Adding SME1 based kernel for FP32 GEMM support

filelist.json

@@ -1807,6 +1807,7 @@
               "src/core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_1VLx4VL/generic.cpp",
               "src/core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_2VLx2VL/generic.cpp",
               "src/core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_4VLx1VL/generic.cpp",
+              "src/core/NEON/kernels/arm_gemm/kernels/sme1_interleaved_nomerge_fp32_mopa_2VLx2VL/generic.cpp",
               "src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_bf16fp32_dot_6x4VL/generic.cpp",
               "src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_bf16fp32_mmla_6x4VL/generic.cpp",
               "src/core/NEON/kernels/arm_gemm/kernels/sve_hybrid_fp16_mla_6x4VL/a64fx.cpp",

src/BUILD.bazel

@@ -282,6 +282,7 @@ filegroup(
 	"core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_1VLx4VL/generic.cpp",
 	"core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_2VLx2VL/generic.cpp",
 	"core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_4VLx1VL/generic.cpp",
+	"core/NEON/kernels/arm_gemm/kernels/sme1_interleaved_nomerge_fp32_mopa_2VLx2VL/generic.cpp",
 	"core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_bf16fp32_mmla_6x4VL/generic.cpp",
 	"core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_fp16_mla_6x4VL/a64fx.cpp",
 	"core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_fp16_mla_6x4VL/generic.cpp",

src/CMakeLists.txt

@@ -251,6 +251,7 @@ target_sources(
 	core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_1VLx4VL/generic.cpp
 	core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_2VLx2VL/generic.cpp
 	core/NEON/kernels/arm_gemm/kernels/sme2_interleaved_nomerge_u8q_mopa_4VLx1VL/generic.cpp
+	core/NEON/kernels/arm_gemm/kernels/sme1_interleaved_nomerge_fp32_mopa_2VLx2VL/generic.cpp
 	core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_bf16fp32_mmla_6x4VL/generic.cpp
 	core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_fp16_mla_6x4VL/a64fx.cpp
 	core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_fp16_mla_6x4VL/generic.cpp
@@ -1081,4 +1082,4 @@ target_sources(
 	cpu/kernels/select/generic/neon/fp16.cpp
 	cpu/kernels/softmax/generic/neon/fp16.cpp
 	cpu/kernels/sub/neon/fp16.cpp
-)
+)

src/core/NEON/kernels/arm_gemm/gemm_fp32.cpp

@@ -56,6 +56,11 @@
 #include "kernels/sve_ffinterleaved_fp32_mla_8x3VL.hpp"
 #include "kernels/sve_ffinterleaved_bf16fp32_mmla_8x3VL.hpp"
 #endif // ARM_COMPUTE_ENABLE_FIXED_FORMAT_KERNELS
+
+#ifdef ARM_COMPUTE_ENABLE_SME
+#include "kernels/sme1_interleaved_nomerge_fp32_mopa_2VLx2VL.hpp"
+#endif // ARM_COMPUTE_ENABLE_SME
+
 #ifdef ARM_COMPUTE_ENABLE_SME2
 #include "kernels/sme2_gemv_fp32_mla_16VL.hpp"
 #include "kernels/sme2_gemv_fp32bf16fp32_dot_16VL.hpp"
@@ -118,6 +123,15 @@ GemmImplementation<float, float, float>::with_estimate(
     [](const GemmArgs &args) { return new GemmHybridIndirect<cls_a64_hybrid_fp32bf16fp32_mmla_4x24, float, float, float>(args); }
 ),
 #endif // ARM_COMPUTE_ENABLE_BF16
+#ifdef ARM_COMPUTE_ENABLE_SME
+{
+    GemmMethod::GEMM_INTERLEAVED,
+    "sme1_interleaved_nomerge_fp32_mopa_2VLx2VL",
+    [](const GemmArgs &args) { return args._ci->has_sme() && !args._accumulate; },
+    nullptr,
+    [](const GemmArgs &args) { return new GemmInterleavedNoMerge<cls_sme1_interleaved_nomerge_fp32_mopa_2VLx2VL, float, float>(args); }
+},
+#endif // ARM_COMPUTE_ENABLE_SME
 #ifdef ARM_COMPUTE_ENABLE_SVE
 #ifdef ARM_COMPUTE_ENABLE_SME2
 // SME kernels
@@ -187,6 +201,7 @@ GemmImplementation<float, float, float>::with_estimate(
     nullptr,
     [](const GemmArgs &args) { return new GemmInterleavedNoMerge<cls_sme2_interleaved_nomerge_fp32_mopa_2VLx2VL, float, float>(args); }
 },
+
 #endif // ARM_COMPUTE_ENABLE_SME2
 #ifdef ARM_COMPUTE_ENABLE_BF16
 GemmImplementation<float, float, float>::with_estimate(

src/core/NEON/kernels/arm_gemm/kernels/sme1_interleaved_nomerge_fp32_mopa_2VLx2VL.hpp

@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2022-2024 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+// + Changes from Qualcomm Technologies, Inc. are provided under the following license:
+// + Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries.
+// + SPDX-License-Identifier:  MIT
+//
+
+
+#pragma once
+
+#ifdef ARM_COMPUTE_ENABLE_SME
+
+
+#include "../std_transforms_sme.hpp"
+
+namespace arm_gemm
+{
+
+// Implementations
+void sme1_interleaved_nomerge_fp32_mopa_2VLx2VL(const float *const A, const float *const B, float *const C, int ldc, const int M, const int N, const int K, const float *const bias, const Activation act, bool accumulate, float *const accumulator_buffer);
+
+class cls_sme1_interleaved_nomerge_fp32_mopa_2VLx2VL
+{
+public:
+  typedef float lhs_operand_type;
+  typedef float rhs_operand_type;
+  typedef float result_type;
+
+  typedef void (*kern_type)(const float *const A, const float *const B, float *const C, int ldc, const int M, const int N, const int K, const float *const bias, const Activation act, bool accumulate, float *const accumulator_buffer);
+
+  /* Kernel blocking parameters */
+  static unsigned int out_height()
+  {
+    return sme::get_vector_length<float>() * 2;
+  }
+
+  static unsigned int out_width()
+  {
+    return sme::get_vector_length<float>() * 2;
+  }
+
+  static constexpr unsigned int k_unroll()
+  {
+    return 1;
+  }
+
+  static constexpr bool supports_accumulate()
+  {
+    return true;
+  }
+
+  static constexpr bool supports_bias()
+  {
+    return true;
+  }
+
+  static constexpr bool supports_activation()
+  {
+    return true;
+  }
+
+  static constexpr bool is_sme()
+  {
+    return true;
+  }
+
+  // Default to the generic kernel
+  kern_type kernel = sme1_interleaved_nomerge_fp32_mopa_2VLx2VL;
+
+  StdTransformsSME<lhs_operand_type, result_type, 2, 2, 1> transforms = {};
+
+  cls_sme1_interleaved_nomerge_fp32_mopa_2VLx2VL(const CPUInfo *)
+  {
+  }
+};
+
+} // namespace arm_gemm
+
+#endif // ARM_COMPUTE_ENABLE_SME