PCS parser is aware of external correlation IDs (#639)

* PCS parser is aware of external correlation IDs

* source formatting (clang-format v11) (#640)

Co-authored-by: vlaindic <139573562+vlaindic@users.noreply.github.com>

---------

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

source/lib/rocprofiler-sdk/pc_sampling/parser/correlation.hpp

@@ -22,6 +22,8 @@
 
 #pragma once
 
+#include <rocprofiler-sdk/fwd.h>
+
 #include <cstdint>
 #include <iostream>
 #include <memory>
@@ -133,7 +135,7 @@ public:
      * Given a device dev, doorbell and and wrapped dispatch_id,
      * @returns the correlation_id set by dispatch_pkt_id_t
      */
-    uint64_t get(device_handle dev, trap_correlation_id_t correlation_in)
+    rocprofiler_correlation_id_t get(device_handle dev, trap_correlation_id_t correlation_in)
     {
 #ifndef _PARSER_CORRELATION_DISABLE_CACHE
         if(dev.handle == cache_dev_id && correlation_in == cache_correlation_id_in)
@@ -166,12 +168,14 @@ public:
     }
 
 private:
-    std::unordered_map<DispatchPkt, uint64_t> dispatch_to_correlation{};
+    std::unordered_map<DispatchPkt, rocprofiler_correlation_id_t> dispatch_to_correlation{};
 
     // Making get() const and these cache variables mutable causes performance to be unstable
-    trap_correlation_id_t cache_correlation_id_in{.raw = ~0ul};  // Invalid value in cache
-    uint64_t              cache_correlation_id_out = ~0ul;
-    uint64_t              cache_dev_id             = ~0ul;  // Invalid device Id in cache
+    trap_correlation_id_t        cache_correlation_id_in{.raw = ~0ul};  // Invalid value in cache
+    rocprofiler_correlation_id_t cache_correlation_id_out{
+        .internal = ~0ul,
+        .external = rocprofiler_user_data_t{.value = ~0ul}};
+    uint64_t cache_dev_id = ~0ul;  // Invalid device Id in cache
 };
 }  // namespace Parser
 

source/lib/rocprofiler-sdk/pc_sampling/parser/parser_types.h

@@ -22,6 +22,8 @@
 
 #pragma once
 
+#include <rocprofiler-sdk/fwd.h>
+
 /**
  * ######## Parser Definitions ########
  */
@@ -130,9 +132,9 @@ typedef struct
     uint32_t workgroup_id_y;
     uint32_t workgroup_id_z;
 
-    uint32_t wave_count;
-    uint64_t timestamp;
-    uint64_t correlation_id;
+    uint32_t                     wave_count;
+    uint64_t                     timestamp;
+    rocprofiler_correlation_id_t correlation_id;
 
     pcsample_snapshot_v1_t snapshot;
 

source/lib/rocprofiler-sdk/pc_sampling/parser/pc_record_interface.cpp

@@ -61,7 +61,7 @@ PCSamplingParserContext::newDispatch(const dispatch_pkt_id_t& pkt)
 {
     std::unique_lock<std::shared_mutex> lock(mut);
     corr_map->newDispatch(pkt);
-    active_dispatches[pkt.correlation_id] = pkt;
+    active_dispatches[pkt.correlation_id.internal] = pkt;
 }
 
 void

source/lib/rocprofiler-sdk/pc_sampling/parser/pc_record_interface.hpp

@@ -148,6 +148,7 @@ protected:
     //! Data allocated to store samples. Temporary.
     std::vector<std::unique_ptr<PCSamplingData>> data;
     //! Dispatches not yet completed.
+    // Uses only the internal correlation_id.
     std::unordered_map<uint64_t, dispatch_pkt_id_t> active_dispatches;
     //! List of correlation ids whose dispatches have been completed and can be forgotten after the
     //! buffer flip.

source/lib/rocprofiler-sdk/pc_sampling/parser/rocr.h

@@ -22,6 +22,8 @@
 
 #pragma once
 
+#include <rocprofiler-sdk/fwd.h>
+
 #include <stdint.h>
 
 /**
@@ -69,8 +71,9 @@ typedef struct
     uint64_t      queue_size;
     uint64_t      write_index;
     uint64_t      read_index;
-    uint64_t      correlation_id;
-    reserved_type _[4];
+    /// both internal and external correlation ID.
+    rocprofiler_correlation_id_t correlation_id;
+    reserved_type                _[2];
 } dispatch_pkt_id_t;
 
 typedef struct

source/lib/rocprofiler-sdk/pc_sampling/parser/tests/correlation_id_test.cpp

@@ -52,7 +52,7 @@ static bool
 check_samples(pcsample_v1_t* samples, uint64_t size)
 {
     for(size_t i = 0; i < size; i++)
-        if(samples[i].correlation_id != samples[i].pc) return false;
+        if(samples[i].correlation_id.internal != samples[i].pc) return false;
     return true;
 }
 

source/lib/rocprofiler-sdk/pc_sampling/parser/tests/gfx9test.cpp

@@ -296,8 +296,8 @@ class WaveIssueAndErrorTest : public WaveSnapTest
     {
         pcsample_v1_t sample;
         ::memset(&sample, 0, sizeof(sample));
-        sample.pc             = dispatch->unique_id;
-        sample.correlation_id = dispatch->getMockId().raw;
+        sample.pc                      = dispatch->unique_id;
+        sample.correlation_id.internal = dispatch->getMockId().raw;
 
         sample.flags.valid              = valid && !error;
         sample.wave_issued              = issued;
@@ -353,7 +353,7 @@ class WaveOtherFieldsTest : public WaveSnapTest
             assert(compare[i].chiplet == parsed[0][i].chiplet);
             assert(compare[i].wave_id == parsed[0][i].wave_id);
             assert(compare[i].hw_id == parsed[0][i].hw_id);
-            assert(compare[i].correlation_id == parsed[0][i].correlation_id);
+            assert(compare[i].correlation_id.internal == parsed[0][i].correlation_id.internal);
         }
     }
 
@@ -367,10 +367,10 @@ class WaveOtherFieldsTest : public WaveSnapTest
         sample.workgroup_id_y = blky;
         sample.workgroup_id_z = blkz;
 
-        sample.chiplet        = chip;
-        sample.wave_id        = wave;
-        sample.hw_id          = hwid;
-        sample.correlation_id = dispatch->unique_id;
+        sample.chiplet                 = chip;
+        sample.wave_id                 = wave;
+        sample.hw_id                   = hwid;
+        sample.correlation_id.internal = dispatch->unique_id;
 
         compare.push_back(sample);
 

source/lib/rocprofiler-sdk/pc_sampling/parser/tests/mocks.hpp

@@ -188,12 +188,12 @@ public:
 
         packet_union_t uni;
         ::memset(&uni, 0, sizeof(uni));
-        uni.dispatch_id.type           = AMD_DISPATCH_PKT_ID;
-        uni.dispatch_id.doorbell_id    = doorbell_id;
-        uni.dispatch_id.queue_size     = queue->size;
-        uni.dispatch_id.write_index    = dispatch_id;
-        uni.dispatch_id.read_index     = queue->read_index;
-        uni.dispatch_id.correlation_id = unique_id;
+        uni.dispatch_id.type                    = AMD_DISPATCH_PKT_ID;
+        uni.dispatch_id.doorbell_id             = doorbell_id;
+        uni.dispatch_id.queue_size              = queue->size;
+        uni.dispatch_id.write_index             = dispatch_id;
+        uni.dispatch_id.read_index              = queue->read_index;
+        uni.dispatch_id.correlation_id.internal = unique_id;
         queue->submit(uni);
         queue->write_index++;
     };