diff options
author | Steven Fuerst <svfuerst@gmail.com> | 2012-08-15 18:07:15 -0400 |
---|---|---|
committer | Alex Deucher <alexander.deucher@amd.com> | 2012-09-20 13:10:34 -0400 |
commit | 747f49ba67b8895a5831ab539de551b916f3738c (patch) | |
tree | b1b1561ae97f9f7fdab8ae44e019bd5dca0e62e4 /drivers/gpu/drm | |
parent | 7ff64fcaa7b7ba62d12758e49643f31dd9e90ece (diff) |
Replace int2float() with an optimized version.
We use __fls() to find the most significant bit. Using that, the
loop can be avoided. A second trick is to use the behaviour of the
rotate instructions to expand the range of the unsigned int to float
conversion to the full 32 bits in a branchless way.
The routine is now exact up to 2^24. Above that, we truncate which
is equivalent to rounding towards zero.
Signed-off-by: Steven Fuerst <svfuerst@gmail.com>
Diffstat (limited to 'drivers/gpu/drm')
-rw-r--r-- | drivers/gpu/drm/radeon/r600_blit.c | 51 |
1 files changed, 28 insertions, 23 deletions
diff --git a/drivers/gpu/drm/radeon/r600_blit.c b/drivers/gpu/drm/radeon/r600_blit.c index ee1b815b5689..7d8ac42e3846 100644 --- a/drivers/gpu/drm/radeon/r600_blit.c +++ b/drivers/gpu/drm/radeon/r600_blit.c | |||
@@ -489,31 +489,36 @@ set_default_state(drm_radeon_private_t *dev_priv) | |||
489 | ADVANCE_RING(); | 489 | ADVANCE_RING(); |
490 | } | 490 | } |
491 | 491 | ||
492 | uint32_t int2float(uint32_t input) | 492 | /* 23 bits of float fractional data */ |
493 | #define I2F_FRAC_BITS 23 | ||
494 | #define I2F_MASK ((1 << I2F_FRAC_BITS) - 1) | ||
495 | |||
496 | /* | ||
497 | * Converts unsigned integer into 32-bit IEEE floating point representation. | ||
498 | * Will be exact from 0 to 2^24. Above that, we round towards zero | ||
499 | * as the fractional bits will not fit in a float. (It would be better to | ||
500 | * round towards even as the fpu does, but that is slower.) | ||
501 | */ | ||
502 | uint32_t int2float(uint32_t x) | ||
493 | { | 503 | { |
494 | u32 result, i, exponent, fraction; | 504 | uint32_t msb, exponent, fraction; |
495 | 505 | ||
496 | if ((input & 0x3fff) == 0) | 506 | /* Zero is special */ |
497 | result = 0; /* 0 is a special case */ | 507 | if (!x) return 0; |
498 | else { | 508 | |
499 | exponent = 140; /* exponent biased by 127; */ | 509 | /* Get location of the most significant bit */ |
500 | fraction = (input & 0x3fff) << 10; /* cheat and only | 510 | msb = __fls(x); |
501 | handle numbers below 2^^15 */ | ||
502 | for (i = 0; i < 14; i++) { | ||
503 | if (fraction & 0x800000) | ||
504 | break; | ||
505 | else { | ||
506 | fraction = fraction << 1; /* keep | ||
507 | shifting left until top bit = 1 */ | ||
508 | exponent = exponent - 1; | ||
509 | } | ||
510 | } | ||
511 | result = exponent << 23 | (fraction & 0x7fffff); /* mask | ||
512 | off top bit; assumed 1 */ | ||
513 | } | ||
514 | return result; | ||
515 | } | ||
516 | 511 | ||
512 | /* | ||
513 | * Use a rotate instead of a shift because that works both leftwards | ||
514 | * and rightwards due to the mod(32) behaviour. This means we don't | ||
515 | * need to check to see if we are above 2^24 or not. | ||
516 | */ | ||
517 | fraction = ror32(x, (msb - I2F_FRAC_BITS) & 0x1f) & I2F_MASK; | ||
518 | exponent = (127 + msb) << I2F_FRAC_BITS; | ||
519 | |||
520 | return fraction + exponent; | ||
521 | } | ||
517 | 522 | ||
518 | static int r600_nomm_get_vb(struct drm_device *dev) | 523 | static int r600_nomm_get_vb(struct drm_device *dev) |
519 | { | 524 | { |