From 28e79a42c221e858afb4ebf67ba72a3ccf36dd9f Mon Sep 17 00:00:00 2001
From: Hari Thangirala <hari.thangirala@amd.com>
Date: Sat, 7 Apr 2018 17:33:20 -0500
Subject: [PATCH] Allow HSA_ENABLE_SDMA to override runtime defaults.

Change-Id: I2305304228010157bfb589c365f4a998577231cd


[ROCm/ROCR-Runtime commit: 3e0cd85d6978625e3a06fc62da618d9f9029574a]
---
 .../hsa-runtime/core/runtime/amd_gpu_agent.cpp     | 14 ++++++++------
 .../runtime/hsa-runtime/core/util/flag.h           | 10 +++++-----
 2 files changed, 13 insertions(+), 11 deletions(-)

diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
index 6421eb06e7..cceda10c6f 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
@@ -558,14 +558,16 @@ void GpuAgent::InitDma() {
   // Share utility queue with device-to-host blits.
   queues_[QueueUtility].reset(queue_lambda);
 
-  // Blits, try create SDMA blit first.
-  // Disable SDMA on specific ISA targets until they are fully qualified.
+  // Decide which engine to use for blits.
   auto blit_lambda = [this](bool h2d, lazy_ptr<core::Queue>& queue) {
-    if ((isa_->GetMajorVersion() != 8) && core::Runtime::runtime_singleton_->flag().enable_sdma() &&
-        (HSA_PROFILE_BASE == profile_)) {
-      auto ret = CreateBlitSdma(h2d);
-      if (ret != nullptr) return ret;
+    std::string sdma_override = core::Runtime::runtime_singleton_->flag().enable_sdma();
+    bool use_sdma = (sdma_override.size() == 0) ? (isa_->GetMajorVersion() != 8) : (sdma_override == "1");
+
+    if (use_sdma && (HSA_PROFILE_BASE == profile_)) {
+        auto ret = CreateBlitSdma(h2d);
+        if (ret != nullptr) return ret;
     }
+
     auto ret = CreateBlitKernel((*queue).get());
     if (ret == nullptr)
       throw AMD::hsa_exception(HSA_STATUS_ERROR_OUT_OF_RESOURCES, "Blit creation failed.");
diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/util/flag.h b/projects/rocr-runtime/runtime/hsa-runtime/core/util/flag.h
index e9049f36b0..7593bf36e2 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/util/flag.h
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/util/flag.h
@@ -69,8 +69,7 @@ class Flag {
     var = os::GetEnvVar("HSA_ENABLE_INTERRUPT");
     enable_interrupt_ = (var == "0") ? false : true;
 
-    var = os::GetEnvVar("HSA_ENABLE_SDMA");
-    enable_sdma_ = (var == "0") ? false : true;
+    enable_sdma_ = os::GetEnvVar("HSA_ENABLE_SDMA");
 
     var = os::GetEnvVar("HSA_RUNNING_UNDER_VALGRIND");
     running_valgrind_ = (var == "1") ? true : false;
@@ -104,8 +103,6 @@ class Flag {
 
   bool enable_interrupt() const { return enable_interrupt_; }
 
-  bool enable_sdma() const { return enable_sdma_; }
-
   bool enable_sdma_hdp_flush() const { return enable_sdma_hdp_flush_; }
 
   bool running_valgrind() const { return running_valgrind_; }
@@ -116,6 +113,8 @@ class Flag {
 
   bool disable_fragment_alloc() const { return disable_fragment_alloc_; }
 
+  std::string enable_sdma() const { return enable_sdma_; }
+
   uint32_t max_queues() const { return max_queues_; }
 
   size_t scratch_mem_size() const { return scratch_mem_size_; }
@@ -126,7 +125,6 @@ class Flag {
   bool check_flat_scratch_;
   bool enable_vm_fault_message_;
   bool enable_interrupt_;
-  bool enable_sdma_;
   bool enable_sdma_hdp_flush_;
   bool running_valgrind_;
   bool sdma_wait_idle_;
@@ -134,6 +132,8 @@ class Flag {
   bool report_tool_load_failures_;
   bool disable_fragment_alloc_;
 
+  std::string enable_sdma_;
+
   uint32_t max_queues_;
 
   size_t scratch_mem_size_;