intel/mi-builder: add framework for self modifying batches

v2: Use Jason's idea to store addresses to modify

v3: Add ALU flushes (Jason)

v4: Remove ALU flush from gen_mi_self_mod_barrier() (Jason)

Signed-off-by: Lionel Landwerlin <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]> (v2)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2775>

1 files changed, 69 insertions, 0 deletions

diff --git a/src/intel/common/gen_mi_builder.h b/src/intel/common/gen_mi_builder.h
index bcd685754a3..5f3f4e28c1f 100644
--- a/src/intel/common/gen_mi_builder.h
+++ b/src/intel/common/gen_mi_builder.h
@@ -24,6 +24,7 @@
 #ifndef GEN_MI_BUILDER_H
 #define GEN_MI_BUILDER_H
 
+#include "genxml/genX_bits.h"
 #include "util/bitscan.h"
 #include "util/fast_idiv_by_const.h"
 #include "util/u_math.h"
@@ -45,6 +46,18 @@
  * __gen_address_type
  * __gen_address_offset(__gen_address_type addr, uint64_t offset);
  *
+ *
+ * If self-modifying batches are supported, we must be able to pass batch
+ * addresses around as void*s so pinning as well as batch chaining or some
+ * other mechanism for ensuring batch pointers remain valid during building is
+ * required. The following function must also be defined, it returns an
+ * address in canonical form:
+ *
+ * uint64_t
+ * __gen_get_batch_address(__gen_user_data *user_data, void *location);
+ *
+ * Also, __gen_combine_address must accept a location value of NULL and return
+ * a fully valid 64-bit address.
  */
 
 /*
@@ -831,4 +844,60 @@ gen_mi_udiv32_imm(struct gen_mi_builder *b,
 
 #endif /* MI_MATH section */
 
+/* This assumes addresses of strictly more than 32bits (aka. Gen8+). */
+#if GEN_MI_BUILDER_CAN_WRITE_BATCH
+
+struct gen_mi_address_token {
+   /* Pointers to address memory fields in the batch. */
+   uint64_t *ptrs[2];
+};
+
+static inline struct gen_mi_address_token
+gen_mi_store_address(struct gen_mi_builder *b,
+                     struct gen_mi_value addr_reg)
+{
+   gen_mi_builder_flush_math(b);
+
+   assert(addr_reg.type == GEN_MI_VALUE_TYPE_REG64);
+
+   struct gen_mi_address_token token = {};
+
+   for (unsigned i = 0; i < 2; i++) {
+      gen_mi_builder_emit(b, GENX(MI_STORE_REGISTER_MEM), srm) {
+         srm.RegisterAddress = addr_reg.reg + (i * 4);
+
+         const unsigned addr_dw =
+            GENX(MI_STORE_REGISTER_MEM_MemoryAddress_start) / 8;
+         token.ptrs[i] = (void *)_dst + addr_dw;
+      }
+   }
+
+   gen_mi_value_unref(b, addr_reg);
+   return token;
+}
+
+static inline void
+gen_mi_self_mod_barrier(struct gen_mi_builder *b)
+{
+   /* Documentation says Gen11+ should be able to invalidate the command cache
+    * but experiment show it doesn't work properly, so for now just get over
+    * the CS prefetch.
+    */
+   for (uint32_t i = 0; i < 128; i++)
+      gen_mi_builder_emit(b, GENX(MI_NOOP), noop);
+}
+
+static inline void
+_gen_mi_resolve_address_token(struct gen_mi_builder *b,
+                              struct gen_mi_address_token token,
+                              void *batch_location)
+{
+   uint64_t addr_addr_u64 = __gen_get_batch_address(b->user_data,
+                                                    batch_location);
+   *(token.ptrs[0]) = addr_addr_u64;
+   *(token.ptrs[1]) = addr_addr_u64 + 4;
+}
+
+#endif /* GEN_MI_BUILDER_CAN_WRITE_BATCH */
+
 #endif /* GEN_MI_BUILDER_H */