pan/mdg: Add pack_unorm_4x8 via 8-bit

More efficient than the 32-bit version in NIR.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5265>

src/panfrost/midgard/midgard_compile.c

@@ -383,6 +383,8 @@ optimise_nir(nir_shader *nir, unsigned quirks)
         if (quirks & MIDGARD_BROKEN_LOD)
                 NIR_PASS_V(nir, midgard_nir_lod_errata);
 
+        NIR_PASS(progress, nir, midgard_nir_lower_algebraic_early);
+
         do {
                 progress = false;
 

src/panfrost/midgard/midgard_nir.h

@@ -1,6 +1,7 @@
 #include <stdbool.h>
 #include "nir.h"
 
+bool midgard_nir_lower_algebraic_early(nir_shader *shader);
 bool midgard_nir_lower_algebraic_late(nir_shader *shader);
 bool midgard_nir_scale_trig(nir_shader *shader);
 bool midgard_nir_cancel_inot(nir_shader *shader);

src/panfrost/midgard/midgard_nir_algebraic.py

@@ -30,6 +30,10 @@ a = 'a'
 b = 'b'
 c = 'c'
 
+algebraic = [
+   (('pack_unorm_4x8', a), ('pack_32_4x8', ('f2u8', ('fround_even', ('fmul', ('fsat', a), 255.0)))))
+]
+
 algebraic_late = [
     # ineg must be lowered late, but only for integers; floats will try to
     # have modifiers attached... hence why this has to be here rather than
@@ -141,6 +145,9 @@ def run():
 
     print('#include "midgard_nir.h"')
 
+    print(nir_algebraic.AlgebraicPass("midgard_nir_lower_algebraic_early",
+                                      algebraic).render())
+
     print(nir_algebraic.AlgebraicPass("midgard_nir_lower_algebraic_late",
                                       algebraic_late + converts + constant_switch).render())