While it's discussed a bit later on, I think what's interesting is that the shader code, which is somewhat reminiscent of RISC-V and instruction commands, clearly would invoke some form of stalls in the mul, 2x madd, and clmp instructions, since all of them are writing to r3. It be interesting to see how compile time reorganization of the order of these instructions might help.
While it's discussed a bit later on, I think what's interesting is that the shader code, which is somewhat reminiscent of RISC-V and instruction commands, clearly would invoke some form of stalls in the mul, 2x madd, and clmp instructions, since all of them are writing to r3. It be interesting to see how compile time reorganization of the order of these instructions might help.