Implement code cache (SQC I$, SQC K$, TCP, GL1, GL2) invalidation for gfx10

Change-Id: I8b2a59118094fbb55e3f575fa9f79959d3725d7d


[ROCm/ROCR-Runtime commit: 5b64fbd0e5]

projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_gpu_pm4.h

@@ -72,6 +72,12 @@
 #  define PM4_ACQUIRE_MEM_COHER_CNTL_SH_ICACHE_ACTION_ENA  (1 << 29)
 #define PM4_ACQUIRE_MEM_DW2_COHER_SIZE(x)                  (((x) & 0xFFFFFFFF) << 0)
 #define PM4_ACQUIRE_MEM_DW3_COHER_SIZE_HI(x)               (((x) & 0xFF) << 0)
+#define PM4_ACQUIRE_MEM_DW7_GCR_CNTL(x)                    (((x) & 0x7FFFF) << 0)
+#  define PM4_ACQUIRE_MEM_GCR_CNTL_GLI_INV(x)              (((x) & 0x3) << 0)
+#  define PM4_ACQUIRE_MEM_GCR_CNTL_GLK_INV                 (1 << 7)
+#  define PM4_ACQUIRE_MEM_GCR_CNTL_GLV_INV                 (1 << 8)
+#  define PM4_ACQUIRE_MEM_GCR_CNTL_GL1_INV                 (1 << 9)
+#  define PM4_ACQUIRE_MEM_GCR_CNTL_GL2_INV                 (1 << 14)
 
 #define PM4_RELEASE_MEM_DW1_EVENT_INDEX(x)                 (((x) & 0xF) << 8)
 #  define PM4_RELEASE_MEM_EVENT_INDEX_AQL                  0x7

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp

@@ -1203,29 +1203,40 @@ void GpuAgent::InvalidateCodeCaches() {
       // Microcode is handling code cache invalidation.
       return;
     }
-  } else if (isa_->GetMajorVersion() > 9) {
+  } else if (isa_->GetMajorVersion() > 10) {
     assert(false && "Code cache invalidation not implemented for this agent");
   }
 
   // Invalidate caches which may hold lines of code object allocation.
-  constexpr uint32_t cache_inv_size_dw = 7;
-  uint32_t cache_inv[cache_inv_size_dw];
+  uint32_t cache_inv[8] = {0};
+  uint32_t cache_inv_size_dw;
+
+  if (isa_->GetMajorVersion() < 10) {
+      cache_inv[1] = PM4_ACQUIRE_MEM_DW1_COHER_CNTL(
+          PM4_ACQUIRE_MEM_COHER_CNTL_SH_ICACHE_ACTION_ENA |
+          PM4_ACQUIRE_MEM_COHER_CNTL_SH_KCACHE_ACTION_ENA |
+          PM4_ACQUIRE_MEM_COHER_CNTL_TC_ACTION_ENA |
+          PM4_ACQUIRE_MEM_COHER_CNTL_TC_WB_ACTION_ENA);
+
+      cache_inv_size_dw = 7;
+  } else {
+      cache_inv[7] = PM4_ACQUIRE_MEM_DW7_GCR_CNTL(
+          PM4_ACQUIRE_MEM_GCR_CNTL_GLI_INV(1) |
+          PM4_ACQUIRE_MEM_GCR_CNTL_GLK_INV |
+          PM4_ACQUIRE_MEM_GCR_CNTL_GLV_INV |
+          PM4_ACQUIRE_MEM_GCR_CNTL_GL1_INV |
+          PM4_ACQUIRE_MEM_GCR_CNTL_GL2_INV);
+
+      cache_inv_size_dw = 8;
+  }
 
   cache_inv[0] = PM4_HDR(PM4_HDR_IT_OPCODE_ACQUIRE_MEM, cache_inv_size_dw,
-                         isa_->GetMajorVersion());
-  cache_inv[1] = PM4_ACQUIRE_MEM_DW1_COHER_CNTL(
-      PM4_ACQUIRE_MEM_COHER_CNTL_SH_ICACHE_ACTION_ENA |
-      PM4_ACQUIRE_MEM_COHER_CNTL_SH_KCACHE_ACTION_ENA |
-      PM4_ACQUIRE_MEM_COHER_CNTL_TC_ACTION_ENA |
-      PM4_ACQUIRE_MEM_COHER_CNTL_TC_WB_ACTION_ENA);
+             isa_->GetMajorVersion());
   cache_inv[2] = PM4_ACQUIRE_MEM_DW2_COHER_SIZE(0xFFFFFFFF);
   cache_inv[3] = PM4_ACQUIRE_MEM_DW3_COHER_SIZE_HI(0xFF);
-  cache_inv[4] = 0;
-  cache_inv[5] = 0;
-  cache_inv[6] = 0;
 
   // Submit the command to the utility queue and wait for it to complete.
-  queues_[QueueUtility]->ExecutePM4(cache_inv, sizeof(cache_inv));
+  queues_[QueueUtility]->ExecutePM4(cache_inv, cache_inv_size_dw * sizeof(uint32_t));
 }
 
 lazy_ptr<core::Blit>& GpuAgent::GetXgmiBlit(const core::Agent& dst_agent) {