[E2E][CI] Automatically add device family feature from architecture reported by sycl-ls (#16220)

The PR adds GPU family features like `gpu-intel-gen12`, `gpu-intel-dg2`
based on the specific architecture type reported by `sycl-ls`.
After this change, you won't have to add `--param
gpu-intel-gen12=True`/`--param gpu-intel-dg1=True`, etc LIT options when
running E2E tests.

Author: uditagarwal97

.github/workflows/sycl-linux-precommit.yml

@@ -143,15 +143,15 @@ jobs:
             image_options: -u 1001 --device=/dev/dri -v /dev/dri/by-path:/dev/dri/by-path --privileged --cap-add SYS_ADMIN
             target_devices: level_zero:gpu;opencl:gpu
             reset_intel_gpu: true
-            extra_lit_opts: --param matrix-xmx8=True --param gpu-intel-dg2=True
+            extra_lit_opts: --param matrix-xmx8=True
             env: '{"LIT_FILTER":${{ needs.determine_arc_tests.outputs.arc_tests }} }'
           - name: E2E tests with dev igc on Intel Arc A-Series Graphics
             runner: '["Linux", "arc"]'
             image: ghcr.io/intel/llvm/ubuntu2404_intel_drivers:devigc
             image_options: -u 1001 --device=/dev/dri -v /dev/dri/by-path:/dev/dri/by-path --privileged --cap-add SYS_ADMIN
             target_devices: level_zero:gpu;opencl:gpu
             reset_intel_gpu: true
-            extra_lit_opts: --param matrix-xmx8=True --param gpu-intel-dg2=True
+            extra_lit_opts: --param matrix-xmx8=True
             env: '{"LIT_FILTER":${{ needs.determine_arc_tests.outputs.arc_tests }} }'
             use_igc_dev: true
           - name: E2E tests on Intel Ponte Vecchio GPU

.github/workflows/sycl-nightly.yml

@@ -58,7 +58,6 @@ jobs:
             target_devices: level_zero:gpu
             reset_intel_gpu: true
             tests_selector: e2e
-            extra_lit_opts: --param gpu-intel-gen12=True
 
           - name: Intel OCL GPU
             runner: '["Linux", "gen12"]'
@@ -67,7 +66,6 @@ jobs:
             target_devices: opencl:gpu
             reset_intel_gpu: true
             tests_selector: e2e
-            extra_lit_opts: --param gpu-intel-gen12=True
 
           - name: OCL CPU (AMD)
             runner: '["Linux", "amdgpu"]'
@@ -126,7 +124,6 @@ jobs:
       name: Intel GEN12 Graphics with Level Zero
       runner: '["Windows","gen12"]'
       sycl_toolchain_archive: ${{ needs.build-win.outputs.artifact_archive_name }}
-      extra_lit_opts: --param gpu-intel-gen12=True
 
   cuda-aws-start:
     needs: [ubuntu2204_build]

.github/workflows/sycl-post-commit.yml

@@ -47,12 +47,11 @@ jobs:
         include:
           - name: Intel GEN12 Graphics with Level Zero
             runner: '["Linux", "gen12"]'
-            extra_lit_opts: --param gpu-intel-gen12=True
             target_devices: level_zero:gpu;opencl:fpga
             reset_intel_gpu: true
           - name: Intel Arc A-Series Graphics with Level Zero
             runner: '["Linux", "arc"]'
-            extra_lit_opts: --param matrix-xmx8=True --param gpu-intel-dg2=True
+            extra_lit_opts: --param matrix-xmx8=True
             reset_intel_gpu: true
           - name: AMD/HIP
             runner: '["Linux", "amdgpu"]'
@@ -69,7 +68,7 @@ jobs:
           - name: Perf tests on Intel GEN12 Graphics system
             runner: '["Linux", "gen12"]'
             env: '{"LIT_FILTER":"PerformanceTests/"}'
-            extra_lit_opts: -a -j 1 --param enable-perf-tests=True --param gpu-intel-gen12=True
+            extra_lit_opts: -a -j 1 --param enable-perf-tests=True
             target_devices: all
             reset_intel_gpu: true
           - name: Perf tests on Intel Arc A-Series Graphics system
@@ -120,7 +119,6 @@ jobs:
       name: Intel GEN12 Graphics with Level Zero
       runner: '["Windows","gen12"]'
       sycl_toolchain_archive: ${{ needs.build-win.outputs.artifact_archive_name }}
-      extra_lit_opts: --param gpu-intel-gen12=True
       compiler: icx
 
   macos_default:

.github/workflows/sycl-windows-precommit.yml

@@ -55,4 +55,3 @@ jobs:
       name: Intel GEN12 Graphics with Level Zero
       runner: '["Windows","gen12"]'
       sycl_toolchain_archive: ${{ needs.build.outputs.artifact_archive_name }}
-      extra_lit_opts: --param gpu-intel-gen12=True

sycl/test-e2e/README.md

@@ -239,6 +239,12 @@ environment:
 * **arch-\<name\>** - [SYCL architecture](https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/experimental/sycl_ext_oneapi_device_architecture.asciidoc)
   of a device (e.g. `arch-intel_gpu_pvc`, the name matches what you can pass
   into `-fsycl-targets` compiler flag);
+* **gpu-intel-dg2**  - Intel GPU DG2 availability; Automatically set if device
+  architecture belongs to DG2 family;
+* **gpu-intel-gen11**  - Intel GPU Gen11 availability; Automatically set if device
+  architecture belongs to Gen11 family;
+* **gpu-intel-gen12**  - Intel GPU Gen12 availability; Automatically set if device
+  architecture belongs to Gen12 family;
 
 #### Manually-set features
 
@@ -247,17 +253,9 @@ section below). All these features are related to HW detection and they should
 be considered deprecated, because we have HW auto-detection functionality in
 place. No new tests should use these features:
 
-* **gpu-intel-gen11** - Intel GPU Gen11 availability;
-* **gpu-intel-gen12** - Intel GPU Gen12 availability;
-* **gpu-intel-dg2** - Intel GPU DG2 availability;
 * **gpu-intel-pvc** - Intel GPU PVC availability;
 * **gpu-intel-pvc-vg** - Intel GPU PVC-VG availability;
 
-Note: some of those features describing whole GPU families and auto-detection of
-HW does not provide this functionality at the moment. As an improvement, we
-could add those features even with auto-detection, because the only alternative
-at the moment is to explicitly list every architecture from a family.
-
 ### llvm-lit parameters
 
 Following options can be passed to llvm-lit tool through --param option to

sycl/test-e2e/lit.cfg.py

@@ -176,6 +176,31 @@
 if lit_config.params.get("igc-dev", False):
     config.available_features.add("igc-dev")
 
+# Map between device family and architecture types.
+device_family_arch_map = {
+    # <Family name> : Set of architectures types (and aliases)
+    # DG2
+    "gpu-intel-dg2": {
+        "intel_gpu_acm_g12",
+        "intel_gpu_dg2_g12",
+        "intel_gpu_acm_g11",
+        "intel_gpu_dg2_g11",
+        "intel_gpu_acm_g10",
+        "intel_gpu_dg2_g10",
+    },
+    # Gen12
+    "gpu-intel-gen12": {"intel_gpu_tgllp", "intel_gpu_tgl"},
+    # Gen11
+    "gpu-intel-gen11": {"intel_gpu_icllp", "intel_gpu_icl"},
+}
+
+
+def get_device_family_from_arch(arch):
+    for device_family, arch_set in device_family_arch_map.items():
+        if arch in arch_set:
+            return device_family
+    return None
+
 def check_igc_tag_and_add_feature():
     if os.path.isfile(config.igc_tag_file):
         with open(config.igc_tag_file, "r") as tag_file:
@@ -773,10 +798,27 @@ def open_check_file(file_name):
     aspect_features = set("aspect-" + a for a in aspects)
     sg_size_features = set("sg-" + s for s in sg_sizes)
     architecture_feature = set("arch-" + s for s in architectures)
+    # Add device family features like intel-gpu-gen12, intel-gpu-dg2 based on
+    # the architecture reported by sycl-ls.
+    device_family = set(
+        get_device_family_from_arch(arch)
+        for arch in architectures
+        if get_device_family_from_arch(arch) is not None
+    )
+
+    # Print the detected GPU family name.
+    if len(device_family) > 0:
+        lit_config.note(
+            "Detected GPU family for {}: {}".format(
+                sycl_device, ", ".join(device_family)
+            )
+        )
+
     features = set()
     features.update(aspect_features)
     features.update(sg_size_features)
     features.update(architecture_feature)
+    features.update(device_family)
 
     be, dev = sycl_device.split(":")
     features.add(dev.replace("fpga", "accelerator"))