Add chip specs (#681)

* Add perfmon config spec, enhance memory partition info.

* Add gfx950 perfmon config.

* Add High Freq variants in gfx942.

* Add backup detection methods for gpu model.

* Improve get_num_xcds logic by adding detection of 1to1 arch-to-compute_partition logic.

* Add default compute partition settings spx:8 for when gpu_model=None.

* Update gpu spec tests.

* Add backup compute partition detection.

---------

Signed-off-by: xuchen-amd <xuchen@amd.com>

tests/test_gpu_specs.py

@@ -22,6 +22,10 @@ GFX942_CHIP_IDS_TO_NUM_XCDS = {
     "29878": {"spx": 4, "dpx": 2, "cpx": 1},
     "29861": {"spx": 8, "dpx": 4, "qpx": 2, "cpx": 1},
     "29881": {"spx": 8, "dpx": 4, "qpx": 2, "cpx": 1},
+    "29864": {"spx": 4, "dpx": 2, "cpx": 1},
+    "29884": {"spx": 4, "dpx": 2, "cpx": 1},
+    "29865": {"spx": 8, "dpx": 4, "qpx": 2, "cpx": 1},
+    "29885": {"spx": 8, "dpx": 4, "qpx": 2, "cpx": 1},
 }
 
 # helper to strip ANSI color codes if your app uses them
@@ -91,17 +95,37 @@ def get_num_xcds():
     if str(chip_id) in GFX942_CHIP_IDS_TO_NUM_XCDS.keys():
         num_xcds = GFX942_CHIP_IDS_TO_NUM_XCDS[str(chip_id)]
 
+    if num_xcds is None:
+        return
+
     return num_xcds
 
 
+def get_gpu_arch():
+
+    rocminfo = str(
+        # decode with utf-8 to account for rocm-smi changes in latest rocm
+        subprocess.run(
+            ["rocminfo"], stdout=subprocess.PIPE, stderr=subprocess.PIPE
+        ).stdout.decode("utf-8")
+    )
+    rocminfo = rocminfo.split("\n")
+
+    soc_regex = re.compile(r"^\s*Name\s*:\s+ ([a-zA-Z0-9]+)\s*$", re.MULTILINE)
+    devices = list(filter(soc_regex.match, rocminfo))
+    gpu_arch = devices[0].split()[1]
+    return gpu_arch
+
+
 @pytest.mark.num_xcds_spec_class
 def test_num_xcds_spec_class(monkeypatch):
-    num_xcds = get_num_xcds()
-
     # 1. Check if gfx942 soc
-    if not num_xcds:
+    gpu_arch = get_gpu_arch()
+    if gpu_arch is None or gpu_arch.lower() != "gfx942":
         pytest.skip("Skipping num xcds test for non-gfx942 socs.")
 
+    num_xcds = get_num_xcds()
+
     # 2. load machine specs
     machine_spec = generate_machine_specs(None)
 
@@ -114,12 +138,13 @@ def test_num_xcds_spec_class(monkeypatch):
 
 @pytest.mark.num_xcds_cli_output
 def test_num_xcds_cli_output():
-    num_xcds = get_num_xcds()
-
     # 1. Check if gfx942 soc
-    if not num_xcds:
+    gpu_arch = get_gpu_arch()
+    if gpu_arch is None or gpu_arch.lower() != "gfx942":
         pytest.skip("Skipping num xcds test for non-gfx942 socs.")
 
+    num_xcds = get_num_xcds()
+
     # 2. Run rocprof-compute -s and grab rocprof-compute num_xcd
     proc = subprocess.run(
         ["rocprof-compute", "-s"],