What is the value of making packets bigger than the SIMD width? I read through the following slides but the reason is not obvious to me.
colorblue
@chuangxuean The basic idea is that you amortize the computation over the the larger packet size. I found online this excerpt on one of the main reason to make the packets bigger in ray packets "typically by performing some conservative full-packet rejection tests based on bounding frusta, bounding rays, or interval arithmetic. All of these techniques rely on high ray coherence to deliver benefits over SIMD packet tracing."
From what I understand the basic idea is that you want to make packets larger for higher level of BVH. Then as you go down in utilization you want to go more towards the SIMD approach.
What is the value of making packets bigger than the SIMD width? I read through the following slides but the reason is not obvious to me.
@chuangxuean The basic idea is that you amortize the computation over the the larger packet size. I found online this excerpt on one of the main reason to make the packets bigger in ray packets "typically by performing some conservative full-packet rejection tests based on bounding frusta, bounding rays, or interval arithmetic. All of these techniques rely on high ray coherence to deliver benefits over SIMD packet tracing." From what I understand the basic idea is that you want to make packets larger for higher level of BVH. Then as you go down in utilization you want to go more towards the SIMD approach.