From c788f2cfa2bbe83bc7cbf0b2e8e6903c3d8d9436 Mon Sep 17 00:00:00 2001 From: lkcl Date: Sun, 3 Oct 2021 15:36:39 +0100 Subject: [PATCH] --- 3d_gpu/architecture/dynamic_simd.mdwn | 14 ++++++++++++++ 1 file changed, 14 insertions(+) diff --git a/3d_gpu/architecture/dynamic_simd.mdwn b/3d_gpu/architecture/dynamic_simd.mdwn index 7680b0911..f333f5911 100644 --- a/3d_gpu/architecture/dynamic_simd.mdwn +++ b/3d_gpu/architecture/dynamic_simd.mdwn @@ -61,6 +61,20 @@ constructs in full OO fashion, yet the high-level dsl.Module language concepts remain true to their intended characteristics and behaviour and need neither duplication nor alteration. +Typical use-cases are, just as is the driving force behind PartitionedSignal, +the sharing at the gate level of arithmetic primitives that would +otherwise require costly duplication in a final product, or be too +complex or costly to develop without such abstraction at the higher +Type 2 (dsl.Module) level. + +Examples include an ALU that may dynamically +at runtime switch between Complex (imaginary) arithmetic +and Real (scalar) arithmetic, whilst *through the same +datapath* automatically sharing the arithmetic logic +gates between the two abstract concepts. Given that +64 bit multipliers are 12,000 to 15,000 gates the savings +can be enormous. + # Rationale / Introduction The Dynamic Partitioned SIMD Signal is effectively a parallelisation -- 2.30.2