Merge branch 'main' into thread_groups_for_conv

.github/workflows/ci-linux.yml

@@ -91,12 +91,14 @@ jobs:
 
   test_linux:
     name: E2E Test linux
-    runs-on: amd7940hs
     needs: build_and_ctest
     strategy:
-      fail-fast: true
+      fail-fast: false
+      matrix:
+        runs-on: [linux-phoenix-20240606, linux-phoenix-20240819]
+    runs-on: ${{ matrix.runs-on }}
     env:
-      XILINXD_LICENSE_FILE: /home/svcnod/Xilinx.lic
+      XILINXD_LICENSE_FILE: /opt/xilinx/Xilinx.lic
     steps:
       - name: "Checking out repository" # for test scripts
         uses: actions/checkout@8f4b7f84864484a7bf31766abe9204da3cbe65b3 # v3.5.0

build_tools/ci/cpu_comparison/run_test.py

@@ -306,6 +306,15 @@ def __init__(
         if xrt_hash:
             self.xrt_hash = xrt_hash[0]
 
+        xdna_datetime = re.findall(
+            # eg 2.18.0_20240606
+            r"amdxdna\s+:\s\d\.\d+\.\d+_(\d+)",
+            xrt_info,
+            flags=re.MULTILINE | re.IGNORECASE,
+        )
+        if xdna_datetime:
+            self.xdna_datetime = int(xdna_datetime[0])
+
         # Try and get the peano commit hash. This is a bit of a hack, if it fails
         # peano_commit_has is left as "undetermined".
         self.peano_commit_hash = "undetermined"
@@ -589,38 +598,45 @@ def run(self, config):
         test_files_dir = config.file_dir / "test_files"
         output_dir = config.output_dir
 
-        for name in [
-            "two_matmul_switching",
-            "matmul_f32_8_8_4",
-            "matmul_f32_8_4_8",
-        ]:
-            aie_vs_llvm_cpu(config, test_files_dir / f"{name}.mlir")
-
-        aie_vs_llvm_cpu(
-            config,
-            test_files_dir / "three_matmuls.mlir",
-            function_name="three_$mm$",
-        )
-
-        # Test(s) of the form matmul(A,B) where A:MxK, B:KxN
-        test_name = output_dir / "test_from_template.mlir"
-        template_name = matmul_template_dir / "matmul_MxK_KxN.mlir"
-        generate_matmul_test(test_name, template_name, 32, 32, 64, "bf16", "f32")
-        aie_vs_llvm_cpu(config, test_name)
-
-        # Test(s) of the form matmul(A,B) + C where A:MxK, B:KxN, C:N
-        test_name = output_dir / "test_from_template_bias_N.mlir"
-        template_name = matmul_template_dir / "matmul_bias_MxK_KxN_N.mlir"
-        generate_matmul_test(test_name, template_name, 1024, 1024, 512, "bf16", "f32")
-        aie_vs_llvm_cpu(config, test_name, tile_pipeline="pack-peel", use_ukernel=True)
-        aie_vs_llvm_cpu(config, test_name, tile_pipeline="pack-peel", use_ukernel=False)
-
         # Test(s) of the form matmul(A,B) + C where A:MxK, B:KxN, C:MxN
         test_name = output_dir / "test_from_template_full_bias.mlir"
         template_name = matmul_template_dir / "matmul_bias_MxK_KxN_MxN.mlir"
         generate_matmul_test(test_name, template_name, 128, 128, 256, "i32", "i32")
         aie_vs_llvm_cpu(config, test_name, tile_pipeline="pack-peel", rtol=0, atol=0)
 
+        if config.xdna_datetime and config.xdna_datetime < 20240819:
+            for name in [
+                "two_matmul_switching",
+                "matmul_f32_8_8_4",
+                "matmul_f32_8_4_8",
+            ]:
+                aie_vs_llvm_cpu(config, test_files_dir / f"{name}.mlir")
+
+            aie_vs_llvm_cpu(
+                config,
+                test_files_dir / "three_matmuls.mlir",
+                function_name="three_$mm$",
+            )
+
+            # Test(s) of the form matmul(A,B) where A:MxK, B:KxN
+            test_name = output_dir / "test_from_template.mlir"
+            template_name = matmul_template_dir / "matmul_MxK_KxN.mlir"
+            generate_matmul_test(test_name, template_name, 32, 32, 64, "bf16", "f32")
+            aie_vs_llvm_cpu(config, test_name)
+
+            # Test(s) of the form matmul(A,B) + C where A:MxK, B:KxN, C:N
+            test_name = output_dir / "test_from_template_bias_N.mlir"
+            template_name = matmul_template_dir / "matmul_bias_MxK_KxN_N.mlir"
+            generate_matmul_test(
+                test_name, template_name, 1024, 1024, 512, "bf16", "f32"
+            )
+            aie_vs_llvm_cpu(
+                config, test_name, tile_pipeline="pack-peel", use_ukernel=True
+            )
+            aie_vs_llvm_cpu(
+                config, test_name, tile_pipeline="pack-peel", use_ukernel=False
+            )
+
 
 class SmokeSet(TestSet):
     def __init__(self):

compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/AIETarget.cpp

@@ -363,6 +363,24 @@ LogicalResult AIETargetBackend::serializeExecutable(
       deviceOps[i] = cast<xilinx::AIE::DeviceOp>(repl);
     }
 
+    // TODO(max): this should be an enum
+    // TODO(max): this needs to be pulled from PCIE
+    std::string npuVersion;
+    switch (clAMDAIETargetDevice) {
+      case AMDAIEDevice::npu1:
+      case AMDAIEDevice::npu1_1col:
+      case AMDAIEDevice::npu1_2col:
+      case AMDAIEDevice::npu1_3col:
+      case AMDAIEDevice::npu1_4col:
+        npuVersion = "npu1";
+        break;
+      case AMDAIEDevice::npu4:
+        npuVersion = "npu4";
+        break;
+      default:
+        llvm::report_fatal_error("unhandled NPU partitioning.\n");
+    }
+
     if (failed(aie2xclbin(
             /*ctx=*/variantOp->getContext(), deviceOps[i],
             /*outputNPU=*/npuInstPath.str().str(),
@@ -377,7 +395,9 @@ LogicalResult AIETargetBackend::serializeExecutable(
             /*vitisDir=*/options.vitisInstallDir.empty()
                 ? std::nullopt
                 : std::optional<std::string>{options.vitisInstallDir},
+            // TODO(max): not right for strix
             /*targetArch=*/"AIE2",
+            /*npuVersion=*/npuVersion,
             /*peanoDir=*/options.peanoInstallDir,
             /*xclBinKernelID=*/ordinalHex.str(),
             /*xclBinKernelName=*/entryPointNamesFb[ordinal],

compiler/plugins/target/AMD-AIE/iree-amd-aie/Target/AIETarget.h

@@ -43,7 +43,6 @@ struct AMDAIEOptions {
   // Print MLIR timing summary for the MLIR passes in aie2xclbin.
   bool aie2xclbinTiming{false};
 
- public:
   void bindOptions(OptionsBinder &binder) {
     static llvm::cl::OptionCategory category("AMD AIE Options");
     binder.opt<std::string>(