diff options
author | James Benton <[email protected]> | 2012-04-05 20:32:51 +0100 |
---|---|---|
committer | José Fonseca <[email protected]> | 2012-04-05 20:32:54 +0100 |
commit | 7c639feb2f6697b5da94e81e7901436ce1620c66 (patch) | |
tree | 99ad663eddb85433d07f5ef83ba69878f699baa7 /src/gallium/auxiliary/gallivm/lp_bld_arit.c | |
parent | 129d5138d63697a9812d531895293313dfad6d57 (diff) |
gallivm: Updated lp_build_polynomial to compute odd and even terms separately to decrease data dependency for faster runtime.
Signed-off-by: José Fonseca <[email protected]>
Diffstat (limited to 'src/gallium/auxiliary/gallivm/lp_bld_arit.c')
-rw-r--r-- | src/gallium/auxiliary/gallivm/lp_bld_arit.c | 32 |
1 files changed, 25 insertions, 7 deletions
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c index 6b17fbd2d31..d07cf689de3 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c @@ -2163,7 +2163,8 @@ lp_build_polynomial(struct lp_build_context *bld, unsigned num_coeffs) { const struct lp_type type = bld->type; - LLVMValueRef res = NULL; + LLVMValueRef even = NULL, odd = NULL; + LLVMValueRef x2; unsigned i; assert(lp_check_value(bld->type, x)); @@ -2175,19 +2176,36 @@ lp_build_polynomial(struct lp_build_context *bld, __FUNCTION__); } + /* + * Calculate odd and even terms seperately to decrease data dependency + * Ex: + * c[0] + x^2 * c[2] + x^4 * c[4] ... + * + x * (c[1] + x^2 * c[3] + x^4 * c[5]) ... + */ + x2 = lp_build_mul(bld, x, x); + for (i = num_coeffs; i--; ) { LLVMValueRef coeff; coeff = lp_build_const_vec(bld->gallivm, type, coeffs[i]); - if(res) - res = lp_build_add(bld, coeff, lp_build_mul(bld, x, res)); - else - res = coeff; + if (i % 2 == 0) { + if (even) + even = lp_build_add(bld, coeff, lp_build_mul(bld, x2, even)); + else + even = coeff; + } else { + if (odd) + odd = lp_build_add(bld, coeff, lp_build_mul(bld, x2, odd)); + else + odd = coeff; + } } - if(res) - return res; + if (odd) + return lp_build_add(bld, lp_build_mul(bld, odd, x), even); + else if (even) + return even; else return bld->undef; } |