[ESIMD] Add a LIT test verifying DPAS with 2 tfloat32 arguments

SYCL/ESIMD/dpas/dpas_tf32.cpp

@@ -0,0 +1,87 @@
+//==---------------- dpas_tf32.cpp  - DPC++ ESIMD on-device test ----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// REQUIRES: gpu-intel-pvc || esimd_emulator
+// UNSUPPORTED: cuda || hip
+// RUN: %clangxx -fsycl -DESIMD_XE_HPC %s -o %t.out
+// RUN: %GPU_RUN_PLACEHOLDER %t.out
+// XFAIL: esimd_emulator
+
+// The test verifies the low-level API for DPAS with 'tfloat32' types.
+// It checks the versions of DPAS with and without the accumulator operand.
+
+#include <sycl/ext/intel/esimd.hpp>
+#include <sycl/sycl.hpp>
+
+using namespace sycl;
+using namespace sycl::ext::intel::esimd;
+using namespace sycl::ext::intel::experimental::esimd;
+
+int main() {
+  queue Q;
+
+  constexpr int REPEAT_COUNT = 8;
+  constexpr int SYSTOLIC_DEPTH = 8;
+  constexpr int EXECUTION_SIZE = 16;
+
+  constexpr int M = REPEAT_COUNT;
+  constexpr int N = EXECUTION_SIZE;
+  constexpr int K = SYSTOLIC_DEPTH; // SYSTOLIC_DEPTH * OPS_PER_CHANNEL
+  float *A = malloc_shared<float>(M * K, Q);
+  float *B = malloc_shared<float>(K * N, Q);
+  float *C = malloc_shared<float>(M * N, Q);
+  float *D = malloc_shared<float>(M * N, Q);
+  for (int I = 0; I < M * K; ++I)
+    A[I] = I;
+  for (int I = 0; I < K * N; ++I)
+    B[I] = I;
+
+  Q.single_task([=]() SYCL_ESIMD_KERNEL {
+     simd<float, M * K> AVec(A);
+     simd<float, K * N> BVec(B);
+     auto AView = AVec.template bit_cast_view<uint>();
+     auto BView = BVec.template bit_cast_view<uint>();
+     // C(MxN) = A(MxK) * B(KxN)
+     simd<float, M *N> CVec =
+         dpas<argument_type::TF32, argument_type::TF32, SYSTOLIC_DEPTH,
+              REPEAT_COUNT, float, uint, uint, M * N, K * N, M * K>(
+             BView.read(), AView.read());
+     CVec.copy_to(C);
+
+     // D(MxN) = D(MxN) + A(MxK) * B(KxN);
+     simd<float, M *N> DVec = 1.0;
+     DVec = dpas<argument_type::TF32, argument_type::TF32, SYSTOLIC_DEPTH,
+                 REPEAT_COUNT, float, uint, uint, M * N, K * N, M * K>(
+         DVec, BView.read(), AView.read());
+     DVec.copy_to(D);
+   }).wait();
+
+  unsigned ErrCnt = 0;
+  for (unsigned I = 0; I < M * N && ErrCnt < 10; ++I) {
+    int m = I / N;
+    int n = I % N;
+    float RefResC = 0.0f;
+    for (int k = 0; k < K; ++k)
+      RefResC += float((m * K + k) * (k * N + n));
+    if (std::abs(RefResC - C[I]) > 0.001) {
+      std::cerr << "C[i] vs ref: " << C[I] << " : " << RefResC << std::endl;
+      ErrCnt++;
+    }
+    float RefResD = RefResC + 1.0;
+    if (std::abs(RefResD - D[I]) > 0.001) {
+      std::cerr << "D[i] vs ref: " << D[I] << " : " << RefResD << std::endl;
+      ErrCnt++;
+    }
+  }
+  free(A, Q);
+  free(B, Q);
+  free(C, Q);
+  free(D, Q);
+
+  std::cout << (ErrCnt > 0 ? "FAILED\n" : "Passed\n");
+  return ErrCnt > 0 ? 1 : 0;
+}