[Mesa-dev] [PATCH 3/3] i965/nir: use vectorization for non-scalar stages

[Ian Romanick]

On 10/17/2018 11:33 AM, Jason Ekstrand wrote:
> From: Connor Abbott <cwabbott0 at gmail.com>
> 
> Shader-db results on Haswell:
> 
>     total instructions in shared programs: 2180337 -> 2154080 (-1.20%)
>     instructions in affected programs: 959766 -> 933509 (-2.74%)
>     helped: 5653
>     HURT: 2560
> 
>     total cycles in shared programs: 12339326 -> 12307102 (-0.26%)
>     cycles in affected programs: 6102794 -> 6070570 (-0.53%)
>     helped: 3838
>     HURT: 4868

Here's the results I got with these 3 patches on 322a919a41f:

total instructions in shared programs: 13674046 -> 13643001 (-0.23%)
instructions in affected programs: 1248672 -> 1217627 (-2.49%)
helped: 7168
HURT: 2841
helped stats (abs) min: 1 max: 39 x̄: 5.40 x̃: 3
helped stats (rel) min: 0.21% max: 33.33% x̄: 4.55% x̃: 3.54%
HURT stats (abs)   min: 1 max: 21 x̄: 2.71 x̃: 3
HURT stats (rel)   min: 0.19% max: 22.73% x̄: 3.86% x̃: 3.53%
95% mean confidence interval for instructions value: -3.23 -2.97
95% mean confidence interval for instructions %-change: -2.28% -2.05%
Instructions are helped.

total cycles in shared programs: 373694400 -> 373745788 (0.01%)
cycles in affected programs: 23171532 -> 23222920 (0.22%)
helped: 4890
HURT: 5632
helped stats (abs) min: 2 max: 1268 x̄: 52.04 x̃: 34
helped stats (rel) min: 0.04% max: 45.71% x̄: 7.43% x̃: 4.64%
HURT stats (abs)   min: 2 max: 6042 x̄: 54.30 x̃: 32
HURT stats (rel)   min: 0.05% max: 60.66% x̄: 8.19% x̃: 6.21%
95% mean confidence interval for cycles value: 1.30 8.47
95% mean confidence interval for cycles %-change: 0.73% 1.14%
Cycles are HURT.

total spills in shared programs: 82569 -> 82572 (<.01%)
spills in affected programs: 70 -> 73 (4.29%)
helped: 0
HURT: 3

total fills in shared programs: 93445 -> 93449 (<.01%)
fills in affected programs: 71 -> 75 (5.63%)
helped: 0
HURT: 4

This is pretty different from your result... and not good. :(  What SHA
of master were you on?

> Most of the hurt programs seem to be because we generate extra MOV's due
> to vectorizing things. For example, in
> shaders/non-free/steam/anomaly-2/158.shader_test, this:
> 
> add(8)          g116<1>.xyF     g12<4,4,1>.xyyyF g1.4<0,4,1>.xyyyF { align16 NoDDClr 1Q };
> add(8)          g117<1>.xyF     g12<4,4,1>.xyyyF g1.4<0,4,1>.zwwwF { align16 NoDDClr 1Q };
> add(8)          g116<1>.zwF     g12<4,4,1>.xxxyF -g1.4<0,4,1>.xxxyF { align16 NoDDChk 1Q };
> add(8)          g117<1>.zwF     g12<4,4,1>.xxxyF -g1.4<0,4,1>.zzzwF { align16 NoDDChk 1Q };
> 
> Turns into this:
> 
> add(8)          g13<1>F         g12<4,4,1>.xyxyF g1.4<0,4,1>F   { align16 1Q };
> add(8)          g14<1>F         g12<4,4,1>.xyxyF -g1.4<0,4,1>F  { align16 1Q };
> mov(8)          g116<1>.xyD     g13<4,4,1>.xyyyD                { align16 NoDDClr 1Q };
> mov(8)          g117<1>.xyD     g13<4,4,1>.zwwwD                { align16 NoDDClr 1Q };
> mov(8)          g116<1>.zwD     g14<4,4,1>.xxxyD                { align16 NoDDChk 1Q };
> mov(8)          g117<1>.zwD     g14<4,4,1>.zzzwD                { align16 NoDDChk 1Q };
> 
> So we eliminated two add's, but then had to introduce four mov's to
> transpose the result.  Some of the hurt is because vectorization is a bit
> over-aggressive and we vectorize something when we should have left it
> as a scalar and CSEd it.  Unfortunately, this is all really tricky to do
> as it involves the interactions between many different components.
> ---
>  src/intel/compiler/brw_nir.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
> index 297845b89b7..564fd004a94 100644
> --- a/src/intel/compiler/brw_nir.c
> +++ b/src/intel/compiler/brw_nir.c
> @@ -568,6 +568,12 @@ brw_nir_optimize(nir_shader *nir, const struct brw_compiler *compiler,
>        OPT(nir_copy_prop);
>        OPT(nir_opt_dce);
>        OPT(nir_opt_cse);
> +
> +      if (!is_scalar) {
> +         OPT(nir_opt_vectorize);
> +         OPT(nir_copy_prop);
> +      }
> +
>        OPT(nir_opt_peephole_select, 0);
>        OPT(nir_opt_intrinsics);
>        OPT(nir_opt_algebraic);
>