SWDEV-402866: Added browser file mode. Fixed caching. Changed ATT buffer size. Added MAX_ATT environment variable. Updated README/Changelog.

Change-Id: I36a6093bb85bf4ef179b59df676fc2e4cbdb5288

CHANGELOG.md

@@ -209,18 +209,26 @@ The resulting `a.out` will depend on
 
 ### Optimized
 - Improved Test Suite
-### Changed
-- ATT analysis will not run by default. For ATT to have the same behaviour as 5.5, use --plugin att <as.s> --mode network
 ### Added
 - 'end_time' need to be disabled in roctx_trace.txt
 - support for hsa_amd_memory_async_copy_on_engine API function trace
+
 ### Fixed
 - rocprof in ROcm/5.4.0 gpu selector broken.
 - rocprof in ROCm/5.4.1 fails to generate kernel info.
 - rocprof clobbers LD_PRELOAD.
 
 ## ROCprofiler for rocm 5.7.0
+### Navi support
+Rocprofiler for ROCm 5.7 added support for counter collection (PMC) and advanced thread tracing (ATT) for Navi21 and Navi31 GPUs.
+- On Navi, specially Navi31, counter collection requires the GPU to be in a stable power state. See README.md for instructions.
+- Navi does not support streaming SQ counters and ATT at the same time, unlike GFX9.
+- On Navi ATT, "att: target_cu" indexes the WGP and the SIMD_MASK parameter is actually the SIMD_ID, in the range [0,3].
+- HIP RT in ATT not yet supported.
+### Changed
+- ATT analysis will not run by default. For ATT to have the same behaviour as 5.5, use --plugin att <as.s> --mode network
 ### Optimized
+- ATT json filesizes
 ### Added
 - Every API trace in V2 reported synchronously will have two records, one for Enter phase and for Exit phase
 - File Plugin now reports the HSA OPS operation kind as part of the output text
@@ -230,9 +238,17 @@ The resulting `a.out` will depend on
 - MI300 individual XCC counters dumped per-xcc as separate records but with same record-id and kernel dispatch info
 - Naming for MPI ranks. Filenames containing "%rank" are replaced by variables "MPI_RANK", "OMPI_COMM_WORLD_RANK" or "MV2_COMM_WORLD_RANK".
 - MPI Rank will appear in perfetto track names.
+- SE_MASK parameter in ATT, a binary mask specifying for which shader engines to run ATT.
+  On GFX9, SEs are masked out completely. On Navi only part of the data is masked.
+  The use of SE_MASK=0x1 is heavily encouraged to avoid packet lost events.
+- "--mode file" option in ATT, which allows for parsed files to be stored. Run python3 httpserver.py from within ./UI/ to view files locally.
+- "ROCPROFILER_MAX_ATT_PROFILES" environment variable can be set. Previously fixed at 16, now the default is 1.
+- Increased ATT buffer size per collection to 1GB.
 ### Fixed
 - Samples are fixed to show the new usage of phases.
 - Plugin option validates the plugin names.
 - Fixing rocsys, for rocsys options, rocsys -h can be called
 - "--output-file" option ignored when no output folder was specified.
 - Perfetto crash when using ROCTX and/or no output file specified.
+- Parsing of the getpc, setpc and swappc instructions with registers loaded from scratch space.
+- Some browsers caching ATT data from older kernels.

README.md

@@ -381,3 +381,7 @@ samples can be run as independent executables once installed
 Please report in the Github Issues
 
 ## Limitations
+- Navi requires a stable power state for counter collection. Currently this state needs to be set by the user.
+  To do so, set "power_dpm_force_performance_level" to be writeable for non-root users with chmod, then:
+  echo profile_standard >> /sys/class/drm/card0/device/power_dpm_force_performance_level
+  Recommended: "auto" or "high" for ATT and "profile_standard" for PMC. Use rocm-smi to verify the current power state.

plugin/att/CMakeLists.txt

@@ -57,11 +57,10 @@ install(TARGETS att_plugin
 
 configure_file(att.py att/att.py COPYONLY)
 configure_file(trace_view.py att/trace_view.py COPYONLY)
-# configure_file(t.db att/t.db COPYONLY)
 configure_file(ui/index.html att/ui/index.html COPYONLY)
 configure_file(ui/logo.svg att/ui/logo.svg COPYONLY)
 configure_file(ui/styles.css att/ui/styles.css COPYONLY)
-# configure_file(ui/trace.json att/ui/trace.json COPYONLY)
+configure_file(ui/httpserver.py att/ui/httpserver.py COPYONLY)
 install(
   DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/att
   DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/rocprofiler

plugin/att/att.py

@@ -19,7 +19,7 @@ from io import BytesIO
 
 class FileBytesIO:
     def __init__(self, iobytes):
-        self.iobytes = iobytes
+        self.iobytes = deepcopy(iobytes)
         self.seek = 0
 
     def __len__(self):
@@ -27,9 +27,9 @@ class FileBytesIO:
 
     def read(self, length=0):
         if length<=0:
-            return bytes(self.getbuffer())
+            return bytes(self.iobytes.getbuffer())
         else:
-            if self.seek >= len(self):
+            if self.seek >= self.iobytes.getbuffer().nbytes:
                 self.seek = 0
                 return None
             response =  self.iobytes.getbuffer()[self.seek:self.seek+length]
@@ -382,14 +382,14 @@ def draw_wave_states(selections, normalize):
     if normalize:
         timelines = np.array(timelines) / np.maximum(np.sum(timelines,0)*1E-2,1E-7)
 
-    kernsize = maxtime//150+1
     trim = max(maxtime//5000,1)
-    cycles = np.arange(timelines[0].size)[::trim]
-
+    cycles = np.arange(0, timelines[0].size//trim, 1)*trim
+    timelines = [time[:trim*(time.size//trim)].reshape((-1, trim)).mean(-1) if len(time) > 0 else cycles*0 for time in timelines]
+    kernsize = 21
     kernel = np.asarray([np.exp(-abs(10*k/kernsize)) for k in range(-kernsize//2,kernsize//2+1)])
     kernel /= np.sum(kernel)
 
-    timelines = [np.convolve(time, kernel)[kernsize//2:-kernsize//2][::trim] if len(time) > 0 else cycles*0 for time in timelines]
+    timelines = [np.convolve(time, kernel)[kernsize//2:-kernsize//2] for time in timelines]
 
     [plt.plot(cycles, t, label='State '+s, linewidth=1.1, color=c)
         for t, s, c, sel in zip(timelines, STATES, colors, selections) if sel]
@@ -456,8 +456,6 @@ if __name__ == "__main__":
         for line in lines:
             if 'PERFCOUNTER=' in line:
                 EVENT_NAMES += [clean(line).split('SQ_')[1].lower()]
-    if len(EVENT_NAMES) == 0:
-        EVENT_NAMES = ['SPI', 'Vdata', 'Sdata', 'LDS']
     if args.target_cu is None:
         args.target_cu = 1
 
@@ -546,7 +544,7 @@ if __name__ == "__main__":
             tuples3 = [(0,df['begin_time'][T]-min_event_time)]+[(int(t[0]),int(t[1])) for t in tuples2]
 
             for state in tuples3:
-                if state[1] > 50E6:
+                if state[1] > 1E8:
                     print('Warning: Time limit reached for ',state[0], state[1])
                     break
                 if time_acc+state[1] > TIMELINES[state[0]].size:

plugin/att/trace_view.py

@@ -798,10 +798,7 @@ def view_trace(args, code, jumps, dbnames, att_filenames, bReturnLoc, pic_callba
             print("Exitting.")
     else:
         os.makedirs('ui', exist_ok=True)
+        os.system('cp ' + os.path.join(os.path.abspath(os.path.dirname(__file__)),'ui') + '/* ui/' )
         for k, v in JSON_GLOBAL_DICTIONARY.items():
-            if '.json' in k:
-                try:
-                    with open(os.path.join('ui',k), 'w') as f:
-                        f.write(v.read())
-                except:
-                    pass
+            with open(os.path.join('ui',k), 'w' if '.json' in k else 'wb') as f:
+                f.write(v.read())

plugin/att/ui/httpserver.py

@@ -0,0 +1,64 @@
+#!/usr/bin/env python3
+import sys
+if sys.version_info[0] < 3:
+    raise Exception("Must be using Python 3")
+
+import http.server
+import socketserver
+import socket
+import os
+import sys
+
+class NoCacheHTTPRequestHandler(http.server.SimpleHTTPRequestHandler):
+    def end_headers(self):
+        self.send_my_headers()
+        http.server.SimpleHTTPRequestHandler.end_headers(self)
+
+    def send_my_headers(self):
+        self.send_header("Cache-Control", "no-cache, no-store, must-revalidate")
+        self.send_header("Pragma", "no-cache")
+        self.send_header("Expires", "0")
+
+    def do_GET(self):
+        if 'timeline.png?' in self.path:
+            self.path = 'timeline.png'
+
+        http.server.SimpleHTTPRequestHandler.do_GET(self)
+
+class RocTCPServer(socketserver.TCPServer):
+    def server_bind(self):
+        self.socket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+        self.socket.bind(self.server_address)
+
+def run_server():
+    Handler = NoCacheHTTPRequestHandler
+    os.chdir(os.path.join(os.path.dirname(os.path.abspath(__file__)),'.'))
+    try:
+        with RocTCPServer((IPAddr, PORT), Handler) as httpd:
+            httpd.serve_forever()
+    except KeyboardInterrupt:
+        pass
+
+def get_ip():
+    s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
+    s.settimeout(0)
+    try:
+        hostname = socket.gethostname()
+        IPAddr = socket.gethostbyname(hostname)
+        s.connect(({IPAddr}, 1))
+    except Exception:
+        IPAddr = '127.0.0.1'
+    finally:
+        return IPAddr
+
+IPAddr = get_ip()
+PORT = 8000
+
+if len(sys.argv) > 1:
+    PORT = int(sys.argv[1])
+print('serving at port: {0}'.format(PORT))
+
+try:
+    run_server()
+except KeyboardInterrupt:
+    print("Exitting.")

plugin/att/ui/index.html

@@ -206,7 +206,7 @@
 		var current_WV = 0
 		var filename_data = {}
 
-		fetch("filenames.json").then(response => response.json()).then(data => {
+		fetch("filenames.json", {cache: "no-store"}).then(response => response.json()).then(data => {
 			filename_data = data.wave_filenames
 
 			wave_cu_index = {};
@@ -272,7 +272,7 @@
 			document.getElementById("minimap").innerHTML = HTML_MINI
 			document.getElementById("Images").innerHTML = HTML_IMAG
 
-			fetch("counters.json").then(response => response.json()).then(data => {
+			fetch("counters.json", {cache: "no-store"}).then(response => response.json()).then(data => {
 				var html_gh = '<input type="checkbox" id="btn_norm" onclick="UpdImageSrc()" checked=true>Normalize\t'
 				for(var key in data.counters) {
 					console.log(key, data.counters[key])
@@ -294,14 +294,14 @@
 			//document.getElementById('what').innerHTML = ""
 
 			d3.select('nav').style('visibility', 'hidden')
-			fetch(file_to_gather)
+			fetch(file_to_gather, {cache: "no-store"})
 				.then(response => response.json())
 				.then(data => {
 				code_data_file = file_to_gather.split('_sm')[0]+'_code.json'
 				console.log("Requestd:", file_to_gather)
 				console.log("Request code:", code_data_file)
 
-				fetch(code_data_file)
+				fetch(code_data_file, {cache: "no-store"})
 				.then(response => response.json())
 				.then(code_data => {
 					const SP = '\u00A0'

src/core/hsa/packets/packets_generator.cpp

@@ -468,7 +468,7 @@ hsa_ven_amd_aqlprofile_profile_t* InitializeDeviceProfilingAqlPackets(
 }
 
 // ATT
-uint32_t g_output_buffer_size = 0x8000000;  // 128M x 16 = 2GB
+uint32_t g_output_buffer_size = 0x40000000;  // 1GB
 bool g_output_buffer_local = true;
 
 // Allocate system memory accessible by both CPU and GPU

src/core/hsa/queues/queue.cpp

@@ -50,7 +50,6 @@
   } while (0)
 
 #define __NR_gettid 186
-#define MAX_ATT_PROFILES 16
 
 std::mutex sessions_pending_signal_lock;
 
@@ -664,6 +663,9 @@ std::atomic<uint32_t> WRITER_ID{0};
  */
 void WriteInterceptor(const void* packets, uint64_t pkt_count, uint64_t user_pkt_index, void* data,
                       hsa_amd_queue_intercept_packet_writer writer) {
+  static const char* env_MAX_ATT_PROFILES = getenv("ROCPROFILER_MAX_ATT_PROFILES");
+  static int MAX_ATT_PROFILES = env_MAX_ATT_PROFILES ? atoi(env_MAX_ATT_PROFILES) : 1;
+
   const Packet::packet_t* packets_arr = reinterpret_cast<const Packet::packet_t*>(packets);
   std::vector<Packet::packet_t> transformed_packets;