radeonsi/nir: lower load constants to scalar

We call nir_lower_load_const_to_scalar in the state trackers linker
however some later passes can reintroduce constant vectors. Here
we lower these to scalar and perform optimisations. The Intel
drivers do a similar call in their backend..

shader-db results VEGA 64:

Totals from affected shaders:
SGPRS: 152168 -> 151976 (-0.13 %)
VGPRS: 135224 -> 135112 (-0.08 %)
Spilled SGPRs: 4027 -> 4163 (3.38 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 10670028 -> 10654776 (-0.14 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 13122 -> 13135 (0.10 %)
Wait states: 0 -> 0 (0.00 %)

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

diff --git a/src/gallium/drivers/radeonsi/si_shader_nir.c b/src/gallium/drivers/radeonsi/si_shader_nir.c
index 85cf53848e66cd5ba807445a9c7a8df5bbd4f408..3bf30c724363b29a75b6c760e071d95cd00005ea 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_nir.c
+++ b/src/gallium/drivers/radeonsi/si_shader_nir.c
@@ -986,6 +986,11 @@ void si_lower_nir(struct si_shader_selector *sel)
        };
        NIR_PASS_V(sel->nir, nir_lower_subgroups, &subgroups_options);
 
+       /* Lower load constants to scalar and then clean up the mess */
+       NIR_PASS_V(sel->nir, nir_lower_load_const_to_scalar);
+       NIR_PASS_V(sel->nir, nir_lower_var_copies);
+       si_nir_opts(sel->nir);
+
        /* Lower large variables that are always constant with load_constant
         * intrinsics, which get turned into PC-relative loads from a data
         * section next to the shader.