feat(dnn/fallback): support mk4 fp32 im2col

GitOrigin-RevId: 178d7231726c18bbc2586797f2b14e4ef3fdb969

dnn/src/aarch64/matrix_mul/int8/kernel_mk4_4x4x16.h

@@ -749,7 +749,7 @@ static void gemm_mk4_s8_4x4_pack_A(dt_int8* outptr, const dt_int8* inptr,
         const int8_t* inptr1 = inptr0 + ldin;
         const int8_t* inptr2 = inptr1 + ldin;
         const int8_t* inptr3 = inptr2 + ldin;
-        int8_t* output = outptr + start_y * out_offset;
+        int8_t* output = outptr + (y - y0) / 4 * out_offset;
         prefetch_2x(inptr0);
         prefetch_2x(inptr1);
         prefetch_2x(inptr2);
@@ -776,7 +776,7 @@ static void gemm_mk4_s8_4x4_pack_A(dt_int8* outptr, const dt_int8* inptr,
     }
     for (; y + 3 < ymax; y += 4, start_y++) {
         const int8_t* inptr0 = inptr + start_y * ldin + k0 * 4;
-        int8_t* output = outptr + start_y * out_offset;
+        int8_t* output = outptr + (y - y0) / 4 * out_offset;
         prefetch_2x(inptr0);
         int K = kmax - k0;
         for (; K > 15; K -= 16) {

dnn/src/armv7/matrix_mul/int8/kernel_mk4_4x2x16.h

@@ -227,7 +227,7 @@ static void gemm_mk4_s8_4x2_pack_A(dt_int8* outptr, const dt_int8* inptr,
         const int8_t* inptr1 = inptr0 + ldin;
         const int8_t* inptr2 = inptr1 + ldin;
         const int8_t* inptr3 = inptr2 + ldin;
-        int8_t* output = outptr + start_y * out_offset;
+        int8_t* output = outptr + (y - y0) / 4 * out_offset;
         prefetch_2x(inptr0);
         prefetch_2x(inptr1);
         prefetch_2x(inptr2);
@@ -254,7 +254,7 @@ static void gemm_mk4_s8_4x2_pack_A(dt_int8* outptr, const dt_int8* inptr,
     }
     for (; y + 3 < ymax; y += 4, start_y++) {
         const int8_t* inptr0 = inptr + start_y * ldin + k0 * 4;
-        int8_t* output = outptr + start_y * out_offset;
+        int8_t* output = outptr + (y - y0) / 4 * out_offset;
         prefetch_2x(inptr0);
         int K = kmax - k0;
         for (; K > 15; K -= 16) {

dnn/src/fallback/conv_bias/conv1x1/conv1x1_strategy.cpp

@@ -22,20 +22,6 @@ namespace conv1x1 {
 
 namespace {
 
-size_t get_format_pack_size(param::ConvBias::Format format) {
-    switch(format){
-        case param::ConvBias::Format::NCHW44:
-        case param::ConvBias::Format::NCHW4:
-            return 4_z;
-        case param::ConvBias::Format::NCHW88:
-            return 8_z;
-        case param::ConvBias::Format::NCHW:
-            return 1_z;
-        default:
-            megdnn_throw("unknow pack size of the format");
-    }
-}
-
 struct StrategyHashParam {
     ConvBiasImpl::NCBKernSizeParam param;
     param::ConvBias::Format format;

dnn/src/fallback/conv_bias/im2col/algos.cpp

@@ -125,13 +125,10 @@ public:
             size_t oc_tile_size) {
         size_t IC = param.filter_meta.icpg, FH = param.filter_meta.spatial[0],
                FW = param.filter_meta.spatial[1];
-        size_t pack_oc_size = 1;
+        size_t pack_oc_size = get_format_pack_size(param.filter_meta.format);
         size_t im2col = 0, packb = 0, bias_temp = 0;
         bool default_pack = matmul_algo->packmode() == Pack_Mode::DEFAULT;
         megdnn_assert(default_pack, "only support default packa");
-        if (param.filter_meta.format == param::ConvBias::Format::NCHW44) {
-            pack_oc_size = 4;
-        }
         size_t im2col_dst_size =
                 IC * FH * FW * ohw_tile_size * sizeof(param.src_type);
         size_t matmul_dst_size = pack_oc_size * oc_tile_size * ohw_tile_size *
@@ -321,14 +318,17 @@ fallback::MatrixMulImpl::KernSizeParam
 ConvBiasImpl::AlgoIm2col ::get_matmul_kern_param(const NCBKernSizeParam& param,
                                                  size_t ohw_tile_size,
                                                  size_t oc_tile_size) const {
-    bool is_nchw44 =
-            param.filter_meta.format == param::ConvBias::Format::NCHW44;
+    auto format = param::MatrixMul::Format::DEFAULT;
+    size_t pack_oc_size = get_format_pack_size(param.filter_meta.format);
+    if (param.filter_meta.format == param::ConvBias::Format::NCHW44) {
+        format = param::MatrixMul::Format::MK4;
+    }
     size_t M = oc_tile_size;
     size_t N = ohw_tile_size;
     size_t K = param.filter_meta.icpg * param.filter_meta.spatial[0] *
                param.filter_meta.spatial[1];
-    size_t pack_oc_size = is_nchw44 ? 4 : 1;
-    size_t LDA = pack_oc_size * K, LDB = pack_oc_size * N, LDC = N;
+    size_t LDA = pack_oc_size * K, LDB = pack_oc_size * N,
+           LDC = N * pack_oc_size;
     bool is_dst_8bit = (param.src_type.enumv() == DTypeEnum::QuantizedS8 &&
                         param.dst_type.enumv() == DTypeEnum::QuantizedS8) ||
                        (param.src_type.enumv() == DTypeEnum::Quantized8Asymm &&
@@ -345,8 +345,7 @@ ConvBiasImpl::AlgoIm2col ::get_matmul_kern_param(const NCBKernSizeParam& param,
             false,
             false,
             param::MatrixMul::ComputeMode::DEFAULT,
-            is_nchw44 ? param::MatrixMul::Format::MK4
-                      : param::MatrixMul::Format::DEFAULT};
+            format};
 }
 
 void ConvBiasImpl::AlgoIm2col::choice_ohw_oc_block(
@@ -356,11 +355,7 @@ void ConvBiasImpl::AlgoIm2col::choice_ohw_oc_block(
     size_t nr_threads = param.nr_threads;
     size_t OC = param.filter_meta.ocpg;
     size_t ohw = param.osz[0] * param.osz[1];
-    //! pay attention please, should not change the 2 line code,
-    //! the opr use the same im2col algo, via choice_ohw_oc_block may change the
-    //! m_ohw_tile_size and m_oc_tile_size， if the two value changed, the
-    //! workspace size may change, will ocur workspace not match problem, so
-    //! should use the original data init them to avoid the problem
+
     oc_tile_size = DEFAULT_OC_TILE_SIZE;
     ohw_tile_size = m_ohw_tile_size;
 
@@ -505,14 +500,13 @@ SmallVector<ConvBiasImpl::NCBKern> ConvBiasImpl::AlgoIm2col::dispatch_kerns(
         size_t ohw_parallel_times = div_ceil(ohw, ohw_tile_size);
         size_t oc_parallel_times = div_ceil<size_t>(OC, oc_tile_size);
         size_t packa_parallel_times = 0;
-        size_t pack_oc_size =
-                (param.filter_meta.format == param::ConvBias::Format::NCHW ? 1
-                                                                           : 4);
+        size_t pack_oc_size = get_format_pack_size(param.filter_meta.format);
+
         if (only_packA) {
             packa_parallel_times = div_ceil<size_t>(OC, oc_tile_size);
         } else if (default_pack) {
             packa_parallel_times = div_ceil<size_t>(
-                    OC, m_matmul_algo->get_inner_block_size().m * pack_oc_size);
+                    OC, m_matmul_algo->get_inner_block_size().m);
         }
 
         auto matmul_param = get_matmul_kern_param(
@@ -659,12 +653,16 @@ bool ConvBiasImpl::AlgoIm2col::usable(
             param.nonlineMode != megdnn::NonlineMode::IDENTITY) {
             return false;
         }
-        //! current now im2col only support int8 quantized s8 nchw44
-        if (opr->param().format == param::ConvBias::Format::NCHW44 &&
-            (param.src_type.enumv() == param.filter_type.enumv() &&
-             (param.src_type.enumv() != DTypeEnum::Int8) &&
-             (param.src_type.enumv() != DTypeEnum::QuantizedS8))) {
+        if (opr->param().format == param::ConvBias::Format::NCHW44) {
+            //! current NCHW44 im2col only support DEFAULT mode matmul
+            if(m_matmul_algo->packmode() != Pack_Mode::DEFAULT) {
                 return false;
+                //! nchw44 hybird mode and channel wise is not support
+            } else if (param.filter_meta.icpg < 4_z ||
+                       param.filter_meta.icpg == 1 ||
+                       param.filter_meta.ocpg == 1) {
+                return false;
+            }
         }
 
         size_t oc_tile_size = 0, ohw_tile_size = 0;

dnn/src/fallback/conv_bias/im2col/factory.h

@@ -221,8 +221,17 @@ public:
         param::ConvBias::Format format = param.filter_meta.format;
         switch (strategytype) {
             case StrategyType::FLOAT:
+                if (format == param::ConvBias::Format::NCHW) {
                     cb1(NCHW, DEFAULT, dt_float32, dt_float32,
-                    PostprocessMode::FLOAT, "DefaultStrategyType::FLOAT"_hash);
+                        PostprocessMode::FLOAT,
+                        "DefaultStrategyType::FLOAT"_hash);
+                } else if (format == param::ConvBias::Format::NCHW44) {
+                    cb1(NCHW44, DEFAULT, dt_float32, dt_float32,
+                        PostprocessMode::FLOAT,
+                        "DefaultStrategyTypeNCHW44::FLOAT"_hash);
+                } else {
+                    megdnn_throw("not support format except nchw44 and nchw\n");
+                }
                 break;
 #if __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
             case StrategyType::FLOAT_FP16:

dnn/src/fallback/conv_bias/im2col/strategy_base.h

@@ -75,15 +75,14 @@ public:
 template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode, PackMode packmode,
-          FormatMode format>
+          FormatMode format = FormatMode::NCHW>
 class Strategy;
 
 template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 class Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-               postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>
-        : public StrategyBase {
+               postprocess_mode, PackMode::DEFAULT> : public StrategyBase {
 public:
     constexpr static size_t BUNDLE_PADDING_INDEX = 0;
     constexpr static size_t BUNDLE_PACKA_INDEX = 1;
@@ -142,8 +141,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
 class Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
                postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW44>
         : public Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-                          postprocess_mode, PackMode::DEFAULT,
-                          FormatMode::NCHW> {
+                          postprocess_mode, PackMode::DEFAULT> {
 public:
     const size_t BUNDLE_PADDING_INDEX = 0;
     const size_t BUNDLE_PACKA_INDEX = 1;
@@ -164,8 +162,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 class Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-               postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>
-        : public StrategyBase {
+               postprocess_mode, PackMode::NO_PACK> : public StrategyBase {
 public:
     constexpr static size_t BUNDLE_PADDING_INDEX = 0;
     constexpr static size_t BUNDLE_PACKA_INDEX = 1;
@@ -231,8 +228,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 class Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-               postprocess_mode, PackMode::ONLY_PACKA, FormatMode::NCHW>
-        : public StrategyBase {
+               postprocess_mode, PackMode::ONLY_PACKA> : public StrategyBase {
 public:
     constexpr static size_t BUNDLE_PADDING_INDEX = 0;
     constexpr static size_t BUNDLE_PACKA_INDEX = 1;

dnn/src/fallback/conv_bias/im2col/strategy_default.cpp

@@ -26,7 +26,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+              postprocess_mode, PackMode::DEFAULT>::
         copy_padding_kern(WorkspaceBundle bundle,
                           const fallback::ConvBiasImpl::NCBKernParam& param,
                           const fallback::ConvBiasImpl::NCBKernIndex& ncb_index,
@@ -93,13 +93,13 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+              postprocess_mode, PackMode::DEFAULT>::
         packA_kern(WorkspaceBundle bundle,
                    const fallback::ConvBiasImpl::NCBKernParam& param,
                    fallback::MatrixMulImpl::KernSizeParam matmulparam,
                    fallback::MatrixMulImpl::AlgoBase* matmul_algo,
                    const fallback::ConvBiasImpl::NCBKernIndex& ncb_index,
-                   size_t pack_oc_size) {
+                   size_t) {
     bundle.set(param.workspace_ptr);
     fallback::MatrixMulImpl::KernParam matmul_param;
     size_t group_id = ncb_index.ndrange_id[0];
@@ -112,19 +112,18 @@ void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
             matmul_algo->get_packA_type_size();
     size_t a_panel_offset = ncb_index.ndrange_id[1] * packed_per_oc_block_size;
     int8_t* a_panel = static_cast<int8_t*>(bundle.get(BUNDLE_PACKA_INDEX)) +
-                      group_id * packA_group_size +
-                      (pack_oc_size == 4 ? 0 : a_panel_offset);
+                      group_id * packA_group_size + a_panel_offset;
     matmul_param.A_ptr =
             const_cast<src_ctype*>(param.filter<src_ctype>(group_id));
     matmul_algo->pack_A(matmul_param, a_panel, ncb_index.ndrange_id[1],
-                        matmul_algo->get_inner_block_size().m * pack_oc_size);
+                        matmul_algo->get_inner_block_size().m);
 }
 
 template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+              postprocess_mode, PackMode::DEFAULT>::
         exec_im2col(WorkspaceBundle bundle, WorkspaceBundle bundle_thread,
                     const StrategyParam& sparam,
                     const fallback::ConvBiasImpl::NCBKernParam& param,
@@ -193,7 +192,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void* Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-               postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+               postprocess_mode, PackMode::DEFAULT>::
         get_matmul_dst_ptr(const fallback::ConvBiasImpl::NCBKernParam& param,
                            const WorkspaceBundle& bundle_thread,
                            const StrategyParam& sparam) {
@@ -212,7 +211,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+              postprocess_mode, PackMode::DEFAULT>::
         exec_matmul(const fallback::ConvBiasImpl::NCBKernParam& param,
                     const StrategyParam& sparam, WorkspaceBundle bundle,
                     WorkspaceBundle bundle_thread,
@@ -249,7 +248,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+              postprocess_mode, PackMode::DEFAULT>::
         exec_postprocess(const fallback::ConvBiasImpl::NCBKernParam& param,
                          const StrategyParam& sparam,
                          WorkspaceBundle bundle_thread) {
@@ -264,12 +263,12 @@ void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
                     ? bias_temp_ptr
                     : static_cast<void*>(const_cast<bias_ctype*>(
                               bias_ptr + sparam.oc_cur_index)));
-
+    size_t pack_oc_size = sparam.pack_oc_size;
     PostProcess<op_ctype, op_dtype, postprocess_mode>::run(
             matmul_dst, bias_preprocess_ptr, matmul_dst, param.bias_mode,
             param.nonlineMode, param.bias_type, param.dst_type, 1_z,
-            sparam.output_block_oc_size, 1_z, sparam.output_block_size,
-            sparam.pack_oc_size);
+            sparam.output_block_oc_size / pack_oc_size, 1_z,
+            sparam.output_block_size, pack_oc_size);
     copy_dst(param, matmul_dst, sparam);
 }
 
@@ -277,7 +276,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+              postprocess_mode, PackMode::DEFAULT>::
         copy_dst(const fallback::ConvBiasImpl::NCBKernParam& param,
                  const void* matmul_dst, const StrategyParam& sparam) {
     if (!sparam.skip_copy_dst) {
@@ -303,7 +302,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void* Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-               postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+               postprocess_mode, PackMode::DEFAULT>::
         get_bias_temp_ptr(const fallback::ConvBiasImpl::NCBKernParam& param,
                           const WorkspaceBundle& bundle_thread) {
     bias_ctype* bias_tmp_ptr =
@@ -318,7 +317,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::DEFAULT, FormatMode::NCHW>::
+              postprocess_mode, PackMode::DEFAULT>::
         copy_bias(const fallback::ConvBiasImpl::NCBKernParam& param,
                   WorkspaceBundle bundle_thread, const StrategyParam& sparam) {
     const bias_ctype* bias_ptr = static_cast<const bias_ctype*>(
@@ -342,8 +341,7 @@ void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
 #define INSTANTIAL_CLASS(_src_ctype, _bias_ctype, _dst_ctype, _op_ctype,    \
                          _op_dtype, _postprocess_mode)                      \
     template class Strategy<_src_ctype, _bias_ctype, _dst_ctype, _op_ctype, \
-                            _op_dtype, _postprocess_mode, PackMode::DEFAULT, \
-                            FormatMode::NCHW>;
+                            _op_dtype, _postprocess_mode, PackMode::DEFAULT>;
 
 INSTANTIAL_CLASS(dt_float32, dt_float32, dt_float32, dt_float32, dt_float32,
                  megdnn::PostprocessMode::FLOAT)

dnn/src/fallback/conv_bias/im2col/strategy_default_nchw44.cpp

@@ -12,10 +12,9 @@
 #include "src/fallback/convolution/img2col_helper.h"
 #if MEGDNN_X86
 #include "src/x86/conv_bias/postprocess_helper.h"
-#elif (MEGDNN_ARMV7 || MEGDNN_AARCH64)
-#include "src/arm_common/conv_bias/postprocess_helper.h"
 #endif
 
+
 using namespace megdnn;
 #if MEGDNN_X86
 using namespace x86;
@@ -101,23 +100,12 @@ void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
 
 INSTANTIAL_CLASS(dt_float32, dt_float32, dt_float32, dt_float32, dt_float32,
                  megdnn::PostprocessMode::FLOAT)
-#if __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-INSTANTIAL_CLASS(dt_float16, dt_float16, dt_float16, __fp16, __fp16,
-                 megdnn::PostprocessMode::FLOAT)
-#else
+
 #if !MEGDNN_DISABLE_FLOAT16
 INSTANTIAL_CLASS(dt_float16, dt_float16, dt_float16, dt_float16, dt_float16,
                  megdnn::PostprocessMode::NO_PROCESS)
 #endif
-#endif
 
-#if MEGDNN_AARCH64 || MEGDNN_ARMV7
-//! x86 do not have uint8 matmul so only armv7 armv8 support uint8
-INSTANTIAL_CLASS(dt_uint8, dt_int32, dt_uint8, dt_qint32, dt_quint8,
-                 megdnn::PostprocessMode::QUANTIZED)
-INSTANTIAL_CLASS(dt_uint8, dt_int32, dt_int32, dt_qint32, dt_qint32,
-                 megdnn::PostprocessMode::NO_PROCESS)
-#endif
 
 INSTANTIAL_CLASS(dt_int8, dt_int32, dt_int8, dt_qint32, dt_qint8,
                  megdnn::PostprocessMode::QUANTIZED)

dnn/src/fallback/conv_bias/im2col/strategy_nopack.cpp

@@ -27,7 +27,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>::
+              postprocess_mode, PackMode::NO_PACK>::
         copy_padding_kern(WorkspaceBundle bundle,
                           const fallback::ConvBiasImpl::NCBKernParam& param,
                           const fallback::ConvBiasImpl::NCBKernIndex& ncb_index,
@@ -90,7 +90,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>::
+              postprocess_mode, PackMode::NO_PACK>::
         packA_kern(WorkspaceBundle bundle,
                    const fallback::ConvBiasImpl::NCBKernParam& param,
                    fallback::MatrixMulImpl::KernSizeParam matmulparam,
@@ -110,7 +110,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void* Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-               postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>::
+               postprocess_mode, PackMode::NO_PACK>::
         get_matmul_dst_ptr(const fallback::ConvBiasImpl::NCBKernParam& param,
                            const WorkspaceBundle& bundle_thread,
                            const StrategyParam& sparam) {
@@ -129,7 +129,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>::
+              postprocess_mode, PackMode::NO_PACK>::
         exec_matmul(const fallback::ConvBiasImpl::NCBKernParam& param,
                     const StrategyParam& sparam, WorkspaceBundle bundle,
                     WorkspaceBundle bundle_thread,
@@ -162,7 +162,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>::
+              postprocess_mode, PackMode::NO_PACK>::
         exec_im2col(WorkspaceBundle bundle, WorkspaceBundle bundle_thread,
                     const StrategyParam& sparam,
                     const fallback::ConvBiasImpl::NCBKernParam& param,
@@ -224,7 +224,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>::
+              postprocess_mode, PackMode::NO_PACK>::
         exec_postprocess(const fallback::ConvBiasImpl::NCBKernParam& param,
                          const StrategyParam& sparam,
                          WorkspaceBundle bundle_thread) {
@@ -252,7 +252,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>::
+              postprocess_mode, PackMode::NO_PACK>::
         copy_dst(const fallback::ConvBiasImpl::NCBKernParam& param,
                  const void* matmul_dst, const StrategyParam& sparam) {
     if (!sparam.skip_copy_dst) {
@@ -274,7 +274,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::NO_PACK, FormatMode::NCHW>::
+              postprocess_mode, PackMode::NO_PACK>::
         copy_bias(const fallback::ConvBiasImpl::NCBKernParam& param,
                   WorkspaceBundle bundle_thread, const StrategyParam& sparam) {
     const bias_ctype* bias_ptr = static_cast<const bias_ctype*>(
@@ -298,8 +298,7 @@ void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
 #define INSTANTIAL_CLASS(_src_ctype, _bias_ctype, _dst_ctype, _op_ctype,    \
                          _op_dtype, _postprocess_mode)                      \
     template class Strategy<_src_ctype, _bias_ctype, _dst_ctype, _op_ctype, \
-                            _op_dtype, _postprocess_mode, PackMode::NO_PACK, \
-                            FormatMode::NCHW>;
+                            _op_dtype, _postprocess_mode, PackMode::NO_PACK>;
 
 INSTANTIAL_CLASS(dt_float32, dt_float32, dt_float32, dt_float32, dt_float32,
                  megdnn::PostprocessMode::FLOAT)

dnn/src/fallback/conv_bias/im2col/strategy_onlypacka.cpp

@@ -27,7 +27,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::ONLY_PACKA, FormatMode::NCHW>::
+              postprocess_mode, PackMode::ONLY_PACKA>::
         copy_padding_kern(WorkspaceBundle bundle,
                           const fallback::ConvBiasImpl::NCBKernParam& param,
                           const fallback::ConvBiasImpl::NCBKernIndex& ncb_index,
@@ -90,7 +90,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::ONLY_PACKA, FormatMode::NCHW>::
+              postprocess_mode, PackMode::ONLY_PACKA>::
         packA_kern(WorkspaceBundle bundle,
                    const fallback::ConvBiasImpl::NCBKernParam& param,
                    fallback::MatrixMulImpl::KernSizeParam matmulparam,
@@ -124,7 +124,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void* Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-               postprocess_mode, PackMode::ONLY_PACKA, FormatMode::NCHW>::
+               postprocess_mode, PackMode::ONLY_PACKA>::
         get_matmul_dst_ptr(const fallback::ConvBiasImpl::NCBKernParam& param,
                            const WorkspaceBundle& bundle_thread,
                            const StrategyParam& sparam) {
@@ -143,7 +143,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::ONLY_PACKA, FormatMode::NCHW>::
+              postprocess_mode, PackMode::ONLY_PACKA>::
         exec_matmul(const fallback::ConvBiasImpl::NCBKernParam& param,
                     const StrategyParam& sparam, WorkspaceBundle bundle,
                     WorkspaceBundle bundle_thread,
@@ -181,7 +181,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::ONLY_PACKA, FormatMode::NCHW>::
+              postprocess_mode, PackMode::ONLY_PACKA>::
         exec_im2col(WorkspaceBundle bundle, WorkspaceBundle bundle_thread,
                     const StrategyParam& sparam,
                     const fallback::ConvBiasImpl::NCBKernParam& param,
@@ -242,7 +242,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::ONLY_PACKA, FormatMode::NCHW>::
+              postprocess_mode, PackMode::ONLY_PACKA>::
         exec_postprocess(const fallback::ConvBiasImpl::NCBKernParam& param,
                          const StrategyParam& sparam,
                          WorkspaceBundle bundle_thread) {
@@ -283,7 +283,7 @@ template <typename src_ctype, typename bias_ctype, typename dst_ctype,
           typename op_ctype, typename op_dtype,
           megdnn::PostprocessMode postprocess_mode>
 void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
-              postprocess_mode, PackMode::ONLY_PACKA, FormatMode::NCHW>::
+              postprocess_mode, PackMode::ONLY_PACKA>::
         copy_dst(const fallback::ConvBiasImpl::NCBKernParam& param,
                  const void* matmul_dst, const StrategyParam& sparam) {
     if (!sparam.skip_copy_dst) {
@@ -305,7 +305,7 @@ void Strategy<src_ctype, bias_ctype, dst_ctype, op_ctype, op_dtype,
                          _op_dtype, _postprocess_mode)                      \
     template class Strategy<_src_ctype, _bias_ctype, _dst_ctype, _op_ctype, \
                             _op_dtype, _postprocess_mode,                   \
-                            PackMode::ONLY_PACKA, FormatMode::NCHW>;
+                            PackMode::ONLY_PACKA>;
 
 INSTANTIAL_CLASS(dt_float32, dt_float32, dt_float32, dt_float32, dt_float32,
                  megdnn::PostprocessMode::FLOAT)

dnn/src/fallback/conv_bias/opr_impl.cpp

@@ -26,6 +26,18 @@
 using namespace megdnn;
 using namespace fallback;
 
+size_t megdnn::fallback::get_format_pack_size(param::ConvBias::Format format) {
+    switch(format){
+        case param::ConvBias::Format::NCHW44:
+        case param::ConvBias::Format::NCHW4:
+            return 4_z;
+        case param::ConvBias::Format::NCHW88:
+            return 8_z;
+        default:
+            return 1_z;
+    }
+}
+
 namespace {
 template <typename T>
 void incr_ptr(T*& dst, ptrdiff_t delta) {

dnn/src/fallback/conv_bias/opr_impl.h

@@ -21,6 +21,11 @@
 namespace megdnn {
 namespace fallback {
 
+/*!
+ * \brief get the pack_size according to the format
+ * */
+size_t get_format_pack_size(param::ConvBias::Format format);
+
 /*!
  * \brief fallback conv bias forward impl
  *

dnn/src/fallback/convolution/img2col_helper.h

@@ -9,9 +9,8 @@
  * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  */
 #include "src/common/utils.h"
-#if MEGDNN_ARMV7 || MEGDNN_AARCH64
-#include "src/arm_common/simd_macro/marm_neon.h"
-#endif
+
+
 namespace {
 
 template <bool is_xcorr, typename dtype>
@@ -268,12 +267,13 @@ void img2col_nchw4(const dtype* __restrict src, dtype* __restrict dst,
                         }
 
                         for (int w = cur_remain_w; w < OW; w++) {
-                            size_t index = ic * IH * IW + (start_h + fh2) * IW +
-                                           (w + fw2);
-                            dst[i++] = src[4 * index];
-                            dst[i++] = src[4 * index + 1];
-                            dst[i++] = src[4 * index + 2];
-                            dst[i++] = src[4 * index + 3];
+                            size_t index =
+                                    4 * (ic * IH * IW + (start_h + fh2) * IW +
+                                         (w + fw2));
+                            dst[i++] = src[index];
+                            dst[i++] = src[index + 1];
+                            dst[i++] = src[index + 2];
+                            dst[i++] = src[index + 3];
                         }
 
                         for (int h = start_h + 1; h < end_h; h++) {
@@ -317,26 +317,11 @@ void img2col_nchw4(const dtype* __restrict src, dtype* __restrict dst,
                             fh2 = FH - fh - 1;
                             fw2 = FW - fw - 1;
                         }
-#if MEGDNN_ARMV7 || MEGDNN_AARCH64
-                        int w = cur_remain_w;
-                        size_t index = (ic * IH * IW + (start_h + fh2) * IW +
-                                        (w + fw2));
-                        for (; w + 3 < end_remain_w; w += 4) {
-                            vst1q_u32(&output[i],
-                                      vld1q_u32(&uint32_src[index]));
-                            i += 4;
-                            index += 4;
-                        }
-                        for (; w < end_remain_w; w++) {
-                            output[i++] = uint32_src[index];
-                        }
-#else
                         for (int w = cur_remain_w; w < end_remain_w; w++) {
                             size_t index = (ic * IH * IW +
                                             (start_h + fh2) * IW + (w + fw2));
                             output[i++] = uint32_src[index];
                         }
-#endif
                     }
                 }
             }
@@ -360,27 +345,11 @@ void img2col_nchw4(const dtype* __restrict src, dtype* __restrict dst,
                         }
 
                         for (int h = start_h + 1; h < end_h; h++) {
-#if MEGDNN_ARMV7 || MEGDNN_AARCH64
-                            int ow = 0;
-                            size_t index = (ic * IH * IW + (h + fh2) * IW +
-                                            (ow + fw2));
-                            for (; ow + 3 < OW; ow += 4) {
-                                vst1q_u32(&output[i],
-                                          vld1q_u32(&uint32_src[index]));
-                                i += 4;
-                                index += 4;
-                            }
-
-                            for (; ow < OW; ow++) {
-                                output[i++] = uint32_src[index++];
-                            }
-#else
                             rep(ow, OW) {
                                 size_t index = (ic * IH * IW + (h + fh2) * IW +
                                                 (ow + fw2));
                                 output[i++] = uint32_src[index];
                             }
-#endif
                         }
 
                         for (int w = 0; w < end_remain_w; w++) {

dnn/test/arm_common/conv_bias_multi_thread.cpp

@@ -1173,10 +1173,10 @@ void checker_conv_bias_mul_int8x8x32(std::vector<conv_bias::TestArg> args,
 
 #if MEGDNN_AARCH64 || MEGDNN_ARMV7
 #if !__ARM_FEATURE_DOTPROD
-TEST_F(ARM_COMMON, CONV_BIAS_IM2COLMATMUL_INT8x8x32NCHW44) {
+TEST_F(ARM_COMMON_MULTI_THREADS, CONV_BIAS_IM2COLMATMUL_INT8x8x32NCHW44_S2) {
     using namespace conv_bias;
     std::vector<conv_bias::TestArg> args =
-            get_nchw44_conv_bias_args({2, 3, 4, 5, 6, 7}, 1, false, true, true);
+            get_nchw44_conv_bias_args({2, 5, 7}, 2, false, true, true);
 
 #define cb(name) checker_conv_bias_mul_int8x8x32(args, handle(), name);
 #if MEGDNN_AARCH64
@@ -1187,10 +1187,10 @@ TEST_F(ARM_COMMON, CONV_BIAS_IM2COLMATMUL_INT8x8x32NCHW44) {
 #undef cb
 }
 
-TEST_F(ARM_COMMON_MULTI_THREADS, CONV_BIAS_IM2COLMATMUL_INT8x8x32NCHW44_MULTI) {
+TEST_F(ARM_COMMON_MULTI_THREADS, CONV_BIAS_IM2COLMATMUL_INT8x8x32NCHW44_S1) {
     using namespace conv_bias;
     std::vector<conv_bias::TestArg> args =
-            get_nchw44_conv_bias_args({2, 3, 4, 5, 6, 7}, 1, false, true, true);
+            get_nchw44_conv_bias_args({3, 4, 6}, 1, false, true, true);
 
 #define cb(name) checker_conv_bias_mul_int8x8x32(args, handle(), name);
 #if MEGDNN_AARCH64
@@ -1202,12 +1202,13 @@ TEST_F(ARM_COMMON_MULTI_THREADS, CONV_BIAS_IM2COLMATMUL_INT8x8x32NCHW44_MULTI) {
 #undef cb
 }
 
-TEST_F(ARM_COMMON, CONV_BIAS_IM2COLMATMUL_QUANTIZEDSYM_NCHW44) {
+TEST_F(ARM_COMMON_MULTI_THREADS,
+       CONV_BIAS_IM2COLMATMUL_QUANTIZEDSYM_NCHW44_S2) {
     UniformIntRNG rng{-50, 50};
 
 #define cb(name)                                                               \
-    checker_conv_bias(get_nchw44_conv_bias_args({2, 3, 4, 5, 6, 7}, 1),     \
-                      handle(), &rng, epsilon, dtype::QuantizedS8(2.5f),    \
+    checker_conv_bias(get_nchw44_conv_bias_args({3, 4, 6}, 2), handle(), &rng, \
+                      epsilon, dtype::QuantizedS8(2.5f),                       \
                       dtype::QuantizedS8(2.5f), dtype::QuantizedS32(6.25f),    \
                       dtype::QuantizedS8(60.25f), name);
     float epsilon = 0.001;
@@ -1220,12 +1221,12 @@ TEST_F(ARM_COMMON, CONV_BIAS_IM2COLMATMUL_QUANTIZEDSYM_NCHW44) {
 }
 
 TEST_F(ARM_COMMON_MULTI_THREADS,
-       CONV_BIAS_IM2COLMATMUL_QUANTIZEDSYM_NCHW44_MULTI) {
+       CONV_BIAS_IM2COLMATMUL_QUANTIZEDSYM_NCHW44_S1) {
     UniformIntRNG rng{-50, 50};
 
 #define cb(name)                                                               \
-    checker_conv_bias(get_nchw44_conv_bias_args({2, 3, 4, 5, 6, 7}, 1),     \
-                      handle(), &rng, epsilon, dtype::QuantizedS8(2.5f),    \
+    checker_conv_bias(get_nchw44_conv_bias_args({2, 5, 7}, 1), handle(), &rng, \
+                      epsilon, dtype::QuantizedS8(2.5f),                       \
                       dtype::QuantizedS8(2.5f), dtype::QuantizedS32(6.25f),    \
                       dtype::QuantizedS8(60.25f), name);
     float epsilon = 0.001;
@@ -1286,6 +1287,24 @@ TEST_F(ARM_COMMON_MULTI_THREADS, CONV_BIAS_IM2COLMATMUL_INT8x8x32) {
 #undef cb
 }
 
+#if MEGDNN_AARCH64
+TEST_F(ARM_COMMON_MULTI_THREADS, CONV_BIAS_IM2COL_S1_MK4_PACK_F32) {
+    using namespace conv_bias;
+    std::vector<conv_bias::TestArg> args =
+            get_nchw44_conv_bias_args({2, 4, 7}, 1);
+    check_conv_bias(args, handle(), "IM2COLMATMUL:AARCH64_F32_MK4_K8X12X1");
+}
+#endif
+
+#if MEGDNN_AARCH64
+TEST_F(ARM_COMMON_MULTI_THREADS, CONV_BIAS_IM2COL_S2_MK4_PACK_F32) {
+    using namespace conv_bias;
+    std::vector<conv_bias::TestArg> args =
+            get_nchw44_conv_bias_args({3, 5, 6}, 2);
+    check_conv_bias(args, handle(), "IM2COLMATMUL:AARCH64_F32_MK4_K8X12X1");
+}
+#endif
+
 /***************************** Conv1x1 Algo Test ***********************/
 TEST_F(ARM_COMMON_MULTI_THREADS, CONV_BIAS_1X1_S1_F32) {
     using namespace conv_bias;