Re: [Mesa-dev] [PATCH 2/3] gallivm: Prefer backend agnostic intrinsic for rounding.

2016-04-03 Thread Jose Fonseca
On 03/04/16 16:20, Roland Scheidegger wrote: Am 03.04.2016 um 12:24 schrieb Jose Fonseca: We could unconditionally use these instrinsics, but performance with SSE2 would suck, as LLVM falls back to calling libm. Would that really work now? Last time I checked, calls out to libm from the jitted

Re: [Mesa-dev] [PATCH 2/3] gallivm: Prefer backend agnostic intrinsic for rounding.

2016-04-03 Thread Roland Scheidegger
Am 03.04.2016 um 12:24 schrieb Jose Fonseca: > We could unconditionally use these instrinsics, but performance with SSE2 > would suck, as LLVM falls back to calling libm. Would that really work now? Last time I checked, calls out to libm from the jitted code weren't just slow, they simply crashed.

[Mesa-dev] [PATCH 2/3] gallivm: Prefer backend agnostic intrinsic for rounding.

2016-04-03 Thread Jose Fonseca
We could unconditionally use these instrinsics, but performance with SSE2 would suck, as LLVM falls back to calling libm. lp_test_arit. --- src/gallium/auxiliary/gallivm/lp_bld_arit.c | 46 - 1 file changed, 39 insertions(+), 7 deletions(-) diff --git