add test and minor fixes

build_tools/ci/cpu_comparison/run_test.sh

@@ -303,6 +303,9 @@ function run_test() {
 run_test \
   --test_file ${THIS_DIR}/test_files/matmul_int32.mlir
 
+run_test \
+  --test_file ${THIS_DIR}/test_files/three_matmuls.mlir
+
 run_test \
    --name_prefix "matmul" \
    --lhs_rhs_type "bf16" \

build_tools/ci/cpu_comparison/test_files/three_matmuls.mlir

@@ -0,0 +1,31 @@
+// This test shows arbitory matmuls that would have producer consumer relationships
+// across different dispatches running on CI.
+
+// These lines are strictly required by the script which generates input data:
+//
+// input 32x32xf32
+// input 32x32xf32
+// input 32x4xf32
+// input 4x32xf32
+
+!A_TYPE = tensor<32x32xf32>
+!B_TYPE = tensor<32x4xf32>
+!C_TYPE = tensor <4x32xf32>
+!D_TYPE = tensor <4x4xf32>
+func.func @two_mm(%lhs : !A_TYPE,
+    %rhs : !A_TYPE, %rhs_2 : !B_TYPE, %lhs_2 : !C_TYPE) -> !D_TYPE {
+  %empty = tensor.empty() : !A_TYPE
+  %empty_2 = tensor.empty() : !B_TYPE
+  %empty_3 = tensor.empty() : !D_TYPE
+  %cst = arith.constant 0.0 : f32
+  %fill = linalg.fill ins(%cst : f32) outs(%empty : !A_TYPE) -> !A_TYPE
+  %fill_2 = linalg.fill ins(%cst : f32) outs(%empty_2 : !B_TYPE) -> !B_TYPE
+  %fill_3 = linalg.fill ins(%cst : f32) outs(%empty_3 : !D_TYPE) -> !D_TYPE
+  %2 = linalg.matmul ins(%lhs, %rhs : !A_TYPE, !A_TYPE)
+      outs(%fill : !A_TYPE) -> !A_TYPE
+  %3 = linalg.matmul ins(%2, %rhs_2 : !A_TYPE, !B_TYPE)
+      outs(%fill_2 : !B_TYPE) -> !B_TYPE
+  %4 = linalg.matmul ins(%lhs_2, %3 : !C_TYPE, !B_TYPE)
+      outs(%fill_3 : !D_TYPE) -> !D_TYPE
+  return %4 : !D_TYPE
+}

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

@@ -302,7 +302,7 @@ LogicalResult AIETargetBackend::serializeExecutable(
     llvm::sys::path::append(npuInstPath,
                             entryPointNamesFb[ordinal] + ".npu.txt");
 
-    // Convert ordinal to hexadecimal string for xclbin kern id
+    // Convert ordinal to hexadecimal string for xclbin kernel id.
     std::stringstream ss;
     ss << "0x" << std::hex << ordinal + 10;
     std::string ordinalHex = ss.str();
@@ -370,7 +370,7 @@ LogicalResult AIETargetBackend::serializeExecutable(
       int result = llvm::sys::ExecuteAndWait(cmdArgs[0], cmdArgs, cmdEnvRefs);
       if (result != 0 && AttemptingMerge) {
         // we failed to create xclbin but maybe we failed becuase we were trying
-        // to merge the kerenel in exisiting kernel, try again to see if perhaps
+        // to merge the kerenel in exisiting xclbin, try again to see if perhaps
         // we have success if we dont try to merge.
         AttemptingMerge = false;
         result =
@@ -400,18 +400,8 @@ LogicalResult AIETargetBackend::serializeExecutable(
     asmInstrIndices[ordinal] = asmInstrRefs.size();
     asmInstrRefs.push_back(
         iree_amd_aie_hal_xrt_AsmInstDef_create(builder, npuInstrsVec));
-    /*
-    xclbinIn = openInputFile(xclbinPath, &errorMessage);
-    if (!xclbinIn) {
-      moduleOp.emitOpError() << "Failed to open xclbin file: " << errorMessage;
-    }
-    auto xclbinStringRef = builder.createString(xclbinIn->getBuffer());
-    xclbinIndices[ordinal] = xclbinRefs.size();
-    xclbinRefs.push_back(
-        iree_amd_aie_hal_xrt_XclbinDef_create(builder, xclbinStringRef));
-    */
   }
-  // write out the final xclbins to flatbuffer
+  // Write out the final xclbins to flatbuffer.
   for (auto xclbinPath : xclbinPaths) {
     llvm::outs() << "writing xclbin from path: " << xclbinPath << "\n";
     std::string errorMessage;
@@ -424,7 +414,7 @@ LogicalResult AIETargetBackend::serializeExecutable(
         iree_amd_aie_hal_xrt_XclbinDef_create(builder, xclbinStringRef));
   }
 
-  // Serialize the executable to flatbuffer format
+  // Serialize the executable to flatbuffer format.
   auto entryPointsRef = builder.createStringVec(entryPointNamesFb);
 
   iree_amd_aie_hal_xrt_ExecutableDef_entry_points_add(builder, entryPointsRef);