drm/format-helper: Optimize 32-to-24-bpp conversion

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 <tzimmermann@suse.de>
Reviewed-by: Jocelyn Falempe <jfalempe@redhat.com>
Link: https://lore.kernel.org/r/20250328141709.217283-7-tzimmermann@suse.de

diff --git a/drivers/gpu/drm/drm_format_helper.c b/drivers/gpu/drm/drm_format_helper.c
index a926aa6671fcd436076208e7c80f8793ea9649b5..daf5a6d4f28355abbef0c8cf574c943ae2c31bc5 100644 (file)
--- a/drivers/gpu/drm/drm_format_helper.c
+++ b/drivers/gpu/drm/drm_format_helper.c
@@ -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 */