Unverified Commit 250d3c87 authored by Chris Austen's avatar Chris Austen Committed by GitHub
Browse files

Merge branch 'develop' into ck-flash-attn

parents 135eb63e f3939b99
......@@ -45,6 +45,9 @@ if(NOT WIN32)
endif()
rocm_clang_tidy_check(driver)
file(STRINGS "${CMAKE_SOURCE_DIR}/test/onnx/.onnxrt-commit" String_output)
target_compile_definitions(driver PUBLIC MIGRAPHX_ORT_SHA1="${String_output}")
target_link_libraries(driver migraphx_all_targets migraphx_onnx migraphx_tf migraphx_py)
rocm_install_targets(
......
......@@ -475,13 +475,15 @@ struct compiler
{
if(is_offload_copy_set(p) and not co.offload_copy)
{
std::cout << "MIGraphX program was likely compiled with offload_copy set, Try "
"passing "
"`--enable-offload-copy` if program run fails.\n";
std::cout
<< "[WARNING]: MIGraphX program was likely compiled with offload_copy "
"set, Try "
"passing "
"`--enable-offload-copy` if program run fails.\n";
}
else if(co.offload_copy)
{
std::cout << "MIGraphX program was likely compiled without "
std::cout << "[WARNING]: MIGraphX program was likely compiled without "
"offload_copy set, Try "
"removing "
"`--enable-offload-copy` flag if passed to driver, if program run "
......@@ -534,13 +536,19 @@ struct params : command<params>
struct verify : command<verify>
{
compiler c;
double tolerance = 80;
migraphx::verify::tolerance tols;
bool per_instruction = false;
bool reduce = false;
void parse(argument_parser& ap)
{
c.parse(ap);
ap(tolerance, {"--tolerance"}, ap.help("Tolerance for errors"));
ap(tols.rms_tol, {"--rms-tol"}, ap.help("Tolerance for the RMS error (Default: 0.001)"));
ap(tols.atol,
{"--atol"},
ap.help("Tolerance for the elementwise absolute difference (Default: 0.001)"));
ap(tols.rtol,
{"--rtol"},
ap.help("Tolerance for the elementwise relative difference (Default: 0.001)"));
ap(per_instruction,
{"-i", "--per-instruction"},
ap.help("Verify each instruction"),
......@@ -565,15 +573,15 @@ struct verify : command<verify>
if(per_instruction)
{
verify_instructions(p, t, c.co, quantize, tolerance);
verify_instructions(p, t, c.co, quantize, tols);
}
else if(reduce)
{
verify_reduced_program(p, t, c.co, quantize, m, tolerance);
verify_reduced_program(p, t, c.co, quantize, m, tols);
}
else
{
verify_program(c.l.file, p, t, c.co, quantize, m, tolerance);
verify_program(c.l.file, p, t, c.co, quantize, m, tols);
}
}
};
......@@ -802,6 +810,13 @@ int main(int argc, const char* argv[])
auto&& m = get_commands();
auto cmd = args.front();
if(cmd == "ort-sha")
{
std::cout << MIGRAPHX_ORT_SHA1 << std::endl;
return 0;
}
if(m.count(cmd) > 0)
{
m.at(cmd)(argv[0], {args.begin() + 1, args.end()});
......
......@@ -30,6 +30,7 @@
#include <migraphx/instruction.hpp>
#include <migraphx/compile_options.hpp>
#include <migraphx/quantization.hpp>
#include <migraphx/ranges.hpp>
namespace migraphx {
namespace driver {
......@@ -76,15 +77,25 @@ void verify_program(const std::string& name,
compile_options options,
precision quantize,
const parameter_map& inputs,
double tolerance)
verify::tolerance tols)
{
auto x = run_ref(p, inputs);
auto y = run_target(p, t, options, quantize, inputs);
auto ref_outs = run_ref(p, inputs);
auto target_outs = run_target(p, t, options, quantize, inputs);
std::size_t output_num = x.size();
std::size_t output_num = ref_outs.size();
for(std::size_t i = 0; i < output_num; ++i)
{
verify_args(name, x[i], y[i], tolerance);
if(ref_outs[i].get_shape().type() != target_outs[i].get_shape().type() or
ref_outs[i].get_shape().lens() != target_outs[i].get_shape().lens())
{
std::cout << "FAILED: " << name << std::endl;
std::cout << "Shape mismatch {" << ref_outs[i].get_shape() << "} != {"
<< target_outs[i].get_shape() << "}" << std::endl;
}
else
{
verify_args(name, target_outs[i], verify::expected{ref_outs[i]}, tols);
}
}
}
......@@ -92,7 +103,7 @@ void verify_instructions(const program& prog,
const target& t,
compile_options options,
precision quantize,
double tolerance)
verify::tolerance tols)
{
const auto* mm_prog = prog.get_main_module();
for(auto&& ins : (*mm_prog))
......@@ -123,8 +134,7 @@ void verify_instructions(const program& prog,
{
std::cout << "Verify: " << ins.name() << std::endl;
std::cout << p << std::endl;
verify_program(
ins.name(), p, t, options, quantize, create_param_map(p, false), tolerance);
verify_program(ins.name(), p, t, options, quantize, create_param_map(p, false), tols);
}
catch(...)
{
......@@ -140,14 +150,22 @@ void verify_reduced(program p,
compile_options options,
precision quantize,
const parameter_map& inputs,
double tolerance)
verify::tolerance tols)
{
auto* mm = p.get_main_module();
auto last = std::prev(mm->end(), n + 1);
auto last = std::prev(mm->end(), n);
mm->remove_instructions(last, mm->end());
std::cout << "Verify: " << n << std::endl;
std::cout << p << std::endl;
verify_program(std::to_string(n), p, t, options, quantize, inputs, tolerance);
try
{
verify_program(std::to_string(n), p, t, options, quantize, inputs, tols);
}
catch(const std::exception& e)
{
std::cout << "FAILED: " << n << std::endl;
std::cout << "Exception: " << e.what() << std::endl;
}
}
void verify_reduced_program(const program& p,
......@@ -155,14 +173,20 @@ void verify_reduced_program(const program& p,
compile_options options,
precision quantize,
const parameter_map& inputs,
double tolerance)
verify::tolerance tols)
{
const auto* mm = p.get_main_module();
auto n = std::distance(mm->begin(), mm->end());
std::cout << "Verify steps: " << n << std::endl;
for(std::size_t i = 0; i < n; i++)
for(std::size_t i = 1; i < n; i++)
{
verify_reduced(p, i, t, options, quantize, inputs, tolerance);
auto last = std::prev(mm->end(), i + 1);
if(contains({"@literal", "@param"}, last->name()))
{
std::cout << "Skip: " << i << std::endl;
continue;
}
verify_reduced(p, i, t, options, quantize, inputs, tols);
}
}
......
......@@ -26,6 +26,7 @@
#include "precision.hpp"
#include <migraphx/program.hpp>
#include <migraphx/verify.hpp>
namespace migraphx {
namespace driver {
......@@ -37,18 +38,18 @@ void verify_program(const std::string& name,
compile_options options = compile_options{},
precision quantize = precision::fp32,
const parameter_map& inputs = {},
double tolerance = 100);
verify::tolerance tols = verify::tolerance{});
void verify_instructions(const program& prog,
const target& t,
compile_options options = compile_options{},
precision quantize = precision::fp32,
double tolerance = 80);
verify::tolerance tols = verify::tolerance{});
void verify_reduced_program(const program& p,
const target& t,
compile_options options = compile_options{},
precision quantize = precision::fp32,
const parameter_map& inputs = {},
double tolerance = 80);
verify::tolerance tols = verify::tolerance{});
} // namespace MIGRAPHX_INLINE_NS
} // namespace driver
......
......@@ -24,11 +24,14 @@
#include <migraphx/fuse_pointwise.hpp>
#include <migraphx/pass_manager.hpp>
#include <migraphx/dead_code_elimination.hpp>
#include <migraphx/simplify_reshapes.hpp>
#include <migraphx/instruction.hpp>
#include <migraphx/program.hpp>
#include <migraphx/make_op.hpp>
#include <migraphx/iterator_for.hpp>
#include <migraphx/ranges.hpp>
#include <migraphx/matcher.hpp>
#include <migraphx/common_dims.hpp>
#include <iterator>
MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_DISABLE_POINTWISE_FUSION)
......@@ -189,6 +192,54 @@ static bool find_pointwise_modules(module& m)
}
return changed;
}
namespace {
struct find_pointwise_reshape_pointwise
{
auto matcher() const
{
auto reshape =
match::name("reshape", "squeeze", "unsqueeze", "flatten")(match::used_once());
auto skip_contiguous = [](auto... ms) {
return match::arg(0)(match::skip(match::name("contiguous")(match::used_once()))(ms...));
};
auto pointwise = match::name("pointwise")(match::used_once());
auto reshape_pointwise = reshape(skip_contiguous(pointwise.bind("x"))).bind("reshape");
return match::name("pointwise")(match::any_of[match::inputs()](reshape_pointwise));
}
void apply(module& m, const match::matcher_result& r) const
{
auto ins = r.result;
auto x_ins = r.instructions["x"];
auto reshape_ins = r.instructions["reshape"];
auto cd = common_dims::compute(ins->get_shape().lens(), x_ins->get_shape().lens());
if(cd.dims.empty())
return;
auto reshape_input = [&](const auto& ins_to_insert) {
return [&](auto input) {
auto c = m.insert_instruction(ins_to_insert, make_op("contiguous"), input);
return m.insert_instruction(
ins_to_insert, make_op("reshape", {{"dims", cd.dims}}), c);
};
};
auto x_inputs = x_ins->inputs();
std::transform(x_inputs.begin(), x_inputs.end(), x_inputs.begin(), reshape_input(x_ins));
auto new_x_ins =
m.insert_instruction(x_ins, x_ins->get_operator(), x_inputs, x_ins->module_inputs());
auto inputs = ins->inputs();
std::transform(inputs.begin(), inputs.end(), inputs.begin(), [&](auto input) {
if(input == reshape_ins)
return new_x_ins;
return reshape_input(ins)(input);
});
auto pw = m.insert_instruction(ins, ins->get_operator(), inputs, ins->module_inputs());
m.replace_instruction(ins, make_op("reshape", {{"dims", ins->get_shape().lens()}}), pw);
}
};
} // namespace
void fuse_pointwise::apply(module_pass_manager& mpm) const
{
......@@ -200,6 +251,8 @@ void fuse_pointwise::apply(module_pass_manager& mpm) const
}
for(int i = 0; i < 8; i++)
{
match::find_matches(mpm.get_module(), find_pointwise_reshape_pointwise{});
mpm.run_pass(simplify_reshapes{1});
if(not find_pointwise_modules(mpm.get_module()))
break;
mpm.run_pass(dead_code_elimination{});
......
......@@ -70,13 +70,19 @@ struct check_shapes
check_dynamic();
}
template <class Op>
template <class Op, MIGRAPHX_REQUIRES(not std::is_convertible<Op, std::string>{})>
check_shapes(const std::vector<shape>& s, const Op& op, const bool d = false)
: begin(s.begin()), end(s.end()), name(op.name()), dynamic_allowed(d)
{
check_dynamic();
}
check_shapes(const std::vector<shape>& s, const std::string& n, const bool d = false)
: begin(s.begin()), end(s.end()), name(n), dynamic_allowed(d)
{
check_dynamic();
}
void check_dynamic() const
{
if(not dynamic_allowed and this->any_of([&](const shape& s) { return s.dynamic(); }))
......@@ -147,7 +153,7 @@ struct check_shapes
{
if(begin != end)
{
if(begin->max_lens().size() != n)
if(begin->ndim() != n)
MIGRAPHX_THROW(prefix() + "Only " + std::to_string(n) + "d supported");
}
return *this;
......@@ -162,7 +168,7 @@ struct check_shapes
{
if(begin != end)
{
if(begin->max_lens().size() > n)
if(begin->ndim() > n)
MIGRAPHX_THROW(prefix() + "Shape must have at most " + std::to_string(n) +
" dimensions");
}
......@@ -178,7 +184,7 @@ struct check_shapes
{
if(begin != end)
{
if(begin->max_lens().size() < n)
if(begin->ndim() < n)
MIGRAPHX_THROW(prefix() + "Shape must have at least " + std::to_string(n) +
" dimensions");
}
......@@ -228,6 +234,16 @@ struct check_shapes
return *this;
}
/*!
* Check all shapes have the same layout.
*/
const check_shapes& same_layout() const
{
if(not this->same([](const shape& s) { return find_permutation(s); }))
MIGRAPHX_THROW(prefix() + "Layouts do not match");
return *this;
}
/*!
* Check all shapes are standard.
*/
......@@ -238,6 +254,16 @@ struct check_shapes
return *this;
}
/*!
* Check all shapes are scalar.
*/
const check_shapes& scalar() const
{
if(not this->all_of([](const shape& s) { return s.scalar(); }))
MIGRAPHX_THROW(prefix() + "Shapes are not a scalar");
return *this;
}
/*!
* Check all shapes are standard or scalar.
*/
......
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
*
* 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.
*/
#ifndef MIGRAPHX_GUARD_MIGRAPHX_COMMON_DIMS_HPP
#define MIGRAPHX_GUARD_MIGRAPHX_COMMON_DIMS_HPP
#include <migraphx/config.hpp>
#include <cstdint>
#include <vector>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
/// This will compute a higher dimensional space that will preserve the axes
/// for both sets of dimensions. Two axes_maps are provided for each of the
/// dims that will map the axis to the axes that are used by the result of
/// common_dims.
struct MIGRAPHX_EXPORT common_dims
{
static common_dims compute(const std::vector<std::size_t>& dims1,
const std::vector<std::size_t>& dims2);
std::vector<std::size_t> dims;
std::vector<std::vector<std::size_t>> axes_map1;
std::vector<std::vector<std::size_t>> axes_map2;
};
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif // MIGRAPHX_GUARD_MIGRAPHX_COMMON_DIMS_HPP
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
......@@ -62,7 +62,7 @@ void convolution(Output output, T input, T weights, Padding padding, Stride stri
shape win_shape{output_shape.type(), win_size};
double acc = 0.0;
shape_for_each(win_shape, [&](auto idx_win) {
shape_for_each(win_shape, [&](const auto& idx_win) {
auto k = idx_win[0];
const auto in_ch = group_id * wei_c + k;
std::vector<std::ptrdiff_t> idx(idx_o.begin(), idx_o.end());
......
......@@ -81,6 +81,7 @@ struct MIGRAPHX_EXPORT instruction
const std::vector<module_ref>& module_inputs() const;
/// Where this instruction is used as an input to another instruction
const std::vector<instruction_ref>& outputs() const;
friend bool operator==(const instruction& x, const instruction& y);
......
......@@ -381,22 +381,24 @@ void find_matches_for(source_location location, Mod& mod, instruction_ref ins, M
const int trace = value_of(MIGRAPHX_TRACE_MATCHES{});
const bool validate = enabled(MIGRAPHX_VALIDATE_MATCHES{});
const auto trace_filter = string_value_of(MIGRAPHX_TRACE_MATCHES_FOR{});
const bool trace_for = not trace_filter.empty() and
(contains(std::string{location.file_name()}, trace_filter) or
contains(std::string{location.function_name()}, trace_filter));
bool match = false;
bool match = false;
each_args(
[&](auto&& m) {
const auto& matcher_name = get_type_name(m);
const bool trace_for = not trace_filter.empty() and
(contains(std::string{location.file_name()}, trace_filter) or
contains(std::string{location.function_name()}, trace_filter) or
contains(matcher_name, trace_filter));
if(match)
return;
if(trace > 1 or trace_for)
std::cout << "Match: " << get_type_name(m) << std::endl;
if(trace > 1 and trace_for)
std::cout << "Match: " << matcher_name << std::endl;
auto r = match_instruction(get_module(mod), ins, m.matcher());
if(r.result == get_module(mod).end())
return;
if(trace > 0 or trace_for)
{
std::cout << "Matched by " << get_type_name(m) << std::endl;
std::cout << "Matched by " << matcher_name << std::endl;
get_module(mod).debug_print(ins);
}
// If its already invalid dont validate it again
......@@ -407,7 +409,7 @@ void find_matches_for(source_location location, Mod& mod, instruction_ref ins, M
auto invalid = get_module(mod).validate();
if(invalid != get_module(mod).end())
{
std::cout << "Invalid program from match: " << get_type_name(m) << std::endl;
std::cout << "Invalid program from match: " << matcher_name << std::endl;
std::cout << "Invalid instructions: " << std::endl;
get_module(mod).debug_print(invalid->inputs());
get_module(mod).debug_print(invalid);
......@@ -621,6 +623,8 @@ MIGRAPHX_PRED_MATCHER(broadcast, instruction_ref ins)
template <class... Ms>
auto skip(Ms... ms)
{
static_assert(((not std::is_convertible<Ms, std::string>{}) and ...),
"Use a matcher not a string for skip.");
auto m = any_of(ms...);
return make_basic_fun_matcher([=](matcher_context& ctx, instruction_ref start) {
return fix<optional<instruction_ref>>(
......
......@@ -36,20 +36,53 @@ namespace op {
struct allocate
{
shape s{};
// for dynamic allocate to set the buffer type
shape::type_t buf_type = shape::half_type;
template <class Self, class F>
static auto reflect(Self& self, F f)
{
return pack(f(self.s, "shape"));
return pack(f(self.s, "shape"), f(self.buf_type, "buf_type"));
}
std::string name() const { return "allocate"; }
shape compute_shape(const std::vector<shape>& inputs) const
{
migraphx::check_shapes{inputs, *this, true}.has(0);
return s;
if(s != shape())
{
if(inputs.size() == 1)
{
migraphx::check_shapes{inputs, *this, false}.only_dims(1);
}
else
{
migraphx::check_shapes{inputs, *this, false}.has(0);
}
return s;
}
else
{
migraphx::check_shapes{inputs, *this, false}.has(1).only_dims(1);
const auto& out_dims = inputs.at(0);
std::size_t max_val = std::numeric_limits<std::size_t>::max();
std::vector<shape::dynamic_dimension> dyn_dims(out_dims.lens().at(0),
shape::dynamic_dimension{0, max_val});
return {buf_type, dyn_dims};
}
}
argument compute(const shape& output_shape, const std::vector<argument>&) const
argument compute(const shape& output_shape, const std::vector<argument>& args) const
{
return {output_shape};
if(args.empty())
{
return {output_shape};
}
else
{
std::vector<std::size_t> output_dims(output_shape.ndim());
args.at(0).visit([&](auto a) { output_dims.assign(a.begin(), a.end()); });
return {shape{buf_type, output_dims}};
}
}
};
......
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
......@@ -33,8 +33,12 @@ namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace op {
// Specifies where to add the "extra" cell of padding if the
// calculated padding is an odd number.
// Padding mode is default_ for fixed shape padding.
// same_lower and same_upper used for dynamic padding.
// same_lower and same_upper specify dynamic padding.
// The odd cell goes at the beginning of the dimension
// (same_lower) or end (same_upper).
enum padding_mode_t
{
default_, // NOLINT
......
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
......
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
......@@ -206,6 +206,7 @@ struct convolution
std::vector<std::size_t> new_padding;
if(padding_mode != op::padding_mode_t::default_)
{
// auto-Calculate the padding sizes with calc_dyn_auto_pad
auto input_lens = args[0].get_shape().lens();
auto weights_lens = args[1].get_shape().lens();
new_padding =
......@@ -217,6 +218,7 @@ struct convolution
}
else
{
// Use the padding that was given
new_padding = padding;
if(output_shape.dynamic())
{
......
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
......@@ -164,7 +164,7 @@ struct convolution_backwards
shape win_shape{dyn_out.computed_shape.type(), win_size};
par_dfor(in_n, wei_c)([&](int o, int k) {
shape_for_each(win_shape, [&](auto idx_win) {
shape_for_each(win_shape, [&](const auto& idx_win) {
const int w = idx_win[0];
auto input_dims_start = idx_win.begin() + 1;
......
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* 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.
*/
#ifndef MIGRAPHX_GUARD_OPERATORS_FILL_HPP
#define MIGRAPHX_GUARD_OPERATORS_FILL_HPP
#include <migraphx/check_shapes.hpp>
#include <migraphx/dyn_output.hpp>
#include <migraphx/par_for.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace op {
/**
* fill(default_value, output_buffer)
* Fill an output buffer with the given default_value.
* Note that if the default_value is a literal and the output_buffer
* has a static shape this operator can be replaced with a literal.
*/
struct fill
{
std::string name() const { return "fill"; }
shape compute_shape(std::vector<shape> inputs) const
{
check_shapes{inputs, *this, true}.has(2).same_type();
if(inputs.at(0).dynamic() or inputs.at(0).elements() != 1)
{
MIGRAPHX_THROW("FILL: default_value is dynamic or more than one element");
}
return inputs.back();
}
argument compute(const dyn_output& dyn_out, std::vector<argument> args) const
{
visit_all(args[0], args[1])([&](auto value, auto output) {
par_for(dyn_out.computed_shape.elements(), [&](auto i) { output[i] = value.front(); });
});
return args[1];
}
std::ptrdiff_t output_alias(const std::vector<shape>&) const { return 1; }
};
} // namespace op
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
......@@ -125,13 +125,12 @@ struct gather
auto out_lens = data.get_shape().lens();
out_lens[axis] = indices.get_shape().elements();
migraphx::shape out_comp_shape{data.get_shape().type(), out_lens};
shape_for_each(out_comp_shape, [&](const auto& out_idx) {
auto data_idx = out_idx;
auto in_index = indices[data_idx[axis]];
in_index = (in_index < 0) ? in_index + axis_dim_size : in_index;
data_idx[axis] = in_index;
output[out_comp_shape.index(out_idx.begin(), out_idx.end())] =
data(data_idx.begin(), data_idx.end());
shape_for_each(out_comp_shape, [&](const auto& out_idx_v, size_t out_idx) {
auto data_idx = out_idx_v;
auto in_index = indices[data_idx[axis]];
in_index = (in_index < 0) ? in_index + axis_dim_size : in_index;
data_idx[axis] = in_index;
output[out_idx] = data(data_idx.begin(), data_idx.end());
});
}
});
......
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
......@@ -258,7 +258,7 @@ struct nonmaxsuppression
selected_boxes_inside_class.reserve(max_output_shape.elements());
// iterate over batches and classes
shape comp_s{shape::double_type, {num_batches, num_classes}};
shape_for_each(comp_s, [&](auto idx) {
shape_for_each(comp_s, [&](const auto& idx) {
auto batch_idx = idx[0];
auto class_idx = idx[1];
// index offset for this class
......
/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
* Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
......@@ -56,10 +56,10 @@ struct nonzero
std::vector<std::vector<std::size_t>> vec_idx;
auto s = args.front().get_shape();
args.front().visit([&](auto v) {
shape_for_each(s, [&](auto idx) {
if(not float_equal(v[s.index(idx)], 0))
shape_for_each(s, [&](const auto& idx_v, size_t idx) {
if(not float_equal(v[idx], 0))
{
vec_idx.push_back(idx);
vec_idx.push_back(idx_v);
}
});
});
......
......@@ -29,6 +29,7 @@
#include <migraphx/config.hpp>
#include <migraphx/value.hpp>
#include <migraphx/argument.hpp>
#include <migraphx/pad_calc.hpp>
#include <migraphx/par_for.hpp>
#include <migraphx/shape_for_each.hpp>
#include <migraphx/dyn_output.hpp>
......@@ -40,10 +41,20 @@ namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace op {
// The Pooling operator mostly follows the specifications for the Onnx pooling op.
// It assumes an NCHW layout, extended to support any number of spatial dimensions
// from 1 on up; dimensions are <batch index, channels, spatial dimensions...>
//
struct pooling
{
// Class members mode, ceil_mode, padding_mode have similar names but refer to separate
// concepts.
pooling_mode mode = {pooling_mode::average};
// If the input has rank other than 4 then padding, stride, lengths must all be specified
// since the defaults have 2-dimensions. Exception: padding not required if
// padding_mode != default_
// Padding along each spatial input dimension
// Can be ndim or 2*ndim values where ndim is size of lengths
// ndim values means pad the same before and after each dimension
......@@ -63,13 +74,14 @@ struct pooling
// ceiling mode is a flag affecting output size
// or equivalently, placements of the pooling kernel.
// When true, round the size upwards, possibly
// including partial placements where the kernel extends beyond the edge
// of input and even padding. When false, round down so that all
// When true, round the size upwards. When false, round down so that all
// kernel placements fit but some input values may be dropped.
bool ceil_mode = false;
int lp_order = 2;
// Mode for auto padding. default_ indicates no auto padding.
padding_mode_t padding_mode = padding_mode_t::default_;
// Global pooling with dynamic shape input
bool dyn_global = false;
......@@ -84,6 +96,7 @@ struct pooling
{
return pack(f(self.mode, "mode"),
f(self.padding, "padding"),
f(self.padding_mode, "padding_mode"),
f(self.stride, "stride"),
f(self.lengths, "lengths"),
f(self.ceil_mode, "ceil_mode"),
......@@ -97,7 +110,8 @@ struct pooling
{
if(dyn_global)
return;
if((padding.size() != stride.size() and (padding.size()) != stride.size() * 2) or
if((padding_mode != default_ and padding.size() != stride.size() and
(padding.size()) != stride.size() * 2) or
stride.size() != lengths.size())
{
MIGRAPHX_THROW("POOLING: inconsistent attribute sizes");
......@@ -137,8 +151,19 @@ struct pooling
std::size_t padding_factor = 2 * padding[i];
if(padding.size() == 2 * kdims)
padding_factor = padding[i] + padding[i + kdims];
assert(input_lens[i + 2] + padding_factor >= lengths[i]);
std::size_t dim_size = input_lens[i + 2] + padding_factor - lengths[i];
std::size_t dim_size;
if(input_lens[i + 2] + padding_factor < lengths[i])
{
if(padding_mode == default_)
MIGRAPHX_THROW("POOLING: not enough padding for the given kernel size");
// lengths can be legitimately larger only if we're doing auto padding
// with a dynamic shape, in which case given padding is ignored. Set a dummy value.
dim_size = 2;
}
else
{
dim_size = input_lens[i + 2] + padding_factor - lengths[i];
}
std::size_t len =
(ceil_mode)
? dim_size / stride[i] +
......@@ -151,17 +176,13 @@ struct pooling
shape normalize_compute_shape(std::vector<shape> inputs) const
{
check_shapes{inputs, *this, true}.has(1);
check_shapes{inputs, *this, true}.has(1).min_ndims(3);
check_attribute_size();
const shape& input = inputs.at(0);
auto padding_size = padding.size();
auto stride_size = stride.size();
size_t kdims = input.ndim() - 2;
if(input.ndim() < 3)
{
MIGRAPHX_THROW("POOLING: input must have 3 or more dimensions and be nonempty");
}
if(input.ndim() * 2 != padding_size + 4 and input.ndim() != padding_size + 2)
if(input.ndim() != stride_size + 2)
{
MIGRAPHX_THROW("POOLING: input and attribute size mismatch!");
}
......@@ -179,6 +200,28 @@ struct pooling
}
return {input.type(), output_dyn_dims};
}
else if(padding_mode != default_)
{
const size_t num_spatial_dims = inputs[0].ndim() - 2;
const shape& x_shape = inputs[0];
// same as convolution::dynamic_compute_shape()
for(std::size_t i = 0; i < num_spatial_dims; ++i)
{
auto ceil_div = [](std::size_t x, std::size_t y) { return (x + y - 1) / y; };
auto s = stride[i];
auto x = x_shape.dyn_dims()[i + 2];
std::set<std::size_t> optimals{};
std::transform(x.optimals.begin(),
x.optimals.end(),
std::inserter(optimals, optimals.begin()),
[&](auto o) { return ceil_div(o, s); });
output_dyn_dims.push_back(
shape::dynamic_dimension{ceil_div(x.min, s), ceil_div(x.max, s), optimals});
}
return {input.type(), output_dyn_dims};
}
else
{
// does not compute optimals
......@@ -267,6 +310,7 @@ struct pooling
Out& output,
const In& input,
const std::vector<std::size_t>& kernel_dims,
const std::vector<std::size_t>& padding_vals,
Op op) const
{
auto in_s = input.get_shape();
......@@ -283,9 +327,9 @@ struct pooling
// For each spatial dimension, find starting and ending index of pooling kernel
for(std::size_t dim = 2; dim < n_dim; ++dim)
{
auto d_2 = dim - 2;
int start =
static_cast<int>(idx_o[dim] * stride[d_2]) - static_cast<int>(padding[d_2]);
auto d_2 = dim - 2;
int start = static_cast<int>(idx_o[dim] * stride[d_2]) -
static_cast<int>(padding_vals[d_2]);
int end;
// NOLINT
if(count_include_pad and ceil_mode and (mode != pooling_mode::max))
......@@ -297,7 +341,7 @@ struct pooling
// Check if this kernel extends beyond the padding at end of dimension
end = std::min(start + kernel_dims[d_2],
in_lens[dim] + static_cast<int>(padding[d_2]));
in_lens[dim] + static_cast<int>(padding_vals[d_2]));
}
else
{
......@@ -316,11 +360,12 @@ struct pooling
}
shape win_shape{output_shape.type(), win_size};
auto pool_size = win_shape.elements();
double output_val = op.template init<Type>();
// for each element in the window...
shape_for_each(win_shape, [&](auto idx_w) {
shape_for_each(win_shape, [&](const auto& idx_w) {
// the coordinates of this element
auto idx = idx_o;
......@@ -354,30 +399,65 @@ struct pooling
argument compute(const dyn_output& dyn_out, std::vector<argument> args) const
{
argument result{dyn_out.computed_shape};
argument result;
auto input_lens = args[0].get_shape().lens();
std::vector<std::size_t> kernel_dims;
shape output_shape;
// If we have to auto-calculate padding, it will be passed to calc_pooling() as an argument
// instead of the member variable padding.
std::vector<std::size_t> temp_padding(padding);
if(dyn_global)
{
// for dynamic GlobalPooling, there's no padding
kernel_dims.insert(kernel_dims.end(), input_lens.begin() + 2, input_lens.end());
output_shape = dyn_out.computed_shape;
result = dyn_out.computed_shape;
}
else
else if((padding_mode != op::padding_mode_t::default_))
{
// if padding_mode is set, input was a dynamic size. Calculate padded size now.
// kernel_lens is the same as kernel_dims, but prepended with the 2 non-
// spatial dimensions. For size computations, it's used like the weights
// tensor for convolutions.
std::vector<std::size_t> kernel_lens;
kernel_lens.insert(kernel_lens.end(), input_lens.begin(), input_lens.begin() + 2);
kernel_lens.insert(kernel_lens.end(), lengths.begin(), lengths.end());
kernel_dims = this->lengths;
auto type = args[0].get_shape().type();
// dilation not currently supported for pooling, so default to all 1's
temp_padding = calc_dyn_auto_pad(
input_lens, kernel_lens, stride, {1, 1}, bool(padding_mode == op::same_upper));
output_shape = compute_padded_pool_shape(
args[0].get_shape(), shape(type, kernel_dims), temp_padding, stride, {1, 1});
result = argument(output_shape);
}
else // fixed/static input
{
kernel_dims = this->lengths;
output_shape = dyn_out.computed_shape;
result = dyn_out.computed_shape;
}
// Perform the computation and populate result
visit_all(result, args[0])([&](auto output, auto input) {
using type = typename decltype(output)::value_type;
switch(mode)
{
case migraphx::op::pooling_mode::average:
calc_pooling<type>(dyn_out.computed_shape, output, input, kernel_dims, avg_pool{});
calc_pooling<type>(
output_shape, output, input, kernel_dims, temp_padding, avg_pool{});
break;
case migraphx::op::pooling_mode::max:
calc_pooling<type>(dyn_out.computed_shape, output, input, kernel_dims, max_pool{});
calc_pooling<type>(
output_shape, output, input, kernel_dims, temp_padding, max_pool{});
break;
case migraphx::op::pooling_mode::lpnorm:
calc_pooling<type>(
dyn_out.computed_shape, output, input, kernel_dims, lpnorm_pool{lp_order});
output_shape, output, input, kernel_dims, temp_padding, lpnorm_pool{lp_order});
break;
}
});
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment