Skip to contentSkip to conversation

drm/format-helper: Add helpers for line conversion

PATCH New Thomas Zimmermann 1 participant 9 messages Mar 28, 2025

Subsystems (3)
DRM drivers DRM misc DRM panic
lore
Share & tools
lore

Thomas Zimmermann

author maintainer
#0 lore raw

v2 cover drm/format-helper: Add helpers for line conversion

Add helpers for converting the pixel format of scanline. This used to be implemented for each format individually, but only the per-pixel code is really different among formats.

Patch 1 moves the per-pixel format helpers from drm_draw.c to an internal header. These functions have equivalents in the format-helper code. Just update the interface to make them exportable and add some comments.

Patches 2 to 5 add generic line-conversion helpers. Each takes a scanline with pixels with a specific number of bits per pixel and converts it to a scanline with a specific number of bits per pixel. Conversion of each individual pixel is done by the provided per-pixel helper that was extracted from drm_draw.c.

Patches 6 to 8 slightly optimize the line-conversion helpers by storing multiple pixels at once. With the current code, there are too many functions to make opimization feasible. But with the new helpers, there are only a few places to optimize to benefit all use cases.

Here's a little benchmark for the common use case of watching a full-screen video under X. The tests measures the average time to display a single frame. It uses an XRGB8888 framebuffer for each frame, but displays it to an RGB565 scanout buffer, thus doing format conversion with the _32to16 helper. The test system is an Intel Core i5-3470 with an Intel HD2500 graphics card. The graphics driver is simpledrm at 1024x768 pixels.

- 32-bit memcpy: 510 µs/frame
- Current code: 1640 µs/frame
- New (unoptimized): 1600 µs/frame
- New (no 64-bit): 1580 µs/frame
- New (optimized): 1470 µs/frame

The first line shows the time to memcpy a single frame to video memory without format conversion. With conversion, the old and new code takes roughly 3 times as long. Both variants are comparable in performance. With 64-bit stores added, the new code is even 10% faster than the current one. (This is not to be considered a full performance test. The main objective is to see whether the new code is on par with the old code, which is the case.)

v2:
- simplify address calculation (Jani)
- fix typos (Jocelyn)

Thomas Zimmermann (8):
drm/format-helper: Move helpers for pixel conversion to header file drm/format-helper: Add generic conversion to 32-bit formats drm/format-helper: Add generic conversion to 24-bit formats drm/format-helper: Add generic conversion to 16-bit formats drm/format-helper: Add generic conversion to 8-bit formats drm/format-helper: Optimize 32-to-24-bpp conversion drm/format-helper: Optimize 32-to-16-bpp conversion drm/format-helper: Optimize 32-to-8-bpp conversion

 drivers/gpu/drm/drm_draw.c            | 100 +------
 drivers/gpu/drm/drm_format_helper.c   | 374 +++++++++++---------------
 drivers/gpu/drm/drm_format_internal.h | 160 +++++++++++
 3 files changed, 335 insertions(+), 299 deletions(-)
 create mode 100644 drivers/gpu/drm/drm_format_internal.h
Signature · 1 line
2.48.1
author maintainer
#1 lore raw

PATCH v2 1/8 drm/format-helper: Move helpers for pixel conversion to header file

The DRM draw helpers contain format-conversion helpers that operate on individual pixels. Move them into an internal header file and adopt them as individual API. Update the draw code accordingly. The pixel helpers will also be useful for other format conversion helpers.

Signed-off-by Thomas Zimmermann <[email protected]>
Reviewed-by Jocelyn Falempe <[email protected]>
---
 drivers/gpu/drm/drm_draw.c            | 100 +++-------------------
 drivers/gpu/drm/drm_format_internal.h | 119 ++++++++++++++++++++++++++
 2 files changed, 130 insertions(+), 89 deletions(-)
 create mode 100644 drivers/gpu/drm/drm_format_internal.h
drivers/gpu/drm/drm_draw.c +11 -89
index cb2ad12bce57f..d41f8ae1c1483 100644
@@ -11,85 +11,7 @@
 #include <drm/drm_fourcc.h>
 
 #include "drm_draw_internal.h"
-
-/*
- * Conversions from xrgb8888
- */
-
-static u16 convert_xrgb8888_to_rgb565(u32 pix)
-{
-	return ((pix & 0x00F80000) >> 8) |
-	       ((pix & 0x0000FC00) >> 5) |
-	       ((pix & 0x000000F8) >> 3);
-}
-
-static u16 convert_xrgb8888_to_rgba5551(u32 pix)
-{
-	return ((pix & 0x00f80000) >> 8) |
-	       ((pix & 0x0000f800) >> 5) |
-	       ((pix & 0x000000f8) >> 2) |
-	       BIT(0); /* set alpha bit */
-}
-
-static u16 convert_xrgb8888_to_xrgb1555(u32 pix)
-{
-	return ((pix & 0x00f80000) >> 9) |
-	       ((pix & 0x0000f800) >> 6) |
-	       ((pix & 0x000000f8) >> 3);
-}
-
-static u16 convert_xrgb8888_to_argb1555(u32 pix)
-{
-	return BIT(15) | /* set alpha bit */
-	       ((pix & 0x00f80000) >> 9) |
-	       ((pix & 0x0000f800) >> 6) |
-	       ((pix & 0x000000f8) >> 3);
-}
-
-static u32 convert_xrgb8888_to_argb8888(u32 pix)
-{
-	return pix | GENMASK(31, 24); /* fill alpha bits */
-}
-
-static u32 convert_xrgb8888_to_xbgr8888(u32 pix)
-{
-	return ((pix & 0x00ff0000) >> 16) <<  0 |
-	       ((pix & 0x0000ff00) >>  8) <<  8 |
-	       ((pix & 0x000000ff) >>  0) << 16 |
-	       ((pix & 0xff000000) >> 24) << 24;
-}
-
-static u32 convert_xrgb8888_to_abgr8888(u32 pix)
-{
-	return ((pix & 0x00ff0000) >> 16) <<  0 |
-	       ((pix & 0x0000ff00) >>  8) <<  8 |
-	       ((pix & 0x000000ff) >>  0) << 16 |
-	       GENMASK(31, 24); /* fill alpha bits */
-}
-
-static u32 convert_xrgb8888_to_xrgb2101010(u32 pix)
-{
-	pix = ((pix & 0x000000FF) << 2) |
-	      ((pix & 0x0000FF00) << 4) |
-	      ((pix & 0x00FF0000) << 6);
-	return pix | ((pix >> 8) & 0x00300C03);
-}
-
-static u32 convert_xrgb8888_to_argb2101010(u32 pix)
-{
-	pix = ((pix & 0x000000FF) << 2) |
-	      ((pix & 0x0000FF00) << 4) |
-	      ((pix & 0x00FF0000) << 6);
-	return GENMASK(31, 30) /* set alpha bits */ | pix | ((pix >> 8) & 0x00300C03);
-}
-
-static u32 convert_xrgb8888_to_abgr2101010(u32 pix)
-{
-	pix = ((pix & 0x00FF0000) >> 14) |
-	      ((pix & 0x0000FF00) << 4) |
-	      ((pix & 0x000000FF) << 22);
-	return GENMASK(31, 30) /* set alpha bits */ | pix | ((pix >> 8) & 0x00300C03);
-}
+#include "drm_format_internal.h"
 
 /**
  * drm_draw_color_from_xrgb8888 - convert one pixel from xrgb8888 to the desired format

@@ -104,28 +26,28 @@ u32 drm_draw_color_from_xrgb8888(u32 color, u32 format)
 {
 	switch (format) {
 	case DRM_FORMAT_RGB565:
-		return convert_xrgb8888_to_rgb565(color);
+		return drm_pixel_xrgb8888_to_rgb565(color);
 	case DRM_FORMAT_RGBA5551:
-		return convert_xrgb8888_to_rgba5551(color);
+		return drm_pixel_xrgb8888_to_rgba5551(color);
 	case DRM_FORMAT_XRGB1555:
-		return convert_xrgb8888_to_xrgb1555(color);
+		return drm_pixel_xrgb8888_to_xrgb1555(color);
 	case DRM_FORMAT_ARGB1555:
-		return convert_xrgb8888_to_argb1555(color);
+		return drm_pixel_xrgb8888_to_argb1555(color);
 	case DRM_FORMAT_RGB888:
 	case DRM_FORMAT_XRGB8888:
 		return color;
 	case DRM_FORMAT_ARGB8888:
-		return convert_xrgb8888_to_argb8888(color);
+		return drm_pixel_xrgb8888_to_argb8888(color);
 	case DRM_FORMAT_XBGR8888:
-		return convert_xrgb8888_to_xbgr8888(color);
+		return drm_pixel_xrgb8888_to_xbgr8888(color);
 	case DRM_FORMAT_ABGR8888:
-		return convert_xrgb8888_to_abgr8888(color);
+		return drm_pixel_xrgb8888_to_abgr8888(color);
 	case DRM_FORMAT_XRGB2101010:
-		return convert_xrgb8888_to_xrgb2101010(color);
+		return drm_pixel_xrgb8888_to_xrgb2101010(color);
 	case DRM_FORMAT_ARGB2101010:
-		return convert_xrgb8888_to_argb2101010(color);
+		return drm_pixel_xrgb8888_to_argb2101010(color);
 	case DRM_FORMAT_ABGR2101010:
-		return convert_xrgb8888_to_abgr2101010(color);
+		return drm_pixel_xrgb8888_to_abgr2101010(color);
 	default:
 		WARN_ONCE(1, "Can't convert to %p4cc\n", &format);
 		return 0;
drivers/gpu/drm/drm_format_internal.h new file +119 -0

119 lines not on this page · raw message

Signature · 1 line
2.48.1
author maintainer
#2 lore raw

PATCH v2 2/8 drm/format-helper: Add generic conversion to 32-bit formats

In reply to Thomas Zimmermann

Add drm_fb_xfrm_line_32to32() to implement conversion from 32-bit pixels to 32-bit pixels. The pixel-conversion is specified by the given callback parameter. Mark the helper as always_inline to avoid overhead from function calls.

Then implement all existing line-conversion functions with the new generic call and the respective pixel-conversion helper.

Signed-off-by Thomas Zimmermann <[email protected]>
Reviewed-by Jocelyn Falempe <[email protected]>
---
 drivers/gpu/drm/drm_format_helper.c | 84 +++++++----------------------
 1 file changed, 19 insertions(+), 65 deletions(-)
drivers/gpu/drm/drm_format_helper.c +19 -65
index 01d3ab307ac3e..abd18c23cfbb5 100644
@@ -20,6 +20,8 @@
 #include <drm/drm_print.h>
 #include <drm/drm_rect.h>
 
+#include "drm_format_internal.h"
+
 /**
  * drm_format_conv_state_init - Initialize format-conversion state
  * @state: The state to initialize

@@ -244,6 +246,18 @@ static int drm_fb_xfrm(struct iosys_map *dst,
 				     xfrm_line);
 }
 
+static __always_inline void drm_fb_xfrm_line_32to32(void *dbuf, const void *sbuf,
+						    unsigned int pixels,
+						    u32 (*xfrm_pixel)(u32))
+{
+	__le32 *dbuf32 = dbuf;
+	const __le32 *sbuf32 = sbuf;
+	const __le32 *send32 = sbuf32 + pixels;
+
+	while (sbuf32 < send32)
+		*dbuf32++ = cpu_to_le32(xfrm_pixel(le32_to_cpup(sbuf32++)));
+}
+
 /**
  * drm_fb_memcpy - Copy clip buffer
  * @dst: Array of destination buffers

@@ -755,16 +769,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_bgr888);
 
 static void drm_fb_xrgb8888_to_argb8888_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le32 *dbuf32 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		pix |= GENMASK(31, 24); /* fill alpha bits */
-		dbuf32[x] = cpu_to_le32(pix);
-	}
+	drm_fb_xfrm_line_32to32(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_argb8888);
 }
 
 /**

@@ -804,19 +809,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_argb8888);
 
 static void drm_fb_xrgb8888_to_abgr8888_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le32 *dbuf32 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		pix = ((pix & 0x00ff0000) >> 16) <<  0 |
-		      ((pix & 0x0000ff00) >>  8) <<  8 |
-		      ((pix & 0x000000ff) >>  0) << 16 |
-		      GENMASK(31, 24); /* fill alpha bits */
-		*dbuf32++ = cpu_to_le32(pix);
-	}
+	drm_fb_xfrm_line_32to32(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_abgr8888);
 }
 
 static void drm_fb_xrgb8888_to_abgr8888(struct iosys_map *dst, const unsigned int *dst_pitch,

@@ -835,19 +828,7 @@ static void drm_fb_xrgb8888_to_abgr8888(struct iosys_map *dst, const unsigned in
 
 static void drm_fb_xrgb8888_to_xbgr8888_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le32 *dbuf32 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		pix = ((pix & 0x00ff0000) >> 16) <<  0 |
-		      ((pix & 0x0000ff00) >>  8) <<  8 |
-		      ((pix & 0x000000ff) >>  0) << 16 |
-		      ((pix & 0xff000000) >> 24) << 24;
-		*dbuf32++ = cpu_to_le32(pix);
-	}
+	drm_fb_xfrm_line_32to32(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_xbgr8888);
 }
 
 static void drm_fb_xrgb8888_to_xbgr8888(struct iosys_map *dst, const unsigned int *dst_pitch,

@@ -866,20 +847,7 @@ static void drm_fb_xrgb8888_to_xbgr8888(struct iosys_map *dst, const unsigned in
 
 static void drm_fb_xrgb8888_to_xrgb2101010_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le32 *dbuf32 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 val32;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		val32 = ((pix & 0x000000FF) << 2) |
-			((pix & 0x0000FF00) << 4) |
-			((pix & 0x00FF0000) << 6);
-		pix = val32 | ((val32 >> 8) & 0x00300C03);
-		*dbuf32++ = cpu_to_le32(pix);
-	}
+	drm_fb_xfrm_line_32to32(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_xrgb2101010);
 }
 
 /**

@@ -920,21 +888,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_xrgb2101010);
 
 static void drm_fb_xrgb8888_to_argb2101010_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le32 *dbuf32 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 val32;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		val32 = ((pix & 0x000000ff) << 2) |
-			((pix & 0x0000ff00) << 4) |
-			((pix & 0x00ff0000) << 6);
-		pix = GENMASK(31, 30) | /* set alpha bits */
-		      val32 | ((val32 >> 8) & 0x00300c03);
-		*dbuf32++ = cpu_to_le32(pix);
-	}
+	drm_fb_xfrm_line_32to32(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_argb2101010);
 }
 
 /**
Signature · 1 line
2.48.1
author maintainer
#3 lore raw

PATCH v2 3/8 drm/format-helper: Add generic conversion to 24-bit formats

In reply to Thomas Zimmermann

Add drm_fb_xfrm_line_32to24() to implement conversion from 32-bit pixels to 24-bit pixels. The pixel-conversion is specified by the given callback parameter. Mark the helper as always_inline to avoid overhead from function calls.

Then implement all existing line-conversion functions with the new generic call and the respective pixel-conversion helper.

Signed-off-by Thomas Zimmermann <[email protected]>
Reviewed-by Jocelyn Falempe <[email protected]>
---
 drivers/gpu/drm/drm_format_helper.c   | 43 ++++++++++++---------------
 drivers/gpu/drm/drm_format_internal.h | 12 ++++++++
 2 files changed, 31 insertions(+), 24 deletions(-)
drivers/gpu/drm/drm_format_helper.c +19 -24
index abd18c23cfbb5..5a2fe3d685a2c 100644
@@ -246,6 +246,23 @@ static int drm_fb_xfrm(struct iosys_map *dst,
 				     xfrm_line);
 }
 
+static __always_inline void drm_fb_xfrm_line_32to24(void *dbuf, const void *sbuf,
+						    unsigned int pixels,
+						    u32 (*xfrm_pixel)(u32))
+{
+	u8 *dbuf8 = dbuf;
+	const __le32 *sbuf32 = sbuf;
+	const __le32 *send32 = sbuf32 + pixels;
+
+	while (sbuf32 < send32) {
+		u32 val24 = xfrm_pixel(le32_to_cpup(sbuf32++));
+		/* write output in reverse order for little endianness */
+		*dbuf8++ = (val24 & 0x000000ff);
+		*dbuf8++ = (val24 & 0x0000ff00) >>  8;
+		*dbuf8++ = (val24 & 0x00ff0000) >> 16;
+	}
+}
+
 static __always_inline void drm_fb_xfrm_line_32to32(void *dbuf, const void *sbuf,
 						    unsigned int pixels,
 						    u32 (*xfrm_pixel)(u32))

@@ -667,18 +684,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_rgba5551);
 
 static void drm_fb_xrgb8888_to_rgb888_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	u8 *dbuf8 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		/* write blue-green-red to output in little endianness */
-		*dbuf8++ = (pix & 0x000000FF) >>  0;
-		*dbuf8++ = (pix & 0x0000FF00) >>  8;
-		*dbuf8++ = (pix & 0x00FF0000) >> 16;
-	}
+	drm_fb_xfrm_line_32to24(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_rgb888);
 }
 
 /**

@@ -718,18 +724,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_rgb888);
 
 static void drm_fb_xrgb8888_to_bgr888_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	u8 *dbuf8 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		/* write red-green-blue to output in little endianness */
-		*dbuf8++ = (pix & 0x00ff0000) >> 16;
-		*dbuf8++ = (pix & 0x0000ff00) >> 8;
-		*dbuf8++ = (pix & 0x000000ff) >> 0;
-	}
+	drm_fb_xfrm_line_32to24(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_bgr888);
 }
 
 /**
drivers/gpu/drm/drm_format_internal.h +12 -0
index 5f82f0b9c8e81..e7fcf260a9146 100644
@@ -68,6 +68,18 @@ static inline u32 drm_pixel_xrgb8888_to_argb1555(u32 pix)
 	       drm_pixel_xrgb8888_to_xrgb1555(pix);
 }
 
+static inline u32 drm_pixel_xrgb8888_to_rgb888(u32 pix)
+{
+	return pix & GENMASK(23, 0);
+}
+
+static inline u32 drm_pixel_xrgb8888_to_bgr888(u32 pix)
+{
+	return ((pix & 0x00ff0000) >> 16) |
+	       ((pix & 0x0000ff00)) |
+	       ((pix & 0x000000ff) << 16);
+}
+
 static inline u32 drm_pixel_xrgb8888_to_argb8888(u32 pix)
 {
 	return GENMASK(31, 24) | /* fill alpha bits */
Signature · 1 line
2.48.1
author maintainer
#4 lore raw

PATCH v2 4/8 drm/format-helper: Add generic conversion to 16-bit formats

In reply to Thomas Zimmermann

Add drm_fb_xfrm_line_32to16() to implement conversion from 32-bit pixels to 16-bit pixels. The pixel-conversion is specified by the given callback parameter. Mark the helper as always_inline to avoid overhead from function calls.

Then implement all existing line-conversion functions with the new generic call and the respective pixel-conversion helper. There's one pixel-conversion helper that swaps output bytes. It is for gud and requires refactoring, so don't move it into the header file.

Signed-off-by Thomas Zimmermann <[email protected]>
Reviewed-by Jocelyn Falempe <[email protected]>
---
 drivers/gpu/drm/drm_format_helper.c   | 118 +++++---------------------
 drivers/gpu/drm/drm_format_internal.h |  12 +++
 2 files changed, 34 insertions(+), 96 deletions(-)
drivers/gpu/drm/drm_format_helper.c +22 -96
index 5a2fe3d685a2c..aecf55c1fc6ba 100644
@@ -246,6 +246,18 @@ static int drm_fb_xfrm(struct iosys_map *dst,
 				     xfrm_line);
 }
 
+static __always_inline void drm_fb_xfrm_line_32to16(void *dbuf, const void *sbuf,
+						    unsigned int pixels,
+						    u32 (*xfrm_pixel)(u32))
+{
+	__le16 *dbuf16 = dbuf;
+	const __le32 *sbuf32 = sbuf;
+	const __le32 *send32 = sbuf32 + pixels;
+
+	while (sbuf32 < send32)
+		*dbuf16++ = cpu_to_le16(xfrm_pixel(le32_to_cpup(sbuf32++)));
+}
+
 static __always_inline void drm_fb_xfrm_line_32to24(void *dbuf, const void *sbuf,
 						    unsigned int pixels,
 						    u32 (*xfrm_pixel)(u32))

@@ -448,38 +460,19 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_rgb332);
 
 static void drm_fb_xrgb8888_to_rgb565_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le16 *dbuf16 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u16 val16;
-	u32 pix;
+	drm_fb_xfrm_line_32to16(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_rgb565);
+}
 
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		val16 = ((pix & 0x00F80000) >> 8) |
-			((pix & 0x0000FC00) >> 5) |
-			((pix & 0x000000F8) >> 3);
-		dbuf16[x] = cpu_to_le16(val16);
-	}
+static __always_inline u32 drm_xrgb8888_to_rgb565_swab(u32 pix)
+{
+	return swab16(drm_pixel_xrgb8888_to_rgb565(pix));
 }
 
 /* TODO: implement this helper as conversion to RGB565|BIG_ENDIAN */
 static void drm_fb_xrgb8888_to_rgb565_swab_line(void *dbuf, const void *sbuf,
 						unsigned int pixels)
 {
-	__le16 *dbuf16 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u16 val16;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		val16 = ((pix & 0x00F80000) >> 8) |
-			((pix & 0x0000FC00) >> 5) |
-			((pix & 0x000000F8) >> 3);
-		dbuf16[x] = cpu_to_le16(swab16(val16));
-	}
+	drm_fb_xfrm_line_32to16(dbuf, sbuf, pixels, drm_xrgb8888_to_rgb565_swab);
 }
 
 /**

@@ -526,19 +519,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_rgb565);
 
 static void drm_fb_xrgb8888_to_xrgb1555_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le16 *dbuf16 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u16 val16;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		val16 = ((pix & 0x00f80000) >> 9) |
-			((pix & 0x0000f800) >> 6) |
-			((pix & 0x000000f8) >> 3);
-		dbuf16[x] = cpu_to_le16(val16);
-	}
+	drm_fb_xfrm_line_32to16(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_xrgb1555);
 }
 
 /**

@@ -578,20 +559,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_xrgb1555);
 
 static void drm_fb_xrgb8888_to_argb1555_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le16 *dbuf16 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u16 val16;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		val16 = BIT(15) | /* set alpha bit */
-			((pix & 0x00f80000) >> 9) |
-			((pix & 0x0000f800) >> 6) |
-			((pix & 0x000000f8) >> 3);
-		dbuf16[x] = cpu_to_le16(val16);
-	}
+	drm_fb_xfrm_line_32to16(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_argb1555);
 }
 
 /**

@@ -631,20 +599,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_argb1555);
 
 static void drm_fb_xrgb8888_to_rgba5551_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	__le16 *dbuf16 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u16 val16;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		val16 = ((pix & 0x00f80000) >> 8) |
-			((pix & 0x0000f800) >> 5) |
-			((pix & 0x000000f8) >> 2) |
-			BIT(0); /* set alpha bit */
-		dbuf16[x] = cpu_to_le16(val16);
-	}
+	drm_fb_xfrm_line_32to16(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_rgba5551);
 }
 
 /**

@@ -980,36 +935,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_gray8);
 
 static void drm_fb_argb8888_to_argb4444_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	unsigned int pixels2 = pixels & ~GENMASK_ULL(0, 0);
-	__le32 *dbuf32 = dbuf;
-	__le16 *dbuf16 = dbuf + pixels2 * sizeof(*dbuf16);
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 val32;
-	u16 val16;
-	u32 pix[2];
-
-	for (x = 0; x < pixels2; x += 2, ++dbuf32) {
-		pix[0] = le32_to_cpu(sbuf32[x]);
-		pix[1] = le32_to_cpu(sbuf32[x + 1]);
-		val32 = ((pix[0] & 0xf0000000) >> 16) |
-			((pix[0] & 0x00f00000) >> 12) |
-			((pix[0] & 0x0000f000) >> 8) |
-			((pix[0] & 0x000000f0) >> 4) |
-			((pix[1] & 0xf0000000) >> 0) |
-			((pix[1] & 0x00f00000) << 4) |
-			((pix[1] & 0x0000f000) << 8) |
-			((pix[1] & 0x000000f0) << 12);
-		*dbuf32 = cpu_to_le32(val32);
-	}
-	for (; x < pixels; x++) {
-		pix[0] = le32_to_cpu(sbuf32[x]);
-		val16 = ((pix[0] & 0xf0000000) >> 16) |
-			((pix[0] & 0x00f00000) >> 12) |
-			((pix[0] & 0x0000f000) >> 8) |
-			((pix[0] & 0x000000f0) >> 4);
-		dbuf16[x] = cpu_to_le16(val16);
-	}
+	drm_fb_xfrm_line_32to16(dbuf, sbuf, pixels, drm_pixel_argb8888_to_argb4444);
 }
 
 /**
drivers/gpu/drm/drm_format_internal.h +12 -0
index e7fcf260a9146..4f20b63cb6f60 100644
@@ -128,4 +128,16 @@ static inline u32 drm_pixel_xrgb8888_to_abgr2101010(u32 pix)
 	       drm_pixel_xrgb8888_to_xbgr2101010(pix);
 }
 
+/*
+ * Conversion from ARGB8888
+ */
+
+static inline u32 drm_pixel_argb8888_to_argb4444(u32 pix)
+{
+	return ((pix & 0xf0000000) >> 16) |
+	       ((pix & 0x00f00000) >> 12) |
+	       ((pix & 0x0000f000) >> 8) |
+	       ((pix & 0x000000f0) >> 4);
+}
+
 #endif
Signature · 1 line
2.48.1
author maintainer
#5 lore raw

PATCH v2 5/8 drm/format-helper: Add generic conversion to 8-bit formats

In reply to Thomas Zimmermann

Add drm_fb_xfrm_line_32to8() to implement conversion from 32-bit pixels to 8-bit pixels. The pixel-conversion is specified by the given callback parameter. Mark the helper as always_inline to avoid overhead from function calls.

Then implement all existing line-conversion functions with the new generic call and the respective pixel-conversion helper.

Signed-off-by Thomas Zimmermann <[email protected]>
Reviewed-by Jocelyn Falempe <[email protected]>
---
 drivers/gpu/drm/drm_format_helper.c   | 38 ++++++++++-----------------
 drivers/gpu/drm/drm_format_internal.h | 17 ++++++++++++
 2 files changed, 31 insertions(+), 24 deletions(-)
drivers/gpu/drm/drm_format_helper.c +14 -24
index aecf55c1fc6ba..a926aa6671fcd 100644
@@ -246,6 +246,18 @@ static int drm_fb_xfrm(struct iosys_map *dst,
 				     xfrm_line);
 }
 
+static __always_inline void drm_fb_xfrm_line_32to8(void *dbuf, const void *sbuf,
+						   unsigned int pixels,
+						   u32 (*xfrm_pixel)(u32))
+{
+	u8 *dbuf8 = dbuf;
+	const __le32 *sbuf32 = sbuf;
+	const __le32 *send32 = sbuf32 + pixels;
+
+	while (sbuf32 < send32)
+		*dbuf8++ = xfrm_pixel(le32_to_cpup(sbuf32++));
+}
+
 static __always_inline void drm_fb_xfrm_line_32to16(void *dbuf, const void *sbuf,
 						    unsigned int pixels,
 						    u32 (*xfrm_pixel)(u32))

@@ -411,17 +423,7 @@ EXPORT_SYMBOL(drm_fb_swab);
 
 static void drm_fb_xrgb8888_to_rgb332_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	u8 *dbuf8 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-	u32 pix;
-
-	for (x = 0; x < pixels; x++) {
-		pix = le32_to_cpu(sbuf32[x]);
-		dbuf8[x] = ((pix & 0x00e00000) >> 16) |
-			   ((pix & 0x0000e000) >> 11) |
-			   ((pix & 0x000000c0) >> 6);
-	}
+	drm_fb_xfrm_line_32to8(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_rgb332);
 }
 
 /**

@@ -879,19 +881,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_argb2101010);
 
 static void drm_fb_xrgb8888_to_gray8_line(void *dbuf, const void *sbuf, unsigned int pixels)
 {
-	u8 *dbuf8 = dbuf;
-	const __le32 *sbuf32 = sbuf;
-	unsigned int x;
-
-	for (x = 0; x < pixels; x++) {
-		u32 pix = le32_to_cpu(sbuf32[x]);
-		u8 r = (pix & 0x00ff0000) >> 16;
-		u8 g = (pix & 0x0000ff00) >> 8;
-		u8 b =  pix & 0x000000ff;
-
-		/* ITU BT.601: Y = 0.299 R + 0.587 G + 0.114 B */
-		*dbuf8++ = (3 * r + 6 * g + b) / 10;
-	}
+	drm_fb_xfrm_line_32to8(dbuf, sbuf, pixels, drm_pixel_xrgb8888_to_r8_bt601);
 }
 
 /**
drivers/gpu/drm/drm_format_internal.h +17 -0
index 4f20b63cb6f60..9f857bfa368d1 100644
@@ -35,6 +35,23 @@
  * Conversions from XRGB8888
  */
 
+static inline u32 drm_pixel_xrgb8888_to_r8_bt601(u32 pix)
+{
+	u32 r = (pix & 0x00ff0000) >> 16;
+	u32 g = (pix & 0x0000ff00) >> 8;
+	u32 b =  pix & 0x000000ff;
+
+	/* ITU-R BT.601: Y = 0.299 R + 0.587 G + 0.114 B */
+	return (3 * r + 6 * g + b) / 10;
+}
+
+static inline u32 drm_pixel_xrgb8888_to_rgb332(u32 pix)
+{
+	return ((pix & 0x00e00000) >> 16) |
+	       ((pix & 0x0000e000) >> 11) |
+	       ((pix & 0x000000c0) >> 6);
+}
+
 static inline u32 drm_pixel_xrgb8888_to_rgb565(u32 pix)
 {
 	return ((pix & 0x00f80000) >> 8) |
Signature · 1 line
2.48.1
author maintainer
#6 lore raw

PATCH v2 6/8 drm/format-helper: Optimize 32-to-24-bpp conversion

In reply to Thomas Zimmermann

For ease of implementation, existing line-conversion functions for 24-bit formats write each byte individually. Optimize the performance by writing 4 pixels in 3 32-bit stores.

v2:
- simplify address calculation (Jani)

Signed-off-by Thomas Zimmermann <[email protected]>
Reviewed-by Jocelyn Falempe <[email protected]>
---
 drivers/gpu/drm/drm_format_helper.c | 37 ++++++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)
drivers/gpu/drm/drm_format_helper.c +36 -1
index a926aa6671fcd..daf5a6d4f2835 100644
@@ -246,6 +246,9 @@ static int drm_fb_xfrm(struct iosys_map *dst,
 				     xfrm_line);
 }
 
+#define ALIGN_DOWN_PIXELS(end, n, a) \
+	((end) - ((n) & ((a) - 1)))
+
 static __always_inline void drm_fb_xfrm_line_32to8(void *dbuf, const void *sbuf,
 						   unsigned int pixels,
 						   u32 (*xfrm_pixel)(u32))

@@ -274,10 +277,42 @@ static __always_inline void drm_fb_xfrm_line_32to24(void *dbuf, const void *sbuf
 						    unsigned int pixels,
 						    u32 (*xfrm_pixel)(u32))
 {
-	u8 *dbuf8 = dbuf;
+	__le32 *dbuf32 = dbuf;
+	u8 *dbuf8;
 	const __le32 *sbuf32 = sbuf;
 	const __le32 *send32 = sbuf32 + pixels;
 
+	/* write pixels in chunks of 4 */
+	while (sbuf32 < ALIGN_DOWN_PIXELS(send32, pixels, 4)) {
+		u32 val24[4] = {
+			xfrm_pixel(le32_to_cpup(sbuf32++)),
+			xfrm_pixel(le32_to_cpup(sbuf32++)),
+			xfrm_pixel(le32_to_cpup(sbuf32++)),
+			xfrm_pixel(le32_to_cpup(sbuf32++)),
+		};
+		u32 out32[3] = {
+			/* write output bytes in reverse order for little endianness */
+			((val24[0] & 0x000000ff)) |
+			((val24[0] & 0x0000ff00)) |
+			((val24[0] & 0x00ff0000)) |
+			((val24[1] & 0x000000ff) << 24),
+			((val24[1] & 0x0000ff00) >> 8) |
+			((val24[1] & 0x00ff0000) >> 8) |
+			((val24[2] & 0x000000ff) << 16) |
+			((val24[2] & 0x0000ff00) << 16),
+			((val24[2] & 0x00ff0000) >> 16) |
+			((val24[3] & 0x000000ff) << 8) |
+			((val24[3] & 0x0000ff00) << 8) |
+			((val24[3] & 0x00ff0000) << 8),
+		};
+
+		*dbuf32++ = cpu_to_le32(out32[0]);
+		*dbuf32++ = cpu_to_le32(out32[1]);
+		*dbuf32++ = cpu_to_le32(out32[2]);
+	}
+
+	/* write trailing pixel */
+	dbuf8 = (u8 __force *)dbuf32;
 	while (sbuf32 < send32) {
 		u32 val24 = xfrm_pixel(le32_to_cpup(sbuf32++));
 		/* write output in reverse order for little endianness */
Signature · 1 line
2.48.1
author maintainer
#7 lore raw

PATCH v2 7/8 drm/format-helper: Optimize 32-to-16-bpp conversion

In reply to Thomas Zimmermann

For ease of implementation, existing line-conversion functions for 16-bit formats write each pixel individually. Optimize the performance by writing multiple pixels in single 64-bit and 32-bit stores.

v2:
- simplify address calculation (Jani)
- fix typo in commit message (Jocelyn)

Signed-off-by Thomas Zimmermann <[email protected]>
Reviewed-by Jocelyn Falempe <[email protected]>
---
 drivers/gpu/drm/drm_format_helper.c | 37 ++++++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)
drivers/gpu/drm/drm_format_helper.c +36 -1
index daf5a6d4f2835..5098a00803d23 100644
@@ -265,10 +265,45 @@ static __always_inline void drm_fb_xfrm_line_32to16(void *dbuf, const void *sbuf
 						    unsigned int pixels,
 						    u32 (*xfrm_pixel)(u32))
 {
-	__le16 *dbuf16 = dbuf;
+	__le64 *dbuf64 = dbuf;
+	__le32 *dbuf32;
+	__le16 *dbuf16;
 	const __le32 *sbuf32 = sbuf;
 	const __le32 *send32 = sbuf32 + pixels;
 
+#if defined(CONFIG_64BIT)
+	/* write 4 pixels at once */
+	while (sbuf32 < ALIGN_DOWN_PIXELS(send32, pixels, 4)) {
+		u32 pix[4] = {
+			le32_to_cpup(sbuf32++),
+			le32_to_cpup(sbuf32++),
+			le32_to_cpup(sbuf32++),
+			le32_to_cpup(sbuf32++),
+		};
+		/* write output bytes in reverse order for little endianness */
+		u64 val64 = ((u64)xfrm_pixel(pix[0])) |
+			    ((u64)xfrm_pixel(pix[1]) << 16) |
+			    ((u64)xfrm_pixel(pix[2]) << 32) |
+			    ((u64)xfrm_pixel(pix[3]) << 48);
+		*dbuf64++ = cpu_to_le64(val64);
+	}
+#endif
+
+	/* write 2 pixels at once */
+	dbuf32 = (__le32 __force *)dbuf64;
+	while (sbuf32 < ALIGN_DOWN_PIXELS(send32, pixels, 2)) {
+		u32 pix[2] = {
+			le32_to_cpup(sbuf32++),
+			le32_to_cpup(sbuf32++),
+		};
+		/* write output bytes in reverse order for little endianness */
+		u32 val32 = xfrm_pixel(pix[0]) |
+			   (xfrm_pixel(pix[1]) << 16);
+		*dbuf32++ = cpu_to_le32(val32);
+	}
+
+	/* write trailing pixel */
+	dbuf16 = (__le16 __force *)dbuf32;
 	while (sbuf32 < send32)
 		*dbuf16++ = cpu_to_le16(xfrm_pixel(le32_to_cpup(sbuf32++)));
 }
Signature · 1 line
2.48.1
author maintainer
#8 lore raw

PATCH v2 8/8 drm/format-helper: Optimize 32-to-8-bpp conversion

In reply to Thomas Zimmermann

For ease of implementation, existing line-conversion functions for 8-bit formats write each pixel individually. Optimize the performance by writing multiple pixels in a single 32-bit store.

v2:
- simplify address calculation (Jani)
- fix typo in commit message (Jocelyn)

Signed-off-by Thomas Zimmermann <[email protected]>
Reviewed-by Jocelyn Falempe <[email protected]>
---
 drivers/gpu/drm/drm_format_helper.c | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)
drivers/gpu/drm/drm_format_helper.c +20 -1
index 5098a00803d23..fc7347caf600e 100644
@@ -253,10 +253,29 @@ static __always_inline void drm_fb_xfrm_line_32to8(void *dbuf, const void *sbuf,
 						   unsigned int pixels,
 						   u32 (*xfrm_pixel)(u32))
 {
-	u8 *dbuf8 = dbuf;
+	__le32 *dbuf32 = dbuf;
+	u8 *dbuf8;
 	const __le32 *sbuf32 = sbuf;
 	const __le32 *send32 = sbuf32 + pixels;
 
+	/* write 4 pixels at once */
+	while (sbuf32 < ALIGN_DOWN_PIXELS(send32, pixels, 4)) {
+		u32 pix[4] = {
+			le32_to_cpup(sbuf32++),
+			le32_to_cpup(sbuf32++),
+			le32_to_cpup(sbuf32++),
+			le32_to_cpup(sbuf32++),
+		};
+		/* write output bytes in reverse order for little endianness */
+		u32 val32 = xfrm_pixel(pix[0]) |
+			   (xfrm_pixel(pix[1]) << 8) |
+			   (xfrm_pixel(pix[2]) << 16) |
+			   (xfrm_pixel(pix[3]) << 24);
+		*dbuf32++ = cpu_to_le32(val32);
+	}
+
+	/* write trailing pixels */
+	dbuf8 = (u8 __force *)dbuf32;
 	while (sbuf32 < send32)
 		*dbuf8++ = xfrm_pixel(le32_to_cpup(sbuf32++));
 }
Signature · 1 line
2.48.1