Missing intrinsics required for building Cycles

[jserv]

Cycles is Blender's physically-based path tracer for production rendering. Recently, Apple Inc. sent Aarch64 patch to Blender developer as shown on D8237: Cycles: support neon instructions for arm64 processors. However, file intern/cycles/util/util_sse_to_neon.h provided by D8237 was incomplete, and Blender developers were discussing the integration of sse2neon.

Missing intrinsics in SSE2NEON required for building Cycles:

• [x] _mm_cmpnle_ps
• [x] _mm_cmpnlt_ps
• [x] _mm_fmadd_ps (FMA)
• [x] _mm_fmsub_ps (FMA)
• [ ] _mm_fnmadd_ps (FMA)
• [ ] _mm_fnmsub_ps (FMA)
• [x] _mm_floor_ps
• [x] _mm_ceil_ps
• [x] _mm_castps_pd
• [x] _mm_stream_ps
• [x] _mm_blendv_ps
• [x] _mm_dp_ps
• [x] _mm_stream_load_si128

D8237 also defines _mm_cmple_epi32 and _mm_cmpge_epi32, which are not part of SSE intrinsics.

Fortunately, the simplified implementations are available in D8237.

[jserv]

Only one instruction is missing for building Cycles. That is _mm_castps_pd as described in #91.