nir,nak: Add a nir_op_prmt_nv

We have this in hardware since forever and it's really useful.  May as
well add it to NIR so we can use it in various lowerings.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30218>

src/compiler/nir/nir_opcodes.py

@@ -1371,6 +1371,19 @@ binop_convert("interleave_agx", tuint32, tuint16, "", """
       be used as-is for Morton encoding.
       """)
 
+# NVIDIA PRMT
+opcode("prmt_nv", 0, tuint32, [0, 0, 0], [tuint32, tuint32, tuint32],
+       False, "", """
+    dst = 0;
+    for (unsigned i = 0; i < 4; i++) {
+        uint8_t byte = (src0 >> (i * 4)) & 0x7;
+        uint8_t x = byte < 4 ? (src1 >> (byte * 8))
+                             : (src2 >> ((byte - 4) * 8));
+        if ((src0 >> (i * 4)) & 0x8)
+            x = ((int8_t)x) >> 7;
+        dst |= ((uint32_t)x) << i * 8;
+    }""")
+
 # 24b multiply into 32b result (with sign extension)
 binop("imul24", tint32, _2src_commutative + associative,
       "(((int32_t)src0 << 8) >> 8) * (((int32_t)src1 << 8) >> 8)")

src/nouveau/compiler/nak/from_nir.rs

@@ -1428,6 +1428,16 @@ impl<'a> ShaderFromNir<'a> {
 
                 b.prmt(low.into(), high.into(), [0, 1, 4, 5])
             }
+            nir_op_prmt_nv => {
+                let dst = b.alloc_ssa(RegFile::GPR, 1);
+                b.push_op(OpPrmt {
+                    dst: dst.into(),
+                    srcs: [srcs[1], srcs[2]],
+                    sel: srcs[0],
+                    mode: PrmtMode::Index,
+                });
+                dst
+            }
             nir_op_sdot_4x8_iadd => {
                 let dst = b.alloc_ssa(RegFile::GPR, 1);
                 b.push_op(OpIDp4 {