pan/bi: Add BI_TABLE for fast table accesses

Used to implement SPECIAL ops. Separate class since they are faster
which means you can pair them with actual work on FMA.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4615>

src/panfrost/bifrost/bi_pack.c

@@ -1140,6 +1140,7 @@ bi_pack_add(bi_clause *clause, bi_bundle bundle, struct bi_registers *regs)
                 return bi_pack_add_st_vary(clause, bundle.add, regs);
         case BI_SPECIAL:
                 return bi_pack_add_special(bundle.add, regs);
+        case BI_TABLE:
         case BI_SWIZZLE:
         case BI_TEX:
         case BI_ROUND:

src/panfrost/bifrost/bi_print.c

@@ -152,6 +152,7 @@ bi_class_name(enum bi_class cl)
         case BI_STORE_VAR: return "store_var";
         case BI_SPECIAL: return "special";
         case BI_SWIZZLE: return "swizzle";
+        case BI_TABLE: return "table";
         case BI_TEX: return "tex";
         case BI_ROUND: return "round";
         default: return "unknown_class";
@@ -251,6 +252,15 @@ bi_bitwise_op_name(enum bi_bitwise_op op)
         }
 }
 
+const char *
+bi_table_op_name(enum bi_table_op op)
+{
+        switch (op) {
+        case BI_TABLE_LOG2_U_OVER_U_1_LOW: return "log2.help";
+        default: return "invalid";
+        }
+}
+
 const char *
 bi_special_op_name(enum bi_special_op op)
 {
@@ -326,6 +336,8 @@ bi_print_instruction(bi_instruction *ins, FILE *fp)
                 fprintf(fp, ins->op.round == BI_ROUND_MODE ? "roundMode": "round");
         else if (ins->type == BI_SPECIAL)
                 fprintf(fp, "%s", bi_special_op_name(ins->op.special));
+        else if (ins->type == BI_TABLE)
+                fprintf(fp, "%s", bi_table_op_name(ins->op.table));
         else if (ins->type == BI_CMP)
                 fprintf(fp, "%s", bi_cond_name(ins->op.compare));
         else

src/panfrost/bifrost/bi_print.h

@@ -41,6 +41,7 @@ const char * bi_ldst_type_name(enum bifrost_ldst_type type);
 const char * bi_class_name(enum bi_class cl);
 const char * bi_cond_name(enum bi_cond cond);
 const char * bi_special_op_name(enum bi_special_op op);
+const char * bi_table_op_name(enum bi_table_op op);
 
 void bi_print_instruction(bi_instruction *ins, FILE *fp);
 void bi_print_bundle(bi_bundle *bundle, FILE *fp);

src/panfrost/bifrost/bi_tables.c

@@ -52,6 +52,7 @@ unsigned bi_class_props[BI_NUM_CLASSES] = {
         [BI_STORE] 		= BI_SCHED_HI_LATENCY | BI_SCHED_ADD | BI_VECTOR | BI_DATA_REG_SRC,
         [BI_STORE_VAR] 		= BI_SCHED_HI_LATENCY | BI_SCHED_ADD | BI_VECTOR | BI_DATA_REG_SRC,
         [BI_SPECIAL] 		= BI_SCHED_ADD | BI_SCHED_SLOW,
+        [BI_TABLE]              = BI_SCHED_ADD,
         [BI_SWIZZLE]            = BI_SCHED_ALL | BI_SWIZZLABLE,
         [BI_TEX] 		= BI_SCHED_HI_LATENCY | BI_SCHED_ADD | BI_VECTOR,
         [BI_ROUND] 		= BI_GENERIC | BI_ROUNDMODE | BI_SCHED_ALL,

src/panfrost/bifrost/compiler.h

@@ -71,8 +71,9 @@ enum bi_class {
         BI_SHIFT,
         BI_STORE,
         BI_STORE_VAR,
-        BI_SPECIAL, /* _FAST, _TABLE on supported GPUs */
+        BI_SPECIAL, /* _FAST on supported GPUs */
         BI_SWIZZLE,
+        BI_TABLE,
         BI_TEX,
         BI_ROUND,
         BI_NUM_CLASSES
@@ -182,6 +183,15 @@ enum bi_round_op {
         BI_ROUND_ROUND /* i.e.: fround() */
 };
 
+enum bi_table_op {
+        /* fp32 log2() with low precision, suitable for GL or half_log2() in
+         * CL. In the first argument, takes x. Letting u be such that x =
+         * 2^{-m} u with m integer and 0.75 <= u < 1.5, returns
+         * log2(u) / (u - 1). */
+
+        BI_TABLE_LOG2_U_OVER_U_1_LOW,
+};
+
 enum bi_special_op {
         BI_SPECIAL_FRCP,
         BI_SPECIAL_FRSQ,
@@ -244,6 +254,7 @@ typedef struct {
                 enum bi_bitwise_op bitwise;
                 enum bi_round_op round;
                 enum bi_special_op special;
+                enum bi_table_op table;
                 enum bi_cond compare;
         } op;