intel/fs: Add the URB fence message

When they re-arranged all the dataport stuff and added the LSC, doing
URB fencing through the dataport no longer makes sense.  Instead, there
is now a fence message on the URB shared function.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Tested-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13092>

diff --git a/src/intel/compiler/brw_disasm.c b/src/intel/compiler/brw_disasm.c
index 1ce0ea2..acbe9d6 100644 (file)
--- a/src/intel/compiler/brw_disasm.c
+++ b/src/intel/compiler/brw_disasm.c
@@ -562,7 +562,8 @@ static const char *const gfx7_urb_opcode[] = {
    [GFX8_URB_OPCODE_ATOMIC_ADD] = "atomic add",  /* Gfx8+ */
    [GFX8_URB_OPCODE_SIMD8_WRITE] = "SIMD8 write", /* Gfx8+ */
    [GFX8_URB_OPCODE_SIMD8_READ] = "SIMD8 read",  /* Gfx8+ */
-   /* [9-15] - reserved */
+   [GFX125_URB_OPCODE_FENCE] = "fence",  /* Gfx12.5+ */
+   /* [10-15] - reserved */
 };
 
 static const char *const urb_swizzle[4] = {
@@ -2189,7 +2190,7 @@ brw_disassemble_inst(FILE *file, const struct intel_device_info *devinfo,
                 opcode == GFX8_URB_OPCODE_SIMD8_READ) {
                if (brw_inst_urb_channel_mask_present(devinfo, inst))
                   string(file, " masked");
-            } else {
+            } else if (opcode != GFX125_URB_OPCODE_FENCE) {
                err |= control(file, "urb swizzle", urb_swizzle,
                               brw_inst_urb_swizzle_control(devinfo, inst),
                               &space);

diff --git a/src/intel/compiler/brw_eu.h b/src/intel/compiler/brw_eu.h
index 995e6d8..0425625 100644 (file)
--- a/src/intel/compiler/brw_eu.h
+++ b/src/intel/compiler/brw_eu.h
@@ -375,6 +375,13 @@ brw_urb_desc_msg_type(ASSERTED const struct intel_device_info *devinfo,
    return GET_BITS(desc, 3, 0);
 }
 
+static inline uint32_t
+brw_urb_fence_desc(const struct intel_device_info *devinfo)
+{
+   assert(devinfo->has_lsc);
+   return brw_urb_desc(devinfo, GFX125_URB_OPCODE_FENCE, false, false, 0);
+}
+
 /**
  * Construct a message descriptor immediate with the specified sampler
  * function controls.

diff --git a/src/intel/compiler/brw_eu_defines.h b/src/intel/compiler/brw_eu_defines.h
index 03263e9..1732abb 100644 (file)
--- a/src/intel/compiler/brw_eu_defines.h
+++ b/src/intel/compiler/brw_eu_defines.h
@@ -1608,6 +1608,7 @@ enum brw_message_target {
 #define GFX8_URB_OPCODE_ATOMIC_ADD  6
 #define GFX8_URB_OPCODE_SIMD8_WRITE 7
 #define GFX8_URB_OPCODE_SIMD8_READ  8
+#define GFX125_URB_OPCODE_FENCE     9
 
 #define BRW_URB_SWIZZLE_NONE          0
 #define BRW_URB_SWIZZLE_INTERLEAVE    1

diff --git a/src/intel/compiler/brw_eu_emit.c b/src/intel/compiler/brw_eu_emit.c
index ad276e6..a8e19cb 100644 (file)
--- a/src/intel/compiler/brw_eu_emit.c
+++ b/src/intel/compiler/brw_eu_emit.c
@@ -3255,17 +3255,22 @@ gfx12_set_memory_fence_message(struct brw_codegen *p,
 
    brw_inst_set_sfid(p->devinfo, insn, sfid);
 
-   enum lsc_fence_scope scope = LSC_FENCE_THREADGROUP;
-   enum lsc_flush_type flush_type = LSC_FLUSH_TYPE_NONE;
+   if (sfid == BRW_SFID_URB) {
+      brw_set_desc(p, insn, brw_urb_fence_desc(p->devinfo) |
+                            brw_message_desc(p->devinfo, mlen, rlen, false));
+   } else {
+      enum lsc_fence_scope scope = LSC_FENCE_THREADGROUP;
+      enum lsc_flush_type flush_type = LSC_FLUSH_TYPE_NONE;
 
-   if (sfid == GFX12_SFID_TGM) {
-      scope = LSC_FENCE_TILE;
-      flush_type = LSC_FLUSH_TYPE_EVICT;
-   }
+      if (sfid == GFX12_SFID_TGM) {
+         scope = LSC_FENCE_TILE;
+         flush_type = LSC_FLUSH_TYPE_EVICT;
+      }
 
-   brw_set_desc(p, insn, lsc_fence_msg_desc(p->devinfo, scope,
-                                            flush_type, false) |
-                         brw_message_desc(p->devinfo, mlen, rlen, false));
+      brw_set_desc(p, insn, lsc_fence_msg_desc(p->devinfo, scope,
+                                               flush_type, false) |
+                            brw_message_desc(p->devinfo, mlen, rlen, false));
+   }
 }
 
 void

diff --git a/src/intel/compiler/brw_ir_performance.cpp b/src/intel/compiler/brw_ir_performance.cpp
index f04694f..2a7dc78 100644 (file)
--- a/src/intel/compiler/brw_ir_performance.cpp
+++ b/src/intel/compiler/brw_ir_performance.cpp
@@ -946,6 +946,7 @@ namespace {
             else
                abort();
 
+         case BRW_SFID_URB:
          case GFX7_SFID_DATAPORT_DATA_CACHE:
          case GFX12_SFID_SLM:
          case GFX12_SFID_TGM: