~ruther/qmk_firmware

84b8cdc1a43245d24b232bc8e28cca7f1183d676 — Karl Shea 3 years ago 23d3ebd
Fix bit loss in cie_lightness() when doing division to resolve #15331 (#15344)

* Fix bit loss in cie_lightness() when doing division.

* Use the right types

* Format

Co-authored-by: zvecr <git@zvecr.com>
1 files changed, 4 insertions(+), 4 deletions(-)

M quantum/backlight/backlight_avr.c
M quantum/backlight/backlight_avr.c => quantum/backlight/backlight_avr.c +4 -4
@@ 232,19 232,19 @@ ISR(TIMERx_OVF_vect) {

// See http://jared.geek.nz/2013/feb/linear-led-pwm
static uint16_t cie_lightness(uint16_t v) {
    if (v <= ICRx / 12)  // If the value is less than or equal to ~8% of max
    if (v <= (uint32_t)ICRx / 12)  // If the value is less than or equal to ~8% of max
    {
        return v / 9;  // Same as dividing by 900%
    } else {
        // In the next two lines values are bit-shifted. This is to avoid loosing decimals in integer math.
        uint32_t y   = (((uint32_t)v + ICRx / 6) << 5) / (ICRx / 6 + ICRx);  // If above 8%, add ~16% of max, and normalize with (max + ~16% max)
        uint32_t out = (y * y * y * ICRx) >> 15;                             // Cube it and undo the bit-shifting. (which is now three times as much due to the cubing)
        uint32_t y   = (((uint32_t)v + (uint32_t)ICRx / 6) << 5) / ((uint32_t)ICRx / 6 + ICRx);  // If above 8%, add ~16% of max, and normalize with (max + ~16% max)
        uint32_t out = (y * y * y * ICRx) >> 15;                                                 // Cube it and undo the bit-shifting. (which is now three times as much due to the cubing)

        if (out > ICRx)  // Avoid overflows
        {
            out = ICRx;
        }
        return out;
        return (uint16_t)out;
    }
}