[Matrix][SYCL] Add bfloat16 support for joint_matrix

sycl/include/sycl/ext/oneapi/matrix/matrix-jit.hpp

@@ -11,6 +11,7 @@
 #include <CL/__spirv/spirv_ops.hpp>
 #include <CL/sycl/detail/defines_elementary.hpp>
 #include <CL/sycl/feature_test.hpp>
+#include <sycl/ext/oneapi/experimental/bfloat16.hpp>
 
 __SYCL_INLINE_NAMESPACE(cl) {
 namespace sycl {
@@ -453,6 +454,156 @@ class wi_element<uint16_t, NumRows, NumCols, Layout, Group> {
 #undef OP
 };
 
+template <size_t NumRows, size_t NumCols, matrix_layout Layout, typename Group>
+class wi_element<sycl::ext::oneapi::experimental::bfloat16, NumRows, NumCols,
+                 Layout, Group> {
+  joint_matrix<sycl::ext::oneapi::experimental::bfloat16, NumRows, NumCols,
+               Layout, Group> &M;
+  std::size_t idx;
+
+public:
+  wi_element(joint_matrix<sycl::ext::oneapi::experimental::bfloat16, NumRows,
+                          NumCols, Layout, Group> &Mat,
+             std::size_t i)
+      : M(Mat), idx(i) {}
+  operator sycl::ext::oneapi::experimental::bfloat16() {
+#ifdef __SYCL_DEVICE_ONLY__
+    return __spirv_VectorExtractDynamic(M.spvm, idx);
+#else
+    throw runtime_error("joint matrix is not supported on host device.",
+                        PI_INVALID_DEVICE);
+#endif // __SYCL_DEVICE_ONLY__
+  }
+
+  explicit operator bool() {
+#ifdef __SYCL_DEVICE_ONLY__
+    return std::fabs(static_cast<float>(__spirv_VectorExtractDynamic(
+               M.spvm, idx))) >= std::numeric_limits<float>::epsilon();
+#else
+    throw runtime_error("joint matrix is not supported on host device.",
+                        PI_INVALID_DEVICE);
+#endif // __SYCL_DEVICE_ONLY__
+  }
+
+  wi_element &operator=(const sycl::ext::oneapi::experimental::bfloat16 &rhs) {
+#ifdef __SYCL_DEVICE_ONLY__
+    M.spvm = __spirv_VectorInsertDynamic(M.spvm, rhs, idx);
+    return *this;
+#else
+    (void)rhs;
+    throw runtime_error("joint matrix is not supported on host device.",
+                        PI_INVALID_DEVICE);
+#endif // __SYCL_DEVICE_ONLY__
+  }
+
+  wi_element &
+  operator=(const wi_element<sycl::ext::oneapi::experimental::bfloat16, NumRows,
+                             NumCols, Layout, Group> &rhs) {
+#ifdef __SYCL_DEVICE_ONLY__
+    M.spvm = __spirv_VectorInsertDynamic(
+        M.spvm, __spirv_VectorExtractDynamic(rhs.M.spvm, rhs.idx), idx);
+    return *this;
+#else
+    (void)rhs;
+    throw runtime_error("joint matrix is not supported on host device.",
+                        PI_INVALID_DEVICE);
+#endif // __SYCL_DEVICE_ONLY__
+  }
+
+#if __SYCL_DEVICE_ONLY__
+#define OP(opassign, op)                                                       \
+  wi_element &operator opassign(                                               \
+      const sycl::ext::oneapi::experimental::bfloat16 &rhs) {                  \
+    M.spvm = __spirv_VectorInsertDynamic(                                      \
+        M.spvm, __spirv_VectorExtractDynamic(M.spvm, idx) op rhs, idx);        \
+    return *this;                                                              \
+  }
+#else // __SYCL_DEVICE_ONLY__
+#define OP(opassign, op)                                                       \
+  wi_element &operator opassign(                                               \
+      const sycl::ext::oneapi::experimental::bfloat16 &rhs) {                  \
+    (void)rhs;                                                                 \
+    throw runtime_error("joint matrix is not supported on host device.",       \
+                        PI_INVALID_DEVICE);                                    \
+  }
+#endif // __SYCL_DEVICE_ONLY__
+  OP(+=, +)
+  OP(-=, -)
+  OP(*=, *)
+  OP(/=, /)
+#undef OP
+
+#if __SYCL_DEVICE_ONLY__
+#define OP(type, op)                                                           \
+  friend type operator op(                                                     \
+      const wi_element<sycl::ext::oneapi::experimental::bfloat16, NumRows,     \
+                       NumCols, Layout, Group> &lhs,                           \
+      const sycl::ext::oneapi::experimental::bfloat16 &rhs) {                  \
+    return __spirv_VectorExtractDynamic(lhs.M.spvm, lhs.idx) op rhs;           \
+  }                                                                            \
+  friend type operator op(                                                     \
+      const sycl::ext::oneapi::experimental::bfloat16 &lhs,                    \
+      const wi_element<sycl::ext::oneapi::experimental::bfloat16, NumRows,     \
+                       NumCols, Layout, Group> &rhs) {                         \
+    return __spirv_VectorExtractDynamic(rhs.M.spvm, rhs.idx) op lhs;           \
+  }
+  OP(sycl::ext::oneapi::experimental::bfloat16, +)
+  OP(sycl::ext::oneapi::experimental::bfloat16, -)
+  OP(sycl::ext::oneapi::experimental::bfloat16, *)
+  OP(sycl::ext::oneapi::experimental::bfloat16, /)
+#undef OP
+#define OP(type, op)                                                           \
+  friend type operator op(                                                     \
+      const wi_element<sycl::ext::oneapi::experimental::bfloat16, NumRows,     \
+                       NumCols, Layout, Group> &lhs,                           \
+      const sycl::ext::oneapi::experimental::bfloat16 &rhs) {                  \
+    return type{static_cast<float>(__spirv_VectorExtractDynamic(               \
+        lhs.M.spvm, lhs.idx)) op static_cast<float>(rhs)};                     \
+  }                                                                            \
+  friend type operator op(                                                     \
+      const sycl::ext::oneapi::experimental::bfloat16 &lhs,                    \
+      const wi_element<sycl::ext::oneapi::experimental::bfloat16, NumRows,     \
+                       NumCols, Layout, Group> &rhs) {                         \
+    return type{static_cast<float>(__spirv_VectorExtractDynamic(               \
+        rhs.M.spvm, rhs.idx)) op static_cast<float>(lhs)};                     \
+  }
+  OP(bool, ==)
+  OP(bool, !=)
+  OP(bool, <)
+  OP(bool, >)
+  OP(bool, <=)
+  OP(bool, >=)
+#undef OP
+#else // __SYCL_DEVICE_ONLY__
+#define OP(type, op)                                                           \
+  friend type operator op(                                                     \
+      const wi_element<sycl::ext::oneapi::experimental::bfloat16, NumRows,     \
+                       NumCols, Layout, Group> &,                              \
+      const sycl::ext::oneapi::experimental::bfloat16 &) {                     \
+    throw runtime_error("joint matrix is not supported on host device.",       \
+                        PI_INVALID_DEVICE);                                    \
+  }                                                                            \
+  friend type operator op(                                                     \
+      const sycl::ext::oneapi::experimental::bfloat16 &,                       \
+      const wi_element<sycl::ext::oneapi::experimental::bfloat16, NumRows,     \
+                       NumCols, Layout, Group> &) {                            \
+    throw runtime_error("joint matrix is not supported on host device.",       \
+                        PI_INVALID_DEVICE);                                    \
+  }
+  OP(sycl::ext::oneapi::experimental::bfloat16, +)
+  OP(sycl::ext::oneapi::experimental::bfloat16, -)
+  OP(sycl::ext::oneapi::experimental::bfloat16, *)
+  OP(sycl::ext::oneapi::experimental::bfloat16, /)
+  OP(bool, ==)
+  OP(bool, !=)
+  OP(bool, <)
+  OP(bool, >)
+  OP(bool, <=)
+  OP(bool, >=)
+#undef OP
+#endif // __SYCL_DEVICE_ONLY__
+};
+
 template <typename T, size_t NumRows, size_t NumCols, matrix_layout Layout,
           typename Group>
 class wi_data {

sycl/test/matrix/matrix-bfloat16-test.cpp

@@ -0,0 +1,191 @@
+// RUN: %clangxx -fsycl -O2 %s -o %t.out
+#include <CL/sycl.hpp>
+#if (SYCL_EXT_ONEAPI_MATRIX == 2)
+#include <iostream>
+
+using namespace sycl::ext::oneapi::experimental::matrix;
+using bfloat16 = sycl::ext::oneapi::experimental::bfloat16;
+
+static constexpr auto TILE_SZ = 16;
+static constexpr auto TM = TILE_SZ - 1;
+static constexpr auto TN = TILE_SZ - 1;
+static constexpr auto TK = 2 * TILE_SZ - 2;
+
+static constexpr auto SG_SZ = 16;
+
+template <typename T, size_t NUM_ROWS, size_t NUM_COLS> struct big_matrix {
+public:
+  T *mat;
+
+public:
+  T *get_data() { return mat; }
+  void set_data(T *data) { mat = data; }
+  big_matrix(T *data) : mat(data) {}
+};
+
+template <typename T1, typename T2, size_t NUM_ROWS_A, size_t NUM_COLS_A,
+          size_t NUM_ROWS_B, size_t NUM_COLS_B, size_t NUM_ROWS_C,
+          size_t NUM_COLS_C>
+void matrix_multiply(big_matrix<T1, NUM_ROWS_C, NUM_COLS_C> &C,
+                     big_matrix<T2, NUM_ROWS_A, NUM_COLS_A> &A,
+                     big_matrix<T2, NUM_ROWS_B, NUM_COLS_B> &B) {
+  size_t M = NUM_ROWS_C;
+  size_t N = NUM_COLS_C;
+  size_t K = NUM_COLS_A;
+  // B => K/4 x N*4, A => M x K, C => M, N
+  // stride should be X's cols, e.g., B's stirde = N*4
+  assert(NUM_ROWS_C == NUM_ROWS_A && NUM_COLS_A == NUM_ROWS_B * 2);
+  size_t NDRangeM = M / TM;
+  size_t NDRangeN = N / TN;
+  sycl::buffer<bfloat16, 2> bufA(A.get_data(), sycl::range<2>(M, K));
+  sycl::buffer<bfloat16, 2> bufB(B.get_data(), sycl::range<2>(K, N));
+  sycl::buffer<float, 2> bufC((float *)C.get_data(), sycl::range<2>(M, N));
+
+  sycl::queue q;
+  q.submit([&](sycl::handler &cgh) {
+     auto accC = bufC.get_access<sycl::access::mode::read_write>(cgh);
+     auto accA = bufA.get_access<sycl::access::mode::read_write>(cgh);
+     auto accB = bufB.get_access<sycl::access::mode::read_write>(cgh);
+
+     cgh.parallel_for<class imatrix>(
+         sycl::nd_range<2>({NDRangeM, NDRangeN * SG_SZ}, {1, 1 * SG_SZ}),
+         [accA, accB, accC, M, N, K](sycl::nd_item<2> spmd_item)
+
+         {
+           // The submatrix API has to be accessed by all the workitems in a
+           // subgroup these functions will be called once by the subgroup no
+           // code divergence between the workitems
+           const auto global_idx = spmd_item.get_global_id(0);
+           const auto global_idy = spmd_item.get_global_id(1);
+           const auto sg_startx = global_idx - spmd_item.get_local_id(0);
+           const auto sg_starty = global_idy - spmd_item.get_local_id(1);
+
+           sycl::ext::oneapi::sub_group sg = spmd_item.get_sub_group();
+           joint_matrix<bfloat16, TM, TK> sub_a(sg);
+           // For B, since current implementation does not support non-packed
+           // layout, users need to specify the updated VNNI sizes along with
+           // the packed_b layout. By default, the layout is row_major and size
+           // is (TK, TN).
+           joint_matrix<bfloat16, TK, TN, matrix_layout::packed_b> sub_b(sg);
+           joint_matrix<float, TM, TN> sub_c(sg);
+
+           // AMX: 8 register tiles : 1k byte size, SMmaxxSKmax =16x64
+           // strideX = X's cols, so strideC = N, strideA = K, strideB = N*4
+           joint_matrix_load(sg, sub_c,
+                             accC.get_pointer() + (sg_startx * TM) * N +
+                                 sg_starty / SG_SZ * TN,
+                             N, matrix_layout::row_major);
+           for (int k = 0; k < K / TK; k += 1) { //
+             joint_matrix_load(
+                 sg, sub_a, accA.get_pointer() + (sg_startx * TM) * K + k * TK,
+                 K, matrix_layout::row_major);
+             // Assuming B data is already in VNNI format.
+             joint_matrix_load(sg, sub_b,
+                               accB.get_pointer() + (k * TK / 2) * (N * 2) +
+                                   sg_starty / SG_SZ * TN * 2,
+                               N * 2, matrix_layout::packed_b);
+             sub_c = joint_matrix_mad(sg, sub_a, sub_b, sub_c);
+           }
+           joint_matrix_store(sg, sub_c,
+                              accC.get_pointer() + (sg_startx * TM) * N +
+                                  sg_starty / SG_SZ * TN,
+                              N, matrix_layout::row_major);
+         }); // parallel for
+   }).wait();
+}
+
+static constexpr size_t MATRIX_M = TM * 2;
+static constexpr size_t MATRIX_N = TN * 2;
+static constexpr size_t MATRIX_K = TK * 2;
+bfloat16 A[MATRIX_M][MATRIX_K];
+bfloat16 B[MATRIX_K / 2][MATRIX_N * 2];
+unsigned short Aref[MATRIX_M][MATRIX_K];
+unsigned short Bref[MATRIX_K / 2][MATRIX_N * 2];
+float C[MATRIX_M][MATRIX_N];
+float D[MATRIX_M][MATRIX_N];
+
+float make_fp32(short x) {
+  unsigned int y = x;
+  y = y << 16;
+  float *res = reinterpret_cast<float *>(&y);
+  return *res;
+}
+
+unsigned short make_bf16(float x) {
+  int *res = reinterpret_cast<int *>(&x);
+  *res = *res >> 16;
+  return (unsigned short)*res;
+}
+
+void matrix_multiply_ref(int *A_mem, int *B_mem, int *C_mem, int M, int N,
+                         int K) {
+  // tiling
+  for (int m = 0; m < M; m++)
+    for (int n = 0; n < N; n++) {
+      for (int k = 0; k < K; k++) {
+        short *va = (short *)(A_mem + m * K + k);
+        short *vb = (short *)(B_mem + k * N + n);
+        float acc = *((float *)(C_mem + m * N + n));
+        // FIXME: Should we do reduce-add in another version?
+        for (int i = 0; i < 2; i++) {
+          acc += (make_fp32(va[i]) * make_fp32(vb[i]));
+        }
+        *((float *)(C_mem + m * N + n)) = acc;
+      }
+    }
+}
+
+int main() {
+  for (int i = 0; i < MATRIX_M; i++) {
+    for (int j = 0; j < MATRIX_K; j++) {
+      // Ee create bfloat16 from unsigned short since float-to-bfloat's
+      // conversion is not allowed.
+      A[i][j] = bfloat16::from_bits(make_bf16(1.0f * (i + j)));
+      Aref[i][j] = make_bf16(1.0f * (i + j));
+    }
+  }
+  for (int i = 0; i < MATRIX_K / 2; i++) {
+    for (int j = 0; j < MATRIX_N * 2; j++) {
+      B[i][j] = bfloat16::from_bits((make_bf16(2.0f * i + 3.0f * j)));
+      Bref[i][j] = make_bf16(2.0f * i + 3.0f * j);
+    }
+  }
+  for (int i = 0; i < MATRIX_M; i++) {
+    for (int j = 0; j < MATRIX_N; j++) {
+      C[i][j] = 1.0;
+      D[i][j] = 1.0;
+    }
+  }
+
+  big_matrix<float, MATRIX_M, MATRIX_N> MC((float *)&C);
+  big_matrix<float, MATRIX_M, MATRIX_N> MD((float *)&D);
+  big_matrix<bfloat16, MATRIX_M, MATRIX_K> MA((bfloat16 *)&A);
+  big_matrix<bfloat16, MATRIX_K / 2, MATRIX_N * 2> MB((bfloat16 *)&B);
+  matrix_multiply(MC, MA, MB);
+  matrix_multiply_ref((int32_t *)Aref, (int32_t *)Bref, (int32_t *)D, MATRIX_M,
+                      MATRIX_N, MATRIX_K / 2);
+
+  bool res = true;
+  for (int i = 0; i < MATRIX_M; i++) {
+    for (int j = 0; j < MATRIX_N; j++) {
+      if (C[i][j] != D[i][j])
+        res = false;
+    }
+  }
+  if (res)
+    std::cout << "passed\n";
+  else
+    std::cout << "failed\n";
+  for (int i = 0; i < MATRIX_M; i++) {
+    for (int j = 0; j < MATRIX_N; j++)
+      std::cout << C[i][j] << ", ";
+    std::cout << "\n";
+  }
+  std::cout << std::endl;
+  for (int i = 0; i < MATRIX_M; i++) {
+    for (int j = 0; j < MATRIX_N; j++)
+      std::cout << D[i][j] << ", ";
+    std::cout << "\n";
+  }
+}
+#endif // (SYCL_EXT_ONEAPI_MATRIX == 2)