mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-20 06:49:15 +00:00
[CK_Tile] Support for a4w4 (fp4) in block scale gemm AB quant (#3603)
* chore: split block scale example instances in more separate files to speed up compile times * wip: fp4 scaffolding for abquant * feat: add fp4 decoding-while-loading to abquant pipeline * feat: add support for fp4 CPU verification in abquant * chore: add time tracking to reference calculation * feat: add a4w4 test for blockscale gemm * feat: optimize reference calculation by preconverting values to AccType * feat: add fp4 to fp8 look-up table * fix: reference to wrong ComputeDataType field in QuantProblem * feat: type utilities for determining MFMA compute types * feat: packed fp4 for abquant weight preshuffle * feat: add separate tests for a4w4 base case, padding and preshuffleB * fix: fp4 conversion on gfx950 attempting to use non-supported method * fix: test case was using quant group sizes which don't work on gfx950 due to larger mfma tile size * chore: add fp4 preshuffleb mode to block scale example * chore: sanity check for packed types being 1 byte * chore: clarify tensor dimension indices with constants * chore: replace traits check with specialized check for packed types * style: some minor refactoring and cleanup * fix: correct conversion table for FNUZ fp8 * chore: add fp4 instances to main abquant instances again * chore: use same initialization branch for int4 and fp4 * chore: add missing initialization for fp4 in block scale gemm example --------- Co-authored-by: Thomas Ning <Thomas.Ning@amd.com>
This commit is contained in:
@@ -91,6 +91,7 @@
|
||||
#include "ck_tile/core/utility/ignore.hpp"
|
||||
#include "ck_tile/core/utility/literals.hpp"
|
||||
#include "ck_tile/core/utility/magic_div.hpp"
|
||||
#include "ck_tile/core/utility/mixed_prec_compute_type.hpp"
|
||||
#include "ck_tile/core/utility/persistent_async_input_scheduler.hpp"
|
||||
#include "ck_tile/core/utility/philox_rand.hpp"
|
||||
#include "ck_tile/core/utility/print.hpp"
|
||||
|
||||
@@ -1544,7 +1544,8 @@ CK_TILE_DEVICE thread_buffer<T, N> amd_buffer_load_impl(int32x4_t src_wave_buffe
|
||||
(N == 1 || N == 2 || N == 4 || N == 8 || N == 16 || N == 32)) ||
|
||||
(std::is_same<T, pk_fp4_raw_t>::value &&
|
||||
(N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
|
||||
(std::is_same<T, pk_fp4_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)),
|
||||
(std::is_same<T, pk_fp4_t>::value &&
|
||||
(N == 1 || N == 2 || N == 4 || N == 8 || N == 16 || N == 32)),
|
||||
"wrong! not implemented");
|
||||
|
||||
using rtn_type = thread_buffer<T, N>;
|
||||
|
||||
@@ -1414,7 +1414,7 @@ CK_TILE_DEVICE thread_buffer<T, N> amd_buffer_load_impl(int32x4_t src_wave_buffe
|
||||
(std::is_same<T, pk_int4_t>::value &&
|
||||
(N == 1 || N == 2 || N == 4 || N == 8 || N == 16 || N == 32) ||
|
||||
(std::is_same<T, pk_fp4_t>::value &&
|
||||
(N == 1 || N == 2 || N == 4 || N == 8 || N == 16))),
|
||||
(N == 1 || N == 2 || N == 4 || N == 8 || N == 16 || N == 32))),
|
||||
"wrong! not implemented");
|
||||
|
||||
using rtn_type = thread_buffer<T, N>;
|
||||
|
||||
@@ -6,6 +6,7 @@
|
||||
#include <cmath>
|
||||
#include "ck_tile/core/config.hpp"
|
||||
#include "ck_tile/core/numeric/half.hpp"
|
||||
#include "ck_tile/core/numeric/float8.hpp"
|
||||
#include "ck_tile/core/numeric/mxfp_convert.hpp"
|
||||
|
||||
#if defined(__gfx950__)
|
||||
@@ -23,6 +24,12 @@ using fp32x2_t = float __attribute__((ext_vector_type(2)));
|
||||
using fp16x2_t = _Float16 __attribute__((ext_vector_type(2)));
|
||||
using bf16x2_t = bfloat16_t __attribute__((ext_vector_type(2)));
|
||||
|
||||
#if CK_TILE_USE_CUSTOM_DATA_TYPE
|
||||
using fp8x2_t = fp8_raw_t __attribute__((ext_vector_type(2)));
|
||||
#else
|
||||
using fp8x2_t = fp8_t __attribute__((ext_vector_type(2)));
|
||||
#endif
|
||||
|
||||
// Helpers: constexpr-safe access to elements of ext_vector_type(2)
|
||||
// Some compilers don't allow operator[] in constant expressions for vector types.
|
||||
// We use bit_cast to a trivially copyable representation to extract lanes.
|
||||
@@ -98,6 +105,8 @@ struct pk_float4_e2m1_t
|
||||
CK_TILE_HOST_DEVICE constexpr fp16x2_t to_fp16x2(float scale = 1.f) const;
|
||||
CK_TILE_HOST_DEVICE constexpr bf16_t to_bf16(float scale = 1.f) const;
|
||||
CK_TILE_HOST_DEVICE constexpr bf16x2_t to_bf16x2(float scale = 1.f) const;
|
||||
CK_TILE_HOST_DEVICE constexpr fp8_t to_fp8(float scale = 1.f) const;
|
||||
CK_TILE_HOST_DEVICE constexpr fp8x2_t to_fp8x2(float scale = 1.f) const;
|
||||
|
||||
CK_TILE_HOST_DEVICE constexpr operator float() const { return to_float(); }
|
||||
CK_TILE_HOST_DEVICE constexpr operator fp32x2_t() const { return to_fp32x2(); }
|
||||
@@ -105,6 +114,8 @@ struct pk_float4_e2m1_t
|
||||
CK_TILE_HOST_DEVICE constexpr operator fp16x2_t() const { return to_fp16x2(); }
|
||||
CK_TILE_HOST_DEVICE constexpr operator bf16_t() const { return to_bf16(); }
|
||||
CK_TILE_HOST_DEVICE constexpr operator bf16x2_t() const { return to_bf16x2(); }
|
||||
CK_TILE_HOST_DEVICE constexpr operator fp8_t() const { return to_fp8(); }
|
||||
CK_TILE_HOST_DEVICE constexpr operator fp8x2_t() const { return to_fp8x2(); }
|
||||
|
||||
template <index_t I>
|
||||
CK_TILE_HOST_DEVICE constexpr pk_float4_e2m1_t unpack(number<I>) const
|
||||
@@ -145,6 +156,49 @@ struct pk_float4_e2m1_t
|
||||
bit_cast<fp16_t>(static_cast<uint16_t>(0xC400)), // -4
|
||||
bit_cast<fp16_t>(static_cast<uint16_t>(0xC600)) // -6
|
||||
};
|
||||
|
||||
#if CK_TILE_USE_OCP_FP8
|
||||
// FP8 EM4E3 (OCP) representation
|
||||
static constexpr fp8_t e2m1_to_fp8_table[16] = {
|
||||
fp8_t(static_cast<uint8_t>(0x00)), // 0
|
||||
fp8_t(static_cast<uint8_t>(0x30)), // 0.5
|
||||
fp8_t(static_cast<uint8_t>(0x38)), // 1
|
||||
fp8_t(static_cast<uint8_t>(0x3C)), // 1.5
|
||||
fp8_t(static_cast<uint8_t>(0x40)), // 2
|
||||
fp8_t(static_cast<uint8_t>(0x44)), // 3
|
||||
fp8_t(static_cast<uint8_t>(0x48)), // 4
|
||||
fp8_t(static_cast<uint8_t>(0x4C)), // 6
|
||||
fp8_t(static_cast<uint8_t>(0x00)), // -0
|
||||
fp8_t(static_cast<uint8_t>(0xB0)), // -0.5
|
||||
fp8_t(static_cast<uint8_t>(0xB8)), // -1
|
||||
fp8_t(static_cast<uint8_t>(0xBC)), // -1.5
|
||||
fp8_t(static_cast<uint8_t>(0xC0)), // -2
|
||||
fp8_t(static_cast<uint8_t>(0xC4)), // -3
|
||||
fp8_t(static_cast<uint8_t>(0xC8)), // -4
|
||||
fp8_t(static_cast<uint8_t>(0xCC)) // -6
|
||||
};
|
||||
#else // CK_TILE_USE_FNUZ_FP8
|
||||
// FP8 E4M3 FNUZ
|
||||
static constexpr fp8_t e2m1_to_fp8_table[16] = {
|
||||
fp8_t(static_cast<uint8_t>(0x00)), // 0
|
||||
fp8_t(static_cast<uint8_t>(0x38)), // 0.5
|
||||
fp8_t(static_cast<uint8_t>(0x40)), // 1
|
||||
fp8_t(static_cast<uint8_t>(0x44)), // 1.5
|
||||
fp8_t(static_cast<uint8_t>(0x48)), // 2
|
||||
fp8_t(static_cast<uint8_t>(0x4C)), // 3
|
||||
fp8_t(static_cast<uint8_t>(0x50)), // 4
|
||||
fp8_t(static_cast<uint8_t>(0x54)), // 6
|
||||
fp8_t(static_cast<uint8_t>(0x00)), // -0
|
||||
fp8_t(static_cast<uint8_t>(0xB8)), // -0.5
|
||||
fp8_t(static_cast<uint8_t>(0xC0)), // -1
|
||||
fp8_t(static_cast<uint8_t>(0xC4)), // -1.5
|
||||
fp8_t(static_cast<uint8_t>(0xC4)), // -2
|
||||
fp8_t(static_cast<uint8_t>(0xCC)), // -3
|
||||
fp8_t(static_cast<uint8_t>(0xD0)), // -4
|
||||
fp8_t(static_cast<uint8_t>(0xD4)) // -6
|
||||
};
|
||||
#endif
|
||||
|
||||
#endif
|
||||
};
|
||||
|
||||
@@ -408,6 +462,27 @@ CK_TILE_HOST_DEVICE constexpr fp16x2_t pk_fp4_t::to_fp16x2(float scale) const
|
||||
type_convert<fp16_t>(convert_to_float<pk_fp4_t>(_unpack(number<1>{}), scale))};
|
||||
#endif
|
||||
}
|
||||
CK_TILE_HOST_DEVICE constexpr fp8_t pk_fp4_t::to_fp8(float scale) const
|
||||
{
|
||||
// NOTE: No specialized fp4 to fp8 instructions are available. Unsure whether fp4 to fp16 to fp8
|
||||
// would be better than the naive implementation below
|
||||
// #if CK_TILE_FP4_CVT_DEVICE
|
||||
// return impl::_from_f4<fp8_t>(data, scale);
|
||||
// #else
|
||||
return fp8_t{type_convert<fp8_t>(convert_to_float<pk_fp4_t>(_unpack(number<0>{}), scale))};
|
||||
// #endif
|
||||
}
|
||||
CK_TILE_HOST_DEVICE constexpr fp8x2_t pk_fp4_t::to_fp8x2(float scale) const
|
||||
{
|
||||
// NOTE: No specialized fp4 to fp8 instructions are available. Unsure whether fp4 to fp16 to fp8
|
||||
// would be better than the naive implementation below
|
||||
// #if CK_TILE_FP4_CVT_DEVICE
|
||||
// return impl::_from_f4<fp8x2_t>(data, scale);
|
||||
// #else
|
||||
return fp8x2_t{type_convert<fp8_t>(convert_to_float<pk_fp4_t>(_unpack(number<0>{}), scale)),
|
||||
type_convert<fp8_t>(convert_to_float<pk_fp4_t>(_unpack(number<1>{}), scale))};
|
||||
// #endif
|
||||
}
|
||||
#else
|
||||
CK_TILE_HOST_DEVICE constexpr float pk_fp4_t::to_float(float scale) const
|
||||
{
|
||||
@@ -415,7 +490,8 @@ CK_TILE_HOST_DEVICE constexpr float pk_fp4_t::to_float(float scale) const
|
||||
}
|
||||
CK_TILE_HOST_DEVICE constexpr fp32x2_t pk_fp4_t::to_fp32x2(float scale) const
|
||||
{
|
||||
return fp32x2_t{e2m1_to_fp32_table[_unpack(number<0>{})] * scale, e2m1_to_fp32_table[_unpack(number<1>{}] * scale};
|
||||
return fp32x2_t{e2m1_to_fp32_table[_unpack(number<0>{})] * scale,
|
||||
e2m1_to_fp32_table[_unpack(number<1>{})] * scale};
|
||||
}
|
||||
CK_TILE_HOST_DEVICE constexpr fp16_t pk_fp4_t::to_fp16(float scale) const
|
||||
{
|
||||
@@ -428,6 +504,16 @@ CK_TILE_HOST_DEVICE constexpr fp16x2_t pk_fp4_t::to_fp16x2(float scale) const
|
||||
type_convert<fp16_t>(type_convert<float>(e2m1_to_fp16_table[_unpack(number<1>{})]) *
|
||||
scale)};
|
||||
}
|
||||
CK_TILE_HOST_DEVICE constexpr fp8_t pk_fp4_t::to_fp8(float scale) const
|
||||
{
|
||||
return type_convert<float>(e2m1_to_fp8_table[_unpack(number<0>{})]) * scale;
|
||||
}
|
||||
CK_TILE_HOST_DEVICE constexpr fp8x2_t pk_fp4_t::to_fp8x2(float scale) const
|
||||
{
|
||||
return fp8x2_t{
|
||||
type_convert<fp8_t>(type_convert<float>(e2m1_to_fp8_table[_unpack(number<0>{})]) * scale),
|
||||
type_convert<fp8_t>(type_convert<float>(e2m1_to_fp8_table[_unpack(number<1>{})]) * scale)};
|
||||
}
|
||||
#endif
|
||||
|
||||
} // namespace ck_tile
|
||||
|
||||
@@ -6,6 +6,7 @@
|
||||
#include "ck_tile/core/numeric/integral_constant.hpp"
|
||||
#include "ck_tile/core/numeric/math.hpp"
|
||||
#include "ck_tile/core/numeric/numeric.hpp"
|
||||
#include "ck_tile/core/numeric/pk_fp4.hpp"
|
||||
#include "ck_tile/core/utility/bit_cast.hpp"
|
||||
#include "ck_tile/core/utility/random.hpp"
|
||||
#include <stdint.h>
|
||||
@@ -23,6 +24,11 @@ struct pk_int4_t
|
||||
type data;
|
||||
CK_TILE_HOST_DEVICE constexpr pk_int4_t() : data{type{}} {}
|
||||
CK_TILE_HOST_DEVICE constexpr pk_int4_t(type init) : data{init} {}
|
||||
|
||||
// NOTE: added for interface compatibility with pk_fp4_t
|
||||
// Other data types could be added for greater similarity
|
||||
CK_TILE_HOST_DEVICE constexpr fp32x2_t to_fp32x2() const;
|
||||
CK_TILE_HOST_DEVICE constexpr operator fp32x2_t() const { return to_fp32x2(); }
|
||||
};
|
||||
|
||||
// limits
|
||||
@@ -186,4 +192,9 @@ CK_TILE_HOST_DEVICE int8x2_t pk_int4_t_to_int8x2_t(const pk_int4_t& x)
|
||||
return res;
|
||||
}
|
||||
|
||||
CK_TILE_HOST_DEVICE constexpr fp32x2_t pk_int4_t::to_fp32x2() const
|
||||
{
|
||||
return pk_int4_t_to_fp32x2_t(*this);
|
||||
}
|
||||
|
||||
} // namespace ck_tile
|
||||
|
||||
@@ -11,6 +11,7 @@
|
||||
#include "ck_tile/core/numeric/half.hpp"
|
||||
#include "ck_tile/core/numeric/bfloat16.hpp"
|
||||
#include "ck_tile/core/numeric/pk_int4.hpp"
|
||||
#include "ck_tile/core/numeric/pk_fp4.hpp"
|
||||
#include "ck_tile/core/numeric/e8m0.hpp"
|
||||
#include "ck_tile/core/utility/type_traits.hpp"
|
||||
|
||||
|
||||
54
include/ck_tile/core/utility/mixed_prec_compute_type.hpp
Normal file
54
include/ck_tile/core/utility/mixed_prec_compute_type.hpp
Normal file
@@ -0,0 +1,54 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core/config.hpp"
|
||||
#include "ck_tile/core/numeric/numeric.hpp"
|
||||
#include "ck_tile/core/utility/type_traits.hpp"
|
||||
|
||||
#include <type_traits>
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
namespace detail {
|
||||
|
||||
// Helper method to automatically determine compute type
|
||||
// Selects the largest type of the two. If both of them are packed data types, defaults to fp8.
|
||||
template <typename ADataType, typename BDataType>
|
||||
struct auto_compute_type
|
||||
{
|
||||
using LargestInputType = largest_type_t<ADataType, BDataType>;
|
||||
|
||||
// Sanity check: there are no packed types larger than 1 byte yet, but if we add them
|
||||
// this logic should change
|
||||
static_assert(!is_packed_type_v<LargestInputType> || sizeof(LargestInputType) == sizeof(fp8_t));
|
||||
|
||||
using type = std::conditional_t<is_packed_type_v<LargestInputType>, fp8_t, LargestInputType>;
|
||||
};
|
||||
|
||||
// Helper method to determine compute type, defaulting an explicitly passed-in compute type
|
||||
template <typename ComputeDataType, typename ADataType, typename BDataType>
|
||||
struct mixed_prec_compute_type
|
||||
{
|
||||
using type = std::conditional_t<std::is_void_v<ComputeDataType>,
|
||||
typename auto_compute_type<ADataType, BDataType>::type,
|
||||
ComputeDataType>;
|
||||
};
|
||||
|
||||
} // namespace detail
|
||||
|
||||
template <typename ComputeDataType, typename ADataType, typename BDataType>
|
||||
using mixed_prec_compute_type_t =
|
||||
typename detail::mixed_prec_compute_type<ComputeDataType, ADataType, BDataType>::type;
|
||||
|
||||
// Helper method to determine compute type, defaulting to input data type
|
||||
// If "ThisDataType" is packed (4-bit), will default to "OtherDataType". If both are packed,
|
||||
// ComputeDataType is used.
|
||||
template <typename ThisDataType, typename OtherDataType, typename ComputeDataType>
|
||||
using mixed_prec_compute_type_from_input_t = std::conditional_t<
|
||||
is_packed_type_v<ThisDataType>,
|
||||
std::conditional_t<is_packed_type_v<OtherDataType>, ComputeDataType, OtherDataType>,
|
||||
ThisDataType>;
|
||||
|
||||
} // namespace ck_tile
|
||||
@@ -4,6 +4,8 @@
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core/config.hpp"
|
||||
#include "ck_tile/core/numeric/numeric.hpp"
|
||||
|
||||
#include <tuple>
|
||||
#include <type_traits>
|
||||
#include <stdint.h>
|
||||
@@ -187,4 +189,19 @@ template <typename Tuple_, std::size_t Idx, typename DefaultType>
|
||||
using tuple_element_or_default_t =
|
||||
typename tuple_element_or_default<Tuple_, Idx, DefaultType>::type;
|
||||
|
||||
// Helper struct to determine if a type is packed (more than 1 element per byte)
|
||||
template <typename T>
|
||||
struct is_packed_type
|
||||
{
|
||||
static constexpr bool value = numeric_traits<T>::PackedSize > 1;
|
||||
};
|
||||
|
||||
template <typename T>
|
||||
static constexpr bool is_packed_type_v = is_packed_type<T>::value;
|
||||
|
||||
// Helper definition to take the largest sizes type
|
||||
template <typename ADataType, typename BDataType>
|
||||
using largest_type_t =
|
||||
std::conditional_t<sizeof(ADataType) >= sizeof(BDataType), ADataType, BDataType>;
|
||||
|
||||
} // namespace ck_tile
|
||||
|
||||
@@ -137,47 +137,55 @@ CK_TILE_HOST void reference_gemm_abquant(const HostTensor<ADataType>& a_m_k,
|
||||
const BElementOp& b_element_op = {},
|
||||
const ACCElementOp& acc_element_op = {})
|
||||
{
|
||||
const std::size_t M = a_m_k.get_length(0);
|
||||
const std::size_t N = b_k_n.get_length(1);
|
||||
const std::size_t K = a_m_k.get_length(1);
|
||||
constexpr auto A_TENSOR_M_DIM = 0;
|
||||
constexpr auto A_TENSOR_K_DIM = 1;
|
||||
constexpr auto B_TENSOR_K_DIM = 0;
|
||||
constexpr auto B_TENSOR_N_DIM = 1;
|
||||
|
||||
const std::size_t M = a_m_k.get_length(A_TENSOR_M_DIM);
|
||||
const std::size_t N = b_k_n.get_length(B_TENSOR_N_DIM);
|
||||
const std::size_t K = a_m_k.get_length(A_TENSOR_K_DIM);
|
||||
|
||||
// Pre-convert A/B tensors to AccData type
|
||||
// This prevents doing slow reconversions for each row/column
|
||||
HostTensor<AccDataType> a_acc(a_m_k.mDesc);
|
||||
HostTensor<AccDataType> b_acc(b_k_n.mDesc);
|
||||
|
||||
a_acc.ForEach([&](auto& self, auto index) {
|
||||
if constexpr(std::is_same_v<ADataType, pk_int4_t> || std::is_same_v<ADataType, pk_fp4_t>)
|
||||
{
|
||||
const ADataType pk_val = a_element_op(a_m_k(index));
|
||||
const fp32x2_t fp32_val = pk_val.to_fp32x2();
|
||||
self(index) = (index[A_TENSOR_K_DIM] & 1) ? fp32_val.hi : fp32_val.lo;
|
||||
}
|
||||
else
|
||||
{
|
||||
self(index) = ck_tile::type_convert<AccDataType>(a_element_op(a_m_k(index)));
|
||||
}
|
||||
});
|
||||
|
||||
b_acc.ForEach([&](auto& self, auto index) {
|
||||
if constexpr(std::is_same_v<BDataType, pk_int4_t> || std::is_same_v<BDataType, pk_fp4_t>)
|
||||
{
|
||||
const BDataType pk_val = b_element_op(b_k_n(index));
|
||||
const fp32x2_t fp32_val = pk_val.to_fp32x2();
|
||||
self(index) = (index[B_TENSOR_K_DIM] & 1) ? fp32_val.hi : fp32_val.lo;
|
||||
}
|
||||
else if constexpr(std::is_same_v<BDataType, fp8_t>)
|
||||
{
|
||||
self(index) = fp8_to_float_raw(b_element_op(b_k_n(index)));
|
||||
}
|
||||
else
|
||||
{
|
||||
self(index) = ck_tile::type_convert<AccDataType>(b_element_op(b_k_n(index)));
|
||||
}
|
||||
});
|
||||
|
||||
auto f_mn = [&](auto m, auto n) {
|
||||
AccDataType v_acc = 0;
|
||||
|
||||
constexpr std::size_t kGroupK = BQuantGroupSize::kK;
|
||||
|
||||
// ---- A loader: dequant A(m,k) into AccDataType ----
|
||||
auto load_a = [&](std::size_t k) -> AccDataType {
|
||||
if constexpr(std::is_same_v<ADataType, pk_int4_t>)
|
||||
{
|
||||
const pk_int4_t pk_val = a_element_op(a_m_k(m, k));
|
||||
const fp32x2_t fp32_val = pk_int4_t_to_fp32x2_t(pk_val);
|
||||
return (k & 1) ? fp32_val.hi : fp32_val.lo;
|
||||
}
|
||||
else
|
||||
{
|
||||
return ck_tile::type_convert<AccDataType>(a_element_op(a_m_k(m, k)));
|
||||
}
|
||||
};
|
||||
|
||||
// ---- B loader: dequant B(k,n) into AccDataType ----
|
||||
auto load_b = [&](std::size_t k) -> AccDataType {
|
||||
if constexpr(std::is_same_v<BDataType, pk_int4_t>)
|
||||
{
|
||||
const pk_int4_t pk_val = b_element_op(b_k_n(k, n));
|
||||
const fp32x2_t fp32_val = pk_int4_t_to_fp32x2_t(pk_val);
|
||||
return (k & 1) ? fp32_val.hi : fp32_val.lo;
|
||||
}
|
||||
else if constexpr(std::is_same_v<BDataType, fp8_t>)
|
||||
{
|
||||
return fp8_to_float_raw(b_element_op(b_k_n(k, n)));
|
||||
}
|
||||
else
|
||||
{
|
||||
return ck_tile::type_convert<AccDataType>(b_element_op(b_k_n(k, n)));
|
||||
}
|
||||
};
|
||||
|
||||
// ---- a scale loader for a given K-group index ----
|
||||
auto load_scale_a = [&](ck_tile::index_t k_group) -> float {
|
||||
const ck_tile::index_t outer_dim = m / AQuantGroupSize::kM;
|
||||
@@ -224,8 +232,8 @@ CK_TILE_HOST void reference_gemm_abquant(const HostTensor<ADataType>& a_m_k,
|
||||
// unscaled accumulation within this K-group
|
||||
for(std::size_t k = k_begin; k < k_end; ++k)
|
||||
{
|
||||
const AccDataType v_a = load_a(k);
|
||||
const AccDataType v_b = load_b(k);
|
||||
const AccDataType v_a = a_acc(m, k);
|
||||
const AccDataType v_b = b_acc(k, n);
|
||||
v_block_acc += v_a * v_b;
|
||||
}
|
||||
|
||||
|
||||
@@ -4,11 +4,12 @@
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core/config.hpp"
|
||||
#include "ck_tile/core/utility/type_traits.hpp"
|
||||
#include "ck_tile/ops/elementwise/unary_element_wise_operation.hpp"
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
template <typename DstDataType, index_t UnaryOpSize>
|
||||
template <typename SrcDataType, typename DstDataType, index_t UnaryOpSize>
|
||||
struct InterleavedPKTypeLoader
|
||||
{
|
||||
template <typename WarpWindow, typename WarpTile>
|
||||
@@ -21,10 +22,15 @@ struct InterleavedPKTypeLoader
|
||||
constexpr index_t thread_buffer_size = WarpTile::get_thread_buffer_size() / UnaryOpSize;
|
||||
const auto in_dstr_tensors = load_tile(warp_window);
|
||||
|
||||
using DstVectorType = DstDataType __attribute__((ext_vector_type(UnaryOpSize)));
|
||||
// NOTE: we rely on types packing neatly here
|
||||
using RawSrcType = typename SrcDataType::type;
|
||||
constexpr auto PackedSize = numeric_traits<SrcDataType>::PackedSize;
|
||||
|
||||
using SrcVectorType = ext_vector_t<RawSrcType, UnaryOpSize / PackedSize>;
|
||||
using DstVectorType = ext_vector_t<DstDataType, UnaryOpSize>;
|
||||
static_for<0, thread_buffer_size, 1>{}([&](auto i) {
|
||||
elementwise_op(warp_tile.get_thread_buffer().template get_as<DstVectorType>()(i),
|
||||
in_dstr_tensors.get_thread_buffer().template get_as<pk_int4x4_t>()[i]);
|
||||
in_dstr_tensors.get_thread_buffer().template get_as<SrcVectorType>()[i]);
|
||||
});
|
||||
}
|
||||
};
|
||||
@@ -37,10 +43,11 @@ template <typename SrcDataType,
|
||||
typename WarpWindow>
|
||||
CK_TILE_DEVICE void load_int4_tile(WarpTile& dst, const WarpWindow& src)
|
||||
{
|
||||
if constexpr(std::is_same_v<SrcDataType, pk_int4_t>)
|
||||
if constexpr(is_packed_type_v<SrcDataType>)
|
||||
{
|
||||
static_assert(!LoadTranspose, "LoadTranspose not supported with pk_int4_t");
|
||||
InterleavedPKTypeLoader<DstDataType, UnaryOpSize>::load_interleaved_pk_type(dst, src);
|
||||
static_assert(!LoadTranspose, "LoadTranspose not supported with pk_int4_t or pk_fp4_t");
|
||||
InterleavedPKTypeLoader<SrcDataType, DstDataType, UnaryOpSize>::load_interleaved_pk_type(
|
||||
dst, src);
|
||||
}
|
||||
else if constexpr(LoadTranspose)
|
||||
{
|
||||
|
||||
@@ -397,6 +397,29 @@ struct PassThroughPack8
|
||||
y.hi = i4_to_bf8x4(bit_cast<int>(x) >> 8);
|
||||
#endif
|
||||
}
|
||||
|
||||
CK_TILE_HOST_DEVICE constexpr void operator()(fp8x8_t& y, const pk_fp4x4_t& x) const
|
||||
{
|
||||
pk_fp4_t f0 = pk_fp4_t{x[0]};
|
||||
pk_fp4_t f1 = pk_fp4_t{x[1]};
|
||||
pk_fp4_t f2 = pk_fp4_t{x[2]};
|
||||
pk_fp4_t f3 = pk_fp4_t{x[3]};
|
||||
|
||||
fp8x2_t x0 = f0.to_fp8x2();
|
||||
fp8x2_t x1 = f1.to_fp8x2();
|
||||
fp8x2_t x2 = f2.to_fp8x2();
|
||||
fp8x2_t x3 = f3.to_fp8x2();
|
||||
|
||||
y[0] = x0[0];
|
||||
y[1] = x0[1];
|
||||
y[2] = x1[0];
|
||||
y[3] = x1[1];
|
||||
y[4] = x2[0];
|
||||
y[5] = x2[1];
|
||||
y[6] = x3[0];
|
||||
y[7] = x3[1];
|
||||
}
|
||||
|
||||
constexpr const static bool is_pack8_invocable = true;
|
||||
};
|
||||
|
||||
|
||||
@@ -4,6 +4,7 @@
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core.hpp"
|
||||
#include "ck_tile/core/numeric/numeric.hpp"
|
||||
#include "ck_tile/ops/gemm/block/block_wp_asmem_breg_creg.hpp"
|
||||
#include "ck_tile/ops/gemm/warp/warp_gemm_dispatcher.hpp"
|
||||
|
||||
@@ -255,17 +256,26 @@ struct UniversalWeightPreshufflePipelineAgBgCrPolicy
|
||||
{
|
||||
using BlockWarps = typename Problem::BlockGemmShape::BlockWarps;
|
||||
using WarpTile = typename Problem::BlockGemmShape::WarpTile;
|
||||
using BTypeToUse =
|
||||
std::conditional_t<std::is_same_v<typename Problem::BDataType, ck_tile::pk_int4_t>,
|
||||
typename Problem::ADataType,
|
||||
typename Problem::BDataType>;
|
||||
|
||||
// Determine compute types to use
|
||||
// This logic defaults to A/B DataType, but if one of them is packed falls back to the other
|
||||
// If both are packed, it falls back to the explicitly defined ComputeDataType in the
|
||||
// problem It might be a good idea to use ComputeDataType anyway, but that would break how
|
||||
// this behaviour used to work
|
||||
using ATypeToUse = mixed_prec_compute_type_from_input_t<typename Problem::ADataType,
|
||||
typename Problem::BDataType,
|
||||
typename Problem::ComputeDataType>;
|
||||
using BTypeToUse = mixed_prec_compute_type_from_input_t<typename Problem::BDataType,
|
||||
typename Problem::ADataType,
|
||||
typename Problem::ComputeDataType>;
|
||||
|
||||
constexpr index_t WaveSize = get_warp_size();
|
||||
constexpr index_t KLane = WarpTile::at(I2) * WarpTile::at(I0) / WaveSize;
|
||||
using BDataType = typename Problem::BDataType;
|
||||
constexpr index_t KLaneBytes =
|
||||
KLane / numeric_traits<BDataType>::PackedSize * sizeof(BDataType);
|
||||
constexpr auto NumAccess = static_cast<WGAttrNumAccessEnum>(max(1, KLaneBytes / 16));
|
||||
using WarpGemm = WarpGemmDispatcher<typename Problem::ADataType,
|
||||
using WarpGemm = WarpGemmDispatcher<ATypeToUse,
|
||||
BTypeToUse,
|
||||
typename Problem::CDataType,
|
||||
WarpTile::at(I0),
|
||||
|
||||
@@ -101,9 +101,11 @@ struct BlockGemmWeightPreshuffleABQuantARegBRegCReg
|
||||
// 4. i4, bf8, (fp8/fp32) -> f32
|
||||
static_assert(
|
||||
(std::is_same_v<ADataType, fp8_t> || std::is_same_v<ADataType, bf8_t> ||
|
||||
std::is_same_v<ADataType, ck_tile::pk_int4_t>) &&
|
||||
std::is_same_v<ADataType, ck_tile::pk_int4_t> ||
|
||||
std::is_same_v<ADataType, ck_tile::pk_fp4_t>) &&
|
||||
(std::is_same_v<BDataType, fp8_t> || std::is_same_v<BDataType, bf8_t> ||
|
||||
std::is_same_v<BDataType, ck_tile::pk_int4_t>) &&
|
||||
std::is_same_v<BDataType, ck_tile::pk_int4_t> ||
|
||||
std::is_same_v<BDataType, ck_tile::pk_fp4_t>) &&
|
||||
(std::is_same_v<AQDataType, float> || std::is_same_v<AQDataType, ck_tile::fp8_t> ||
|
||||
std::is_same_v<AQDataType, ck_tile::bf8_t>) &&
|
||||
(std::is_same_v<BQDataType, float> || std::is_same_v<BQDataType, ck_tile::fp8_t> ||
|
||||
@@ -189,7 +191,8 @@ struct BlockGemmWeightPreshuffleABQuantARegBRegCReg
|
||||
typename BFlatBlockTensor,
|
||||
typename AQBlockTensor,
|
||||
typename BQBlockTensor,
|
||||
typename ABlockWindow>
|
||||
typename ABlockWindow,
|
||||
index_t UnaryOpSize = 8>
|
||||
CK_TILE_DEVICE void operator()(CBlockTensor& c_block_tensor,
|
||||
ABlockTensor& a_warp_tensor,
|
||||
BFlatBlockTensor& b_warp_tensor,
|
||||
@@ -249,8 +252,10 @@ struct BlockGemmWeightPreshuffleABQuantARegBRegCReg
|
||||
{
|
||||
constexpr auto AmIter = (mIter + m_preload) % MIterPerWarp;
|
||||
constexpr auto AkIter = (kIter + (mIter + m_preload) / MIterPerWarp);
|
||||
a_warp_tensor(number<AwarpIter>{}) =
|
||||
load_tile(a_warp_windows(number<AmIter>{})(number<AkIter>{}));
|
||||
|
||||
load_int4_tile<ADataType, ComputeDataType, UnaryOpSize>(
|
||||
a_warp_tensor(number<AwarpIter>{}),
|
||||
a_warp_windows(number<AmIter>{})(number<AkIter>{}));
|
||||
}
|
||||
// barrier
|
||||
// Could be deleted
|
||||
|
||||
@@ -108,9 +108,11 @@ struct ABQuantBlockUniversalGemmAsBsCr : public BlockGemmQuantBase
|
||||
// 4. i4, bf8, (fp8/fp32) -> f32
|
||||
static_assert(
|
||||
(std::is_same_v<ADataType, fp8_t> || std::is_same_v<ADataType, bf8_t> ||
|
||||
std::is_same_v<ADataType, ck_tile::pk_int4_t>) &&
|
||||
std::is_same_v<ADataType, ck_tile::pk_int4_t> ||
|
||||
std::is_same_v<ADataType, ck_tile::pk_fp4_t>) &&
|
||||
(std::is_same_v<BDataType, fp8_t> || std::is_same_v<BDataType, bf8_t> ||
|
||||
std::is_same_v<BDataType, ck_tile::pk_int4_t>) &&
|
||||
std::is_same_v<BDataType, ck_tile::pk_int4_t> ||
|
||||
std::is_same_v<BDataType, ck_tile::pk_fp4_t>) &&
|
||||
(std::is_same_v<AQDataType, float> || std::is_same_v<AQDataType, ck_tile::fp8_t> ||
|
||||
std::is_same_v<AQDataType, ck_tile::bf8_t>) &&
|
||||
(std::is_same_v<BQDataType, float> || std::is_same_v<BQDataType, ck_tile::fp8_t> ||
|
||||
@@ -135,12 +137,9 @@ struct ABQuantBlockUniversalGemmAsBsCr : public BlockGemmQuantBase
|
||||
using ComputeDataType = remove_cvref_t<typename Traits::ComputeDataType>;
|
||||
using CDataType = remove_cvref_t<typename Traits::CDataType>;
|
||||
|
||||
// BDataType gets converted from PkInt4 during loading
|
||||
using OverrideBDataType = std::conditional_t<
|
||||
std::is_same_v<BDataType, pk_int4_t> &&
|
||||
std::is_same_v<typename Traits::BLayout, tensor_layout::gemm::RowMajor>,
|
||||
ADataType,
|
||||
BDataType>;
|
||||
// A/B DataType get converted from PkInt4/PkFp4 during loading
|
||||
using OverrideADataType = ComputeDataType;
|
||||
using OverrideBDataType = ComputeDataType;
|
||||
|
||||
using Base = BlockGemmQuantBase;
|
||||
using WarpGemm = remove_cvref_t<typename Traits::WarpGemm>;
|
||||
@@ -268,9 +267,9 @@ struct ABQuantBlockUniversalGemmAsBsCr : public BlockGemmQuantBase
|
||||
bool_constant<ALoadTranspose> = {},
|
||||
bool_constant<BLoadTranspose> = {})
|
||||
{
|
||||
load_int4_tile<ADataType, ComputeDataType, UnaryOpSize_, ALoadTranspose>(
|
||||
// If A/B datatype were pkint4/pkfp4 it would be converted prior to storing in LDS
|
||||
load_int4_tile<OverrideADataType, ComputeDataType, UnaryOpSize_, ALoadTranspose>(
|
||||
a_warp_tile_, a_block_window);
|
||||
// If B datatype were pkint4 it would be converted prior to storing in LDS
|
||||
load_int4_tile<OverrideBDataType, ComputeDataType, UnaryOpSize_, BLoadTranspose>(
|
||||
b_warp_tile_, b_block_window);
|
||||
}
|
||||
|
||||
@@ -10,9 +10,10 @@
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
struct GemmABQuantPipelineAgBgCrDefaultPolicy : public UniversalGemmPipelineAgBgCrPolicy
|
||||
struct GemmABQuantPipelineAgBgCrDefaultPolicy
|
||||
: public UniversalGemmBasePolicy<GemmABQuantPipelineAgBgCrDefaultPolicy>
|
||||
{
|
||||
using Base = UniversalGemmPipelineAgBgCrPolicy;
|
||||
using Base = UniversalGemmBasePolicy<GemmABQuantPipelineAgBgCrDefaultPolicy>;
|
||||
using Base::I0;
|
||||
using Base::I1;
|
||||
using Base::I2;
|
||||
|
||||
@@ -34,9 +34,6 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
using BlockGemmShape = remove_cvref_t<typename Problem::BlockGemmShape>;
|
||||
using AQuantGroupSize = remove_cvref_t<typename Problem::AQuantGroupSize>;
|
||||
using BQuantGroupSize = remove_cvref_t<typename Problem::BQuantGroupSize>;
|
||||
// BDataType gets converted from PkInt4 during loading
|
||||
using OverrideBDataType =
|
||||
std::conditional_t<std::is_same_v<BDataType, pk_int4_t>, ADataType, BDataType>;
|
||||
|
||||
static_assert(BQuantGroupSize::kM == 1, "only N/K blocks for BQuant kernel!");
|
||||
static_assert(AQuantGroupSize::kN == 1, "only M/K blocks for AQuant kernel!");
|
||||
@@ -67,6 +64,10 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
|
||||
using BlockGemm = remove_cvref_t<decltype(Policy::template GetBlockGemm<Problem>())>;
|
||||
|
||||
// A/B DataType gets converted from PkInt4/PkFp4 during loading
|
||||
using OverrideADataType = BlockGemm::OverrideADataType;
|
||||
using OverrideBDataType = BlockGemm::OverrideBDataType;
|
||||
|
||||
static constexpr index_t BlockSize = Problem::kBlockSize;
|
||||
static constexpr index_t MPerBlock = BlockGemmShape::kM;
|
||||
static constexpr index_t NPerBlock = BlockGemmShape::kN;
|
||||
@@ -281,9 +282,9 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
using AQDramTileWindowStep = typename AQDramBlockWindowTmp::BottomTensorIndex;
|
||||
using BQDramTileWindowStep = typename BQDramBlockWindowTmp::BottomTensorIndex;
|
||||
|
||||
// Note: BDataType PkInt4 gets converted during loading, before going to LDS
|
||||
// Note: A/B DataType PkInt4/PkFp4 gets converted during loading, before going to LDS
|
||||
auto&& [a_lds_block, b_lds_block] =
|
||||
Base::template GetABLdsTensorViews<ADataType, OverrideBDataType>(p_smem);
|
||||
Base::template GetABLdsTensorViews<OverrideADataType, OverrideBDataType>(p_smem);
|
||||
|
||||
constexpr auto a_lds_load_tile_distr =
|
||||
make_static_tile_distribution(BlockGemm::MakeABlockDistributionEncode());
|
||||
@@ -303,9 +304,9 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
using BQBlockTileDistr = decltype(bq_copy_dram_window.get_tile_distribution());
|
||||
|
||||
using ABlockTile =
|
||||
decltype(make_static_distributed_tensor<ADataType>(ABlockTileDistr{}));
|
||||
decltype(make_static_distributed_tensor<OverrideADataType>(ABlockTileDistr{}));
|
||||
using BBlockTile =
|
||||
decltype(make_static_distributed_tensor<BDataType>(BBlockTileDistr{}));
|
||||
decltype(make_static_distributed_tensor<OverrideBDataType>(BBlockTileDistr{}));
|
||||
using AQBlockTile =
|
||||
decltype(make_static_distributed_tensor<AQDataType>(AQBlockTileDistr{}));
|
||||
using BQBlockTile =
|
||||
@@ -361,7 +362,7 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
|
||||
if constexpr(is_a_col_major && !is_a_load_tr_v())
|
||||
{
|
||||
auto a_shuffle_tmp = make_static_distributed_tensor<ADataType>(
|
||||
auto a_shuffle_tmp = make_static_distributed_tensor<OverrideADataType>(
|
||||
Policy::template MakeShuffledARegTileDistribution<Problem>());
|
||||
transpose_tile2d(a_shuffle_tmp, a_block_tile);
|
||||
Base::LocalPrefill(a_copy_lds_window, a_shuffle_tmp, a_element_func);
|
||||
@@ -373,7 +374,7 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
|
||||
if constexpr(is_b_row_major && !is_b_load_tr_v())
|
||||
{
|
||||
auto b_shuffle_tmp = make_static_distributed_tensor<BDataType>(
|
||||
auto b_shuffle_tmp = make_static_distributed_tensor<OverrideBDataType>(
|
||||
Policy::template MakeShuffledBRegTileDistribution<Problem>());
|
||||
transpose_tile2d(b_shuffle_tmp, b_block_tile);
|
||||
Base::LocalPrefill(b_copy_lds_window, b_shuffle_tmp, b_element_func);
|
||||
@@ -409,7 +410,8 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
|
||||
if constexpr(is_a_col_major && !is_a_load_tr_v())
|
||||
{
|
||||
auto a_shuffle_tmp = make_static_distributed_tensor<ADataType>(
|
||||
// Note: ABDataType PkInt4/PkFp4 gets converted during loading earlier
|
||||
auto a_shuffle_tmp = make_static_distributed_tensor<OverrideADataType>(
|
||||
Policy::template MakeShuffledARegTileDistribution<Problem>());
|
||||
transpose_tile2d(a_shuffle_tmp, a_block_tile);
|
||||
Base::LocalPrefill(a_copy_lds_window, a_shuffle_tmp, a_element_func);
|
||||
@@ -420,7 +422,7 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
}
|
||||
if constexpr(is_b_row_major && !is_b_load_tr_v())
|
||||
{
|
||||
// Note: BDataType PkInt4 gets converted during loading earlier
|
||||
// Note: BDataType PkInt4/PkFp4 gets converted during loading earlier
|
||||
auto b_shuffle_tmp = make_static_distributed_tensor<OverrideBDataType>(
|
||||
Policy::template MakeShuffledBRegTileDistribution<Problem>());
|
||||
transpose_tile2d(b_shuffle_tmp, b_block_tile);
|
||||
@@ -493,7 +495,8 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
|
||||
if constexpr(is_a_col_major && !is_a_load_tr_v())
|
||||
{
|
||||
auto a_shuffle_tmp = make_static_distributed_tensor<ADataType>(
|
||||
// Note: ADataType gets converted during loading from PkInt4/PkFp4
|
||||
auto a_shuffle_tmp = make_static_distributed_tensor<OverrideADataType>(
|
||||
Policy::template MakeShuffledARegTileDistribution<Problem>());
|
||||
transpose_tile2d(a_shuffle_tmp, a_block_tile);
|
||||
Base::LocalPrefill(a_copy_lds_window, a_shuffle_tmp, a_element_func);
|
||||
@@ -543,9 +546,9 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
|
||||
return PipelineImpl<Scheduler>{}.template operator()<HasHotLoop, TailNum>(
|
||||
a_dram_block_window_tmp,
|
||||
[](const ADataType& a) { return a; },
|
||||
[](const OverrideADataType& a) { return a; },
|
||||
b_dram_block_window_tmp,
|
||||
[](const BDataType& b) { return b; },
|
||||
[](const OverrideBDataType& b) { return b; },
|
||||
aq_dram_block_window_tmp,
|
||||
bq_dram_block_window_tmp,
|
||||
m,
|
||||
@@ -593,9 +596,10 @@ struct ABQuantGemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Pro
|
||||
|
||||
return PipelineImpl<Scheduler>{}.template operator()<hot_loop, tail_num>(
|
||||
a_dram_block_window_tmp,
|
||||
[](const ADataType& a) { return a; },
|
||||
// Note: ADataType PkInt4/PkFp4 gets converted during loading
|
||||
[](const OverrideADataType& a) { return a; },
|
||||
b_dram_block_window_tmp,
|
||||
// Note: BDataType PkInt4 gets converted during loading
|
||||
// Note: BDataType PkInt4/PkFp4 gets converted during loading
|
||||
[](const OverrideBDataType& b) { return b; },
|
||||
aq_dram_block_window_tmp,
|
||||
bq_dram_block_window_tmp,
|
||||
|
||||
@@ -21,23 +21,27 @@ template <typename ADataType_,
|
||||
typename AQuantGroupSize_,
|
||||
typename BQuantGroupSize_,
|
||||
bool TransposeC_,
|
||||
typename ComputeDataType_ = BDataType_,
|
||||
typename ComputeDataType_ = void,
|
||||
GemmPipelineScheduler Scheduler_ = GemmPipelineScheduler::Intrawave,
|
||||
bool HasHotLoop_ = true,
|
||||
TailNumber TailNum_ = TailNumber::Full>
|
||||
struct GemmQuantPipelineProblemBase : public GemmPipelineProblemBase<ADataType_,
|
||||
BDataType_,
|
||||
CDataType_,
|
||||
BlockGemmShape_,
|
||||
Traits_,
|
||||
ComputeDataType_>
|
||||
struct GemmQuantPipelineProblemBase
|
||||
: public GemmPipelineProblemBase<
|
||||
ADataType_,
|
||||
BDataType_,
|
||||
CDataType_,
|
||||
BlockGemmShape_,
|
||||
Traits_,
|
||||
mixed_prec_compute_type_t<ComputeDataType_, ADataType_, BDataType_>>
|
||||
{
|
||||
using Base = GemmPipelineProblemBase<ADataType_,
|
||||
BDataType_,
|
||||
CDataType_,
|
||||
BlockGemmShape_,
|
||||
Traits_,
|
||||
ComputeDataType_>;
|
||||
|
||||
using Base = GemmPipelineProblemBase<
|
||||
ADataType_,
|
||||
BDataType_,
|
||||
CDataType_,
|
||||
BlockGemmShape_,
|
||||
Traits_,
|
||||
mixed_prec_compute_type_t<ComputeDataType_, ADataType_, BDataType_>>;
|
||||
|
||||
using Traits = typename Base::Traits;
|
||||
|
||||
|
||||
@@ -95,11 +95,6 @@ struct GemmWPABQuantPipelineAgBgCrPolicy : public UniversalWeightPreshufflePipel
|
||||
using BlockWarps = typename Problem::BlockGemmShape::BlockWarps;
|
||||
using WarpTile = typename Problem::BlockGemmShape::WarpTile;
|
||||
|
||||
using BTypeToUse =
|
||||
std::conditional_t<std::is_same_v<typename Problem::BDataType, ck_tile::pk_int4_t>,
|
||||
typename Problem::ADataType,
|
||||
typename Problem::BDataType>;
|
||||
|
||||
constexpr index_t WaveSize = get_warp_size();
|
||||
constexpr index_t KLane = WarpTile::at(I2) * WarpTile::at(I0) / WaveSize;
|
||||
using BDataType = typename Problem::BDataType;
|
||||
@@ -107,8 +102,8 @@ struct GemmWPABQuantPipelineAgBgCrPolicy : public UniversalWeightPreshufflePipel
|
||||
KLane / numeric_traits<BDataType>::PackedSize * sizeof(BDataType);
|
||||
constexpr auto NumAccess = static_cast<WGAttrNumAccessEnum>(max(1, KLaneBytes / 16));
|
||||
|
||||
using WarpGemm = WarpGemmDispatcher<typename Problem::ADataType,
|
||||
BTypeToUse,
|
||||
using WarpGemm = WarpGemmDispatcher<typename Problem::ComputeDataType,
|
||||
typename Problem::ComputeDataType,
|
||||
typename Problem::CDataType,
|
||||
WarpTile::at(I0),
|
||||
WarpTile::at(I1),
|
||||
|
||||
@@ -7,6 +7,7 @@
|
||||
#include <sstream>
|
||||
|
||||
#include "ck_tile/core.hpp"
|
||||
#include "ck_tile/ops/common/load_interleaved_pk_type.hpp"
|
||||
#include "ck_tile/ops/gemm/pipeline/gemm_universal_pipeline_ag_bg_cr_policy.hpp"
|
||||
#include "ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_scheduler.hpp"
|
||||
#include "ck_tile/ops/gemm/pipeline/wp_pipeline_agmem_bgmem_creg_v2.hpp"
|
||||
@@ -239,36 +240,42 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
make_tensor_view<address_space_enum::lds>(p_a_lds_pong, a_lds_block_desc);
|
||||
|
||||
// A DRAM tile window for load
|
||||
auto a_dram_tile_distribution =
|
||||
PipelinePolicy::template MakeADramTileDistribution<Problem>();
|
||||
|
||||
auto a_copy_dram_window =
|
||||
make_tile_window(a_dram_block_window_tmp.get_bottom_tensor_view(),
|
||||
make_tuple(number<kMPerBlock>{}, number<kKPerBlock>{}),
|
||||
a_dram_block_window_tmp.get_window_origin(),
|
||||
PipelinePolicy::template MakeADramTileDistribution<Problem>());
|
||||
a_dram_tile_distribution);
|
||||
|
||||
auto a_copy_lds_window_ping =
|
||||
make_tile_window(a_lds_block_ping,
|
||||
make_tuple(number<kMPerBlock>{}, number<kKPerBlock>{}),
|
||||
{0, 0},
|
||||
PipelinePolicy::template MakeADramTileDistribution<Problem>());
|
||||
a_dram_tile_distribution);
|
||||
|
||||
auto a_copy_lds_window_pong =
|
||||
make_tile_window(a_lds_block_pong,
|
||||
make_tuple(number<kMPerBlock>{}, number<kKPerBlock>{}),
|
||||
{0, 0},
|
||||
PipelinePolicy::template MakeADramTileDistribution<Problem>());
|
||||
a_dram_tile_distribution);
|
||||
|
||||
// ping-pong window for A LDS
|
||||
auto a_warp_tile_distribution =
|
||||
make_static_tile_distribution(typename WG::AWarpDstrEncoding{});
|
||||
|
||||
auto a_warp_window_ping_tmp =
|
||||
make_tile_window(a_lds_block_ping,
|
||||
make_tuple(number<WG::kM>{}, number<WG::kK>{}),
|
||||
{iMWarp * WG::kM, 0},
|
||||
make_static_tile_distribution(typename WG::AWarpDstrEncoding{}));
|
||||
a_warp_tile_distribution);
|
||||
|
||||
auto a_warp_window_pong_tmp =
|
||||
make_tile_window(a_lds_block_pong,
|
||||
make_tuple(number<WG::kM>{}, number<WG::kK>{}),
|
||||
{iMWarp * WG::kM, 0},
|
||||
make_static_tile_distribution(typename WG::AWarpDstrEncoding{}));
|
||||
a_warp_tile_distribution);
|
||||
|
||||
statically_indexed_array<
|
||||
statically_indexed_array<decltype(a_warp_window_ping_tmp), KIterPerWarp>,
|
||||
@@ -314,7 +321,7 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
b_flat_distribution);
|
||||
|
||||
using BTypeToUse =
|
||||
std::conditional_t<std::is_same_v<BDataType, pk_int4_t>, ADataType, BDataType>;
|
||||
mixed_prec_compute_type_from_input_t<BDataType, ADataType, ComputeDataType>;
|
||||
using BTileType = decltype(make_static_distributed_tensor<BTypeToUse>(b_flat_distribution));
|
||||
|
||||
// pingpong buffer for B
|
||||
@@ -354,7 +361,7 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
move_tile_window(b_flat_dram_windows(nIter)(kIter),
|
||||
{nIter * flatNPerWarp, kIter * flatKPerWarp});
|
||||
|
||||
load_int4_tile<BDataType, ADataType, UnaryOpSize_>(
|
||||
load_int4_tile<BDataType, BTypeToUse, UnaryOpSize_>(
|
||||
b_warp_tensor_ping(nIter)(kIter), b_flat_dram_windows(nIter)(kIter));
|
||||
});
|
||||
});
|
||||
@@ -393,15 +400,17 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
block_sync_lds();
|
||||
|
||||
// preload A00,A10 from lds
|
||||
statically_indexed_array<decltype(load_tile(a_warp_windows_ping(number<0>{})(number<0>{}))),
|
||||
m_preload>
|
||||
a_warp_tensor;
|
||||
using ATypeToUse =
|
||||
mixed_prec_compute_type_from_input_t<ADataType, BDataType, ComputeDataType>;
|
||||
using ATileType =
|
||||
decltype(make_static_distributed_tensor<BTypeToUse>(a_warp_tile_distribution));
|
||||
statically_indexed_array<ATileType, m_preload> a_warp_tensor;
|
||||
|
||||
static_for<0, m_preload, 1>{}([&](auto loadIter) {
|
||||
constexpr auto mIter = loadIter % MIterPerWarp;
|
||||
constexpr auto kIter = loadIter / MIterPerWarp;
|
||||
a_warp_tensor(loadIter) =
|
||||
load_tile(a_warp_windows_ping(number<mIter>{})(number<kIter>{}));
|
||||
load_int4_tile<ADataType, ATypeToUse, UnaryOpSize_>(
|
||||
a_warp_tensor(loadIter), a_warp_windows_ping(number<mIter>{})(number<kIter>{}));
|
||||
});
|
||||
__builtin_amdgcn_sched_barrier(0);
|
||||
|
||||
@@ -434,7 +443,7 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
|
||||
move_tile_window(b_flat_dram_windows(nIter)(kIter),
|
||||
{nIter * flatNPerWarp, kIter * flatKPerWarp});
|
||||
load_int4_tile<BDataType, ADataType, UnaryOpSize_>(
|
||||
load_int4_tile<BDataType, BTypeToUse, UnaryOpSize_>(
|
||||
b_warp_tensor_pong(nIter)(kIter), b_flat_dram_windows(nIter)(kIter));
|
||||
});
|
||||
});
|
||||
@@ -450,8 +459,8 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
static_for<0, m_preload, 1>{}([&](auto loadIter) {
|
||||
constexpr auto mIter = loadIter % MIterPerWarp;
|
||||
constexpr auto kIter = loadIter / MIterPerWarp;
|
||||
a_warp_tensor(loadIter) =
|
||||
load_tile(a_warp_windows_pong(number<mIter>{})(number<kIter>{}));
|
||||
load_int4_tile<ADataType, ATypeToUse, UnaryOpSize_>(
|
||||
a_warp_tensor(loadIter), a_warp_windows_pong(number<mIter>{})(number<kIter>{}));
|
||||
});
|
||||
|
||||
// Next K
|
||||
@@ -463,7 +472,7 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
|
||||
move_tile_window(b_flat_dram_windows(nIter)(kIter),
|
||||
{nIter * flatNPerWarp, kIter * flatKPerWarp});
|
||||
load_int4_tile<BDataType, ADataType, UnaryOpSize_>(
|
||||
load_int4_tile<BDataType, BTypeToUse, UnaryOpSize_>(
|
||||
b_warp_tensor_ping(nIter)(kIter), b_flat_dram_windows(nIter)(kIter));
|
||||
});
|
||||
});
|
||||
@@ -495,8 +504,8 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
static_for<0, m_preload, 1>{}([&](auto loadIter) {
|
||||
constexpr auto mIter = loadIter % MIterPerWarp;
|
||||
constexpr auto kIter = loadIter / MIterPerWarp;
|
||||
a_warp_tensor(loadIter) =
|
||||
load_tile(a_warp_windows_ping(number<mIter>{})(number<kIter>{}));
|
||||
load_int4_tile<ADataType, ATypeToUse, UnaryOpSize_>(
|
||||
a_warp_tensor(loadIter), a_warp_windows_ping(number<mIter>{})(number<kIter>{}));
|
||||
});
|
||||
iCounter--;
|
||||
HotLoopScheduler<loop_count>();
|
||||
@@ -513,7 +522,7 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
move_tile_window(b_flat_dram_windows(nIter)(kIter),
|
||||
{nIter * flatNPerWarp, kIter * flatKPerWarp});
|
||||
|
||||
load_int4_tile<BDataType, ADataType, UnaryOpSize_>(
|
||||
load_int4_tile<BDataType, BTypeToUse, UnaryOpSize_>(
|
||||
b_warp_tensor_pong(nIter)(kIter), b_flat_dram_windows(nIter)(kIter));
|
||||
});
|
||||
});
|
||||
@@ -535,8 +544,8 @@ struct WPABQuantBPipelineAgBgCrV2 : public WeightPreshufflePipelineAGmemBGmemCRe
|
||||
static_for<0, m_preload, 1>{}([&](auto loadIter) {
|
||||
constexpr auto mIter = loadIter % MIterPerWarp;
|
||||
constexpr auto kIter = loadIter / MIterPerWarp;
|
||||
a_warp_tensor(loadIter) =
|
||||
load_tile(a_warp_windows_pong(number<mIter>{})(number<kIter>{}));
|
||||
load_int4_tile<ADataType, ATypeToUse, UnaryOpSize_>(
|
||||
a_warp_tensor(loadIter), a_warp_windows_pong(number<mIter>{})(number<kIter>{}));
|
||||
});
|
||||
|
||||
// GEMM loopK
|
||||
|
||||
Reference in New Issue
Block a user