19, 19, 19, 127), /* black */
+ MAKE_RGBA8FP16(rgb, 0, 192, 192, 127), /* cyan */
+ MAKE_RGBA8FP16(rgb, 19, 19, 19, 127), /* black */
+ MAKE_RGBA8FP16(rgb, 192, 192, 192, 127),/* grey */
+ };
+ const uint64_t colors_bottom[] = {
+ MAKE_RGBA8FP16(rgb, 0, 33, 76, 255), /* in-phase */
+ MAKE_RGBA8FP16(rgb, 255, 255, 255, 255),/* super white */
+ MAKE_RGBA8FP16(rgb, 50, 0, 106, 255), /* quadrature */
+ MAKE_RGBA8FP16(rgb, 19, 19, 19, 255), /* black */
+ MAKE_RGBA8FP16(rgb, 9, 9, 9, 255), /* 3.5% */
+ MAKE_RGBA8FP16(rgb, 19, 19, 19, 255), /* 7.5% */
+ MAKE_RGBA8FP16(rgb, 29, 29, 29, 255), /* 11.5% */
+ MAKE_RGBA8FP16(rgb, 19, 19, 19, 255), /* black */
+ };
+ unsigned int x;
+ unsigned int y;
+
+ for (y = 0; y < height * 6 / 9; ++y) {
+ for (x = 0; x < width; ++x)
+ ((uint64_t *)mem)[x] = colors_top[x * 7 / width];
+ mem += stride;
+ }
+
+ for (; y < height * 7 / 9; ++y) {
+ for (x = 0; x < width; ++x)
+ ((uint64_t *)mem)[x] = colors_middle[x * 7 / width];
+ mem += stride;
+ }
+
+ for (; y < height; ++y) {
+ for (x = 0; x < width * 5 / 7; ++x)
+ ((uint64_t *)mem)[x] =
+ colors_bottom[x * 4 / (width * 5 / 7)];
+ for (; x < width * 6 / 7; ++x)
+ ((uint64_t *)mem)[x] =
+ colors_bottom[(x - width * 5 / 7) * 3
+ / (width / 7) + 4];
+ for (; x < width; ++x)
+ ((uint64_t *)mem)[x] = colors_bottom[7];
+ mem += stride;
+ }
+}
+
static void fill_smpte_c8(void *mem, unsigned int width, unsigned int height,
unsigned int stride)
{
@@ -638,6 +753,13 @@ static void fill_smpte(const struct util_format_info *info, void *planes[3],
case DRM_FORMAT_BGRX1010102:
return fill_smpte_rgb32(&info->rgb, planes[0],
width, height, stride);
+
+ case DRM_FORMAT_XRGB16161616F:
+ case DRM_FORMAT_XBGR16161616F:
+ case DRM_FORMAT_ARGB16161616F:
+ case DRM_FORMAT_ABGR16161616F:
+ return fill_smpte_rgb16fp(&info->rgb, planes[0],
+ width, height, stride);
}
}
@@ -849,6 +971,32 @@ static void fill_tiles_rgb32(const struct util_format_info *info, void *mem,
make_pwetty(mem_base, width, height, stride, info->format);
}
+static void fill_tiles_rgb16fp(const struct util_format_info *info, void *mem,
+ unsigned int width, unsigned int height,
+ unsigned int stride)
+{
+ const struct util_rgb_info *rgb = &info->rgb;
+ void *mem_base = mem;
+ unsigned int x, y;
+
+ /* TODO: Give this actual fp16 precision */
+ for (y = 0; y < height; ++y) {
+ for (x = 0; x < width; ++x) {
+ div_t d = div(x+y, width);
+ uint32_t rgb32 = 0x00130502 * (d.quot >> 6)
+ + 0x000a1120 * (d.rem >> 6);
+ uint32_t alpha = ((y < height/2) && (x < width/2)) ? 127 : 255;
+ uint64_t color =
+ MAKE_RGBA8FP16(rgb, (rgb32 >> 16) & 0xff,
+ (rgb32 >> 8) & 0xff, rgb32 & 0xff,
+ alpha);
+
+ ((uint64_t *)mem)[x] = color;
+ }
+ mem += stride;
+ }
+}
+
static void fill_tiles(const struct util_format_info *info, void *planes[3],
unsigned int width, unsigned int height,
unsigned int stride)
@@ -923,14 +1071,32 @@ static void fill_tiles(const struct util_format_info *info, void *planes[3],
case DRM_FORMAT_BGRX1010102:
return fill_tiles_rgb32(info, planes[0],
width, height, stride);
+
+ case DRM_FORMAT_XRGB16161616F:
+ case DRM_FORMAT_XBGR16161616F:
+ case DRM_FORMAT_ARGB16161616F:
+ case DRM_FORMAT_ABGR16161616F:
+ return fill_tiles_rgb16fp(info, planes[0],
+ width, height, stride);
}
}
-static void fill_plain(void *planes[3],
+static void fill_plain(const struct util_format_info *info, void *planes[3],
unsigned int height,
unsigned int stride)
{
- memset(planes[0], 0x77, stride * height);
+ switch (info->format) {
+ case DRM_FORMAT_XRGB16161616F:
+ case DRM_FORMAT_XBGR16161616F:
+ case DRM_FORMAT_ARGB16161616F:
+ case DRM_FORMAT_ABGR16161616F:
+ /* 0x3838 = 0.5273 */
+ memset(planes[0], 0x38, stride * height);
+ break;
+ default:
+ memset(planes[0], 0x77, stride * height);
+ break;
+ }
}
static void fill_gradient_rgb32(const struct util_rgb_info *rgb,
@@ -961,6 +1127,34 @@ static void fill_gradient_rgb32(const struct util_rgb_info *rgb,
}
}
+static void fill_gradient_rgb16fp(const struct util_rgb_info *rgb,
+ void *mem,
+ unsigned int width, unsigned int height,
+ unsigned int stride)
+{
+ int i, j;
+
+ for (i = 0; i < height / 2; i++) {
+ uint64_t *row = mem;
+
+ for (j = 0; j < width / 2; j++) {
+ uint64_t value = MAKE_RGBA10FP16(rgb, j & 0x3ff, j & 0x3ff, j & 0x3ff, 0);
+ row[2*j] = row[2*j+1] = value;
+ }
+ mem += stride;
+ }
+
+ for (; i < height; i++) {
+ uint64_t *row = mem;
+
+ for (j = 0; j < width / 2; j++) {
+ uint64_t value = MAKE_RGBA10FP16(rgb, j & 0x3fc, j & 0x3fc, j & 0x3fc, 0);
+ row[2*j] = row[2*j+1] = value;
+ }
+ mem += stride;
+ }
+}
+
/* The gradient pattern creates two horizontal gray gradients, split
* into two halves. The top half has 10bpc precision, the bottom half
* has 8bpc precision. When using with a 10bpc fb format, there are 3
@@ -1009,6 +1203,13 @@ static void fill_gradient(const struct util_format_info *info, void *planes[3],
case DRM_FORMAT_BGRX1010102:
return fill_gradient_rgb32(&info->rgb, planes[0],
width, height, stride);
+
+ case DRM_FORMAT_XRGB16161616F:
+ case DRM_FORMAT_XBGR16161616F:
+ case DRM_FORMAT_ARGB16161616F:
+ case DRM_FORMAT_ABGR16161616F:
+ return fill_gradient_rgb16fp(&info->rgb, planes[0],
+ width, height, stride);
}
}
@@ -1042,7 +1243,7 @@ void util_fill_pattern(uint32_t format, enum util_fill_pattern pattern,
return fill_smpte(info, planes, width, height, stride);
case UTIL_PATTERN_PLAIN:
- return fill_plain(planes, height, stride);
+ return fill_plain(info, planes, height, stride);
case UTIL_PATTERN_GRADIENT:
return fill_gradient(info, planes, width, height, stride);
--
2.21.0
More information about the dri-devel
mailing list