Abstract: Techniques are disclosed relating to graphics processor that support ray tracing. In particular, shader circuitry may be configured to adjust a scheduling priority of a single-instruction multiple-data (SIMD) group of a shader program based on a hint that the SIMD group has an upcoming ray intersect command for ray intersect accelerator circuitry and based on a resource usage indication from the ray intersect accelerator circuitry. This may advantageously reduce cache thrashing, e.g., when shaders may allocate memory for ray intersect commands and fill a shared cache faster than the ray intersect accelerator circuitry can process the rays.

Abstract: Techniques are disclosed relating to graphics processor that support ray tracing. In particular, shader circuitry may be configured to adjust a scheduling priority of a single-instruction multiple-data (SIMD) group of a shader program based on a hint that the SIMD group has an upcoming ray intersect command for ray intersect accelerator circuitry and based on a resource usage indication from the ray intersect accelerator circuitry. This may advantageously reduce cache thrashing, e.g., when shaders may allocate memory for ray intersect commands and fill a shared cache faster than the ray intersect accelerator circuitry can process the rays.