From ba4be5157191af6fbb83756119ffd29147b7fb99 Mon Sep 17 00:00:00 2001 From: Andrew Mahone Date: Fri, 30 Jan 2009 10:52:30 +0000 Subject: per Jens Arnold's suggestion: use platform clz instruction only on ARMv5+, where we know it returns 32 for a 0 input, and remove the special case for 0 use binary search only to find most significant four bits, then use lookup table to add number of leading zeroes within those bits git-svn-id: svn://svn.rockbox.org/rockbox/trunk@19888 a1c6a512-1295-4272-9138-f99709370657 --- apps/plugins/pictureflow.c | 26 ++++++++------------------ 1 file changed, 8 insertions(+), 18 deletions(-) diff --git a/apps/plugins/pictureflow.c b/apps/plugins/pictureflow.c index 31e4c97468..cd7e54ce35 100644 --- a/apps/plugins/pictureflow.c +++ b/apps/plugins/pictureflow.c @@ -337,11 +337,9 @@ static inline PFreal fmul(PFreal a, PFreal b) return (a*b) >> PFREAL_SHIFT; } -/* ARMv5+ have a clz instruction. So do most modern desktop processors, and the - * simulator doesn't need to be as concerned about savings for inlining a calls - * to a clz function. +/* ARMv5+ has a clz instruction equivalent to our function. */ -#if defined(SIMULATOR) || (defined(CPU_ARM) && (ARM_ARCH > 4)) +#if (defined(CPU_ARM) && (ARM_ARCH > 4)) static inline int clz(uint32_t v) { return __builtin_clz(v); @@ -349,12 +347,14 @@ static inline int clz(uint32_t v) /* Otherwise, use our clz, which can be inlined */ #else +static const char clz_lut[16] = { 4, 3, 2, 2, 1, 1, 1, 1, + 0, 0, 0, 0, 0, 0, 0, 0 }; /* This clz is based on the log2(n) implementation at - * http://graphics.stanford.edu/~seander/bithacks.html#IntegerLog + * http://graphics.stanford.edu/~seander/bithacks.html#IntegerLogLookup */ static inline int clz(uint32_t v) { - uint32_t r = 31; + int r = 28; if (v >= 0x10000) { v >>= 16; @@ -370,14 +370,7 @@ static inline int clz(uint32_t v) v >>= 4; r -= 4; } - if (v & 0xc) - { - v >>= 2; - r -= 2; - } - if (v & 2) - r -= 1; - return r; + return r + clz_lut[v]; } #endif @@ -387,10 +380,7 @@ static inline int clz(uint32_t v) static inline int allowed_shift(int32_t val) { uint32_t uval = val ^ (val >> 31); - if (!uval) - return 31; - else - return clz(uval) - 1; + return clz(uval) - 1; } /* Calculate num/den, with the result shifted left by PFREAL_SHIFT, by shifting -- cgit v1.2.3