ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-07 23:57:14 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
develop
composable_kernel/dispatcher/examples/fmha/cpp/28_bwd_masks_fmha.cpp
Vidyasagar Ananthan b20458e19e [rocm-libraries] ROCm/rocm-libraries#5260 (commit a1834d2) 
[CK] [CK_Tile] Add FMHA scaffolding to CK kernel dispatcher (#5260)

## Motivation

The CK Tile dispatcher currently supports GEMM and Grouped Convolution
but has no support for Fused Multi-Head Attention (FMHA). The
example/ck_tile/01_fmha folder contains a comprehensive FMHA
implementation with forward, backward, split-KV, paged-KV, append-KV,
and batch-prefill kernels across multiple GPU architectures — but there
is no unified dispatch layer for it. This PR ports the FMHA stack into
the dispatcher, following the same architectural patterns established by
GEMM and Grouped Convolution, enabling runtime kernel selection, JIT
compilation from Python, and a declarative C++ example flow. Autotuning
heuristics to follow.

## Technical Details

This PR adds FMHA scaffolding to the CK dispatcher framework, mirroring
GEMM's layered architecture. Seven new C++ runtime headers provide type
definitions (coexisting with upstream headers via __has_include,
requiring zero modifications to example/ck_tile/01_fmha/), a problem
builder with 18+ setters, Signature + Algorithm kernel key matching, a
virtual kernel instance, a DECL_FMHA_KERNEL_SET macro with wildcard
support and named tile/wave/warp setters, arch-aware registry with JSON
export, and a dispatcher with seqtune-aware selection, configurable
timing, and multi-stage execution plans for split-KV (two-stage) and
backward (three-stage). The codegen pipeline is driven by a
fmha_arch_specs.json capturing per-arch tile tables and pipeline
constraints for five architectures (gfx90a/942/950/1100/1201), migrated
from hardcoded logic in 01_fmha/codegen/, with supporting modules for
C++ symbol mappings, validation rules, and named receipt profiles
(ck_default, flash, pytorch, aiter, fp32, fp8). Python integration
(fmha_utils.py) mirrors the C++ layer with JIT compilation, parallel
multi-kernel builds, HIP memory management via ctypes, tolerance-based
validation, and a NumPy CPU reference with GQA support. Twenty-seven C++
and thirty-two Python examples cover the full feature surface — forward,
split-KV, masks, bias, dropout, GQA, backward, append-KV, batch prefill,
fp8, logits soft cap, sink tokens, and parameter sweeps — all
JIT-compiled on the fly.

## Test Plan

Seven test files cover the runtime types, codegen, and end-to-end
correctness. C++ unit tests validate the problem builder, dispatcher
planning (single-stage for forward/paged-KV/append-KV; multi-stage for
split-KV and backward), registry operations, and the kernel-set
declaration macro. Python unit tests verify codegen emission, profile
filtering, and 15 validation rules for masks, hdim constraints, and
pipeline requirements. GPU execution validation in 01_basic_fmha
--validate reports zero errors across 65,536 elements with max absolute
error of 7.29e-05. A gold-standard parity suite (test_fmha_parity.py)
runs 14 configurations through both the upstream tile_example_fmha_fwd
and the dispatcher, comparing exit codes to confirm behavioral parity —
all 14 match.

## Test Result

The C++ smoke test builds and passes all 9 compiled examples, and a
Python JIT sweep (29_sweep_seqlen.py) passes 7/7 configurations reaching
up to 375 TFLOPS at seqlen 2048.

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

---------

Co-authored-by: Yaswanth Raparti <113389104+yraparti@users.noreply.github.com>
Co-authored-by: Mohsen Saffari <mohsen.saffari@amd.com>
Co-authored-by: Maksim (Max) Podkorytov <Maksim.Podkorytov@amd.com>
Co-authored-by: yashagar <yashagar@amd.com>
2026-05-17 00:29:40 -07:00

490 lines
19 KiB
C++
Raw Permalink Blame History

// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
//
// Example 28: FMHA Backward with Causal Mask
//
// Demonstrates:
// 1. Forward kernel with top_left causal mask + LSE
// 2. Backward kernel families (bwd_dot_do_o, bwd_dq_dk_dv, bwd_convert_dq) with causal mask
// 3. GPU forward execution with causal mask validation
// 4. Backward 3-stage plan display
//
// Backward kernels use planning only -- actual backward GPU execution requires
// all 3 stages to compile, and bwd_dq_dk_dv has tile structure issues on gfx950.
#include <hip/hip_runtime.h>
#include <cmath>
#include <iomanip>
#include <iostream>
#include <random>
#include <vector>
#include "ck_tile/dispatcher.hpp"
#include "ck_tile/dispatcher/example_args.hpp"
using namespace ck_tile::dispatcher;
using namespace ck_tile::dispatcher::utils;
DECL_FMHA_KERNEL_SET(bwd_masks_fmha_kernels,
// Forward: causal mask (top_left) with LSE for backward
.add(FmhaSignature()
.family("fwd")
.dtype("fp16")
.mode("batch")
.vlayout("r")
.hdim(128)
.mask("top_left")
.bias("no")
.lse(true)
.dropout(false)
.qscale("no"),
FmhaAlgorithm()
.tile_m0(128)
.tile_n0(128)
.tile_k0(32)
.tile_n1(128)
.tile_k1(32)
.tile_k0max(128)
.wave_m0(4)
.wave_n0(1)
.wave_k0(1)
.wave_m1(4)
.wave_n1(1)
.wave_k1(1)
.warp_m0(32)
.warp_n0(32)
.warp_k0(16)
.warp_m1(32)
.warp_n1(32)
.warp_k1(16)
.pipeline("qr_async")
.padding(true, true, true, true)
.alignments(128, 128)
.selection_rank(0),
"gfx950")
// Backward stage 1: dot(dO, O) with causal mask
.add(FmhaSignature()
.family("bwd_dot_do_o")
.dtype("fp16")
.mode("batch")
.hdim(128)
.mask("top_left")
.bias("no")
.dropout(false)
.dbias(false)
.store_randval(false)
.deterministic(false),
FmhaAlgorithm()
.tile_m0(64)
.tile_n0(128)
.tile_k0(32)
.tile_n1(0)
.tile_k1(0)
.tile_k0max(0)
.padding(true, true, true, true)
.selection_rank(0),
"gfx950")
// Backward stage 2: compute dQ, dK, dV with causal mask
.add(FmhaSignature()
.family("bwd_dq_dk_dv")
.dtype("fp16")
.mode("batch")
.hdim(128)
.mask("top_left")
.bias("no")
.dropout(false)
.dbias(false)
.store_randval(false)
.deterministic(false),
FmhaAlgorithm()
.tile_m0(16)
.tile_n0(128)
.tile_k0(128)
.tile_n1(16)
.tile_k1(128)
.tile_k0max(32)
.wave(1, 4, 1, 4, 1, 1, 1, 4, 1)
.warp(16, 16, 32, 16, 16, 16, 16, 16, 16)
.padding(true, true, true, true)
.max_seq_len_q(0)
.selection_rank(0),
"gfx950")
// Backward stage 3: convert accumulated dQ from fp32 to fp16
.add(FmhaSignature()
.family("bwd_convert_dq")
.dtype("fp16")
.mode("batch")
.hdim(128)
.mask("top_left")
.bias("no")
.dropout(false)
.dbias(false)
.store_randval(false)
.deterministic(false),
FmhaAlgorithm()
.tile_m0(64)
.tile_n0(128)
.tile_k0(0)
.tile_n1(0)
.tile_k1(0)
.tile_k0max(0)
.padding(true, true, true, true)
.selection_rank(0),
"gfx950"));
namespace {
using FmhaDataType = ck_tile::fp16_t;
void cpu_attention_fwd_causal(const std::vector<float>& Q,
const std::vector<float>& K,
const std::vector<float>& V,
std::vector<float>& O,
std::vector<float>& LSE,
int batch,
int nhead,
int seqlen,
int hdim,
float scale)
{
for(int b = 0; b < batch; ++b)
{
for(int h = 0; h < nhead; ++h)
{
for(int sq = 0; sq < seqlen; ++sq)
{
std::vector<float> scores(seqlen, 0.0f);
float max_score = -1e30f;
for(int sk = 0; sk < seqlen; ++sk)
{
float dot = 0.0f;
for(int d = 0; d < hdim; ++d)
{
int q_idx = ((b * nhead + h) * seqlen + sq) * hdim + d;
int k_idx = ((b * nhead + h) * seqlen + sk) * hdim + d;
dot += Q[q_idx] * K[k_idx];
}
float s = dot * scale;
// top_left causal: mask if sk > sq
if(sk > sq)
s = -1e30f;
scores[sk] = s;
max_score = std::max(max_score, scores[sk]);
}
float sum_exp = 0.0f;
for(int sk = 0; sk < seqlen; ++sk)
{
scores[sk] = std::exp(scores[sk] - max_score);
sum_exp += scores[sk];
}
int lse_idx = (b * nhead + h) * seqlen + sq;
LSE[lse_idx] = max_score + std::log(sum_exp);
for(int sk = 0; sk < seqlen; ++sk)
scores[sk] /= sum_exp;
for(int dv = 0; dv < hdim; ++dv)
{
float acc = 0.0f;
for(int sk = 0; sk < seqlen; ++sk)
{
int v_idx = ((b * nhead + h) * seqlen + sk) * hdim + dv;
acc += scores[sk] * V[v_idx];
}
int o_idx = ((b * nhead + h) * seqlen + sq) * hdim + dv;
O[o_idx] = acc;
}
}
}
}
}
} // namespace
int main(int argc, char* argv[])
{
ExampleArgs args("Example 28: FMHA Backward with Masks",
"Causal mask forward (GPU) + backward plan");
args.add_option("--arch", "gfx950", "GPU architecture");
args.add_option("--batch", "2", "Batch size");
args.add_option("--nhead", "4", "Number of heads");
args.add_option("--seqlen", "64", "Sequence length");
args.add_option("--hdim", "128", "Head dimension");
if(!args.parse(argc, argv))
return 0;
const std::string gfx_arch = args.get("--arch", "gfx950");
const int batch = args.get_int("--batch", 2);
const int nhead = args.get_int("--nhead", 4);
const int seqlen = args.get_int("--seqlen", 64);
const int hdim = args.get_int("--hdim", 128);
const float scale = 1.0f / std::sqrt(static_cast<float>(hdim));
print_header("Example 28: FMHA Backward with Causal Mask");
// Step 1: Register kernels
std::cout << "\nStep 1: Register Kernels\n";
FmhaKernelSetRegistry::instance().print();
FmhaRegistry registry;
registry.set_name("bwd_masks_fmha");
REGISTER_GENERATED_KERNELS(registry, gfx_arch);
std::cout << " Registered " << registry.size() << " kernel(s)\n";
FmhaDispatcher dispatcher(&registry);
dispatcher.set_benchmarking(true);
dispatcher.set_timing(1, 3);
// Step 2: Plan backward (3-stage) with causal mask
std::cout << "\nStep 2: Plan Backward (causal mask)\n";
fmha_bwd_traits bwd_traits{};
bwd_traits.hdim_q = hdim;
bwd_traits.hdim_v = hdim;
bwd_traits.data_type = "fp16";
bwd_traits.is_group_mode = false;
bwd_traits.mask_type = mask_enum::mask_top_left;
bwd_traits.bias_type = bias_enum::no_bias;
bwd_traits.has_dbias = false;
bwd_traits.has_dropout = false;
bwd_traits.is_store_randval = false;
bwd_traits.is_deterministic = false;
fmha_bwd_args bwd_args{};
bwd_args.batch = batch;
bwd_args.seqlen_q = seqlen;
bwd_args.seqlen_k = seqlen;
bwd_args.max_seqlen_q = seqlen;
bwd_args.max_seqlen_k = seqlen;
bwd_args.hdim_q = hdim;
bwd_args.hdim_v = hdim;
bwd_args.nhead_q = nhead;
bwd_args.nhead_k = nhead;
auto bwd_plan = dispatcher.plan(
FmhaProblem::from_invocation(FmhaInvocation::make(bwd_traits, bwd_args), gfx_arch));
if(bwd_plan.is_valid() && bwd_plan.stages.size() >= 2)
{
std::cout << " Backward plan stages (" << bwd_plan.stages.size() << "):\n";
for(const auto& stage : bwd_plan.stages)
{
std::cout << " " << to_string(stage.family) << " -> " << stage.kernel_id << "\n";
}
}
else
{
std::cout << " Backward plan: INVALID or single-stage (expected 3 stages)\n";
std::cout << " This is expected -- backward planning shows the pattern\n";
}
// Step 3: Run forward on GPU with causal mask
std::cout << "\nStep 3: Run Forward (causal mask, GPU)\n";
const int64_t qkv_elems = static_cast<int64_t>(batch) * nhead * seqlen * hdim;
const int64_t lse_elems = static_cast<int64_t>(batch) * nhead * seqlen;
GpuBuffer<FmhaDataType> q_dev(qkv_elems);
GpuBuffer<FmhaDataType> k_dev(qkv_elems);
GpuBuffer<FmhaDataType> v_dev(qkv_elems);
GpuBuffer<FmhaDataType> o_dev(qkv_elems);
GpuBuffer<float> lse_dev(lse_elems);
std::mt19937 rng(42);
std::uniform_real_distribution<float> dist(-0.5f, 0.5f);
std::vector<FmhaDataType> q_host(qkv_elems), k_host(qkv_elems), v_host(qkv_elems);
for(auto& x : q_host)
x = FmhaDataType(dist(rng));
for(auto& x : k_host)
x = FmhaDataType(dist(rng));
for(auto& x : v_host)
x = FmhaDataType(dist(rng));
q_dev.copy_from_host(q_host.data());
k_dev.copy_from_host(k_host.data());
v_dev.copy_from_host(v_host.data());
o_dev.zero();
lse_dev.zero();
fmha_fwd_traits fwd_traits{};
fwd_traits.hdim_q = hdim;
fwd_traits.hdim_v = hdim;
fwd_traits.data_type = "fp16";
fwd_traits.is_group_mode = false;
fwd_traits.is_v_rowmajor = true;
fwd_traits.has_logits_soft_cap = false;
fwd_traits.mask_type = mask_enum::mask_top_left;
fwd_traits.bias_type = bias_enum::no_bias;
fwd_traits.has_lse = true;
fwd_traits.has_dropout = false;
fwd_traits.qscale_type = quant_scale_enum::no_scale;
fmha_fwd_args fwd_args{};
fwd_args.q_ptr = q_dev.get();
fwd_args.k_ptr = k_dev.get();
fwd_args.v_ptr = v_dev.get();
fwd_args.o_ptr = o_dev.get();
fwd_args.lse_ptr = lse_dev.get();
fwd_args.bias_ptr = nullptr;
fwd_args.q_descale_ptr = nullptr;
fwd_args.k_descale_ptr = nullptr;
fwd_args.v_descale_ptr = nullptr;
fwd_args.rand_val_ptr = nullptr;
fwd_args.sink_ptr = nullptr;
fwd_args.block_scale_seqstart_q_ptr = nullptr;
fwd_args.block_scale_seqstart_k_ptr = nullptr;
fwd_args.seqlen_q = seqlen;
fwd_args.seqlen_k = seqlen;
fwd_args.batch = batch;
fwd_args.max_seqlen_q = seqlen;
fwd_args.hdim_q = hdim;
fwd_args.hdim_v = hdim;
fwd_args.nhead_q = nhead;
fwd_args.nhead_k = nhead;
fwd_args.scale_s = scale;
fwd_args.logits_soft_cap = 0.0f;
fwd_args.stride_q = hdim;
fwd_args.stride_k = hdim;
fwd_args.stride_v = hdim;
fwd_args.stride_bias = 0;
fwd_args.stride_randval = 0;
fwd_args.stride_o = hdim;
fwd_args.nhead_stride_q = seqlen * hdim;
fwd_args.nhead_stride_k = seqlen * hdim;
fwd_args.nhead_stride_v = seqlen * hdim;
fwd_args.nhead_stride_bias = 0;
fwd_args.nhead_stride_randval = 0;
fwd_args.nhead_stride_lse = seqlen;
fwd_args.nhead_stride_o = seqlen * hdim;
fwd_args.nhead_stride_q_descale = 0;
fwd_args.nhead_stride_k_descale = 0;
fwd_args.nhead_stride_v_descale = 0;
fwd_args.batch_stride_q = nhead * seqlen * hdim;
fwd_args.batch_stride_k = nhead * seqlen * hdim;
fwd_args.batch_stride_v = nhead * seqlen * hdim;
fwd_args.batch_stride_bias = 0;
fwd_args.batch_stride_randval = 0;
fwd_args.batch_stride_lse = nhead * seqlen;
fwd_args.batch_stride_o = nhead * seqlen * hdim;
fwd_args.batch_stride_q_descale = 0;
fwd_args.batch_stride_k_descale = 0;
fwd_args.batch_stride_v_descale = 0;
fwd_args.window_size_left = -1;
fwd_args.window_size_right = 0;
fwd_args.sink_size = 0;
fwd_args.mask_type = 1; // top_left
fwd_args.min_seqlen_q = 0;
fwd_args.p_drop = 0.0f;
fwd_args.s_randval = false;
fwd_args.drop_seed_offset = std::make_pair(uint64_t(0), uint64_t(0));
fwd_args.block_scale_size_q = 0;
fwd_args.block_scale_size_kv = 0;
bool fwd_passed = false;
try
{
float fwd_time = dispatcher.run_fwd(fwd_traits, fwd_args, nullptr);
std::cout << " Forward time: " << std::fixed << std::setprecision(4) << fwd_time
<< " ms\n";
auto problem =
FmhaProblem::from_invocation(FmhaInvocation::make(fwd_traits, fwd_args), gfx_arch);
double tflops = static_cast<double>(problem.num_ops()) / (fwd_time * 1e-3) / 1e12;
std::cout << " TFLOPS: " << std::setprecision(2) << tflops << "\n";
fwd_passed = true;
}
catch(const std::exception& e)
{
std::cerr << " Forward ERROR: " << e.what() << "\n";
}
// Step 4: Validate forward output
std::cout << "\nStep 4: Validate Forward Output\n";
if(fwd_passed)
{
std::vector<FmhaDataType> o_host(qkv_elems);
o_dev.copy_to_host(o_host.data());
std::vector<float> lse_host(lse_elems);
lse_dev.copy_to_host(lse_host.data());
std::vector<float> q_f32(qkv_elems), k_f32(qkv_elems), v_f32(qkv_elems);
for(int64_t i = 0; i < qkv_elems; ++i)
q_f32[i] = static_cast<float>(q_host[i]);
for(int64_t i = 0; i < qkv_elems; ++i)
k_f32[i] = static_cast<float>(k_host[i]);
for(int64_t i = 0; i < qkv_elems; ++i)
v_f32[i] = static_cast<float>(v_host[i]);
std::vector<float> o_ref(qkv_elems, 0.0f);
std::vector<float> lse_ref(lse_elems, 0.0f);
cpu_attention_fwd_causal(
q_f32, k_f32, v_f32, o_ref, lse_ref, batch, nhead, seqlen, hdim, scale);
double max_o_err = 0.0;
int o_errors = 0;
const double rtol = 1e-2;
const double atol = 1e-2;
for(int64_t i = 0; i < qkv_elems; ++i)
{
float gpu_val = static_cast<float>(o_host[i]);
float ref_val = o_ref[i];
double abs_err = std::abs(gpu_val - ref_val);
max_o_err = std::max(max_o_err, abs_err);
if(abs_err > atol + rtol * std::abs(ref_val))
++o_errors;
}
double max_lse_err = 0.0;
int lse_reasonable = 0;
for(int64_t i = 0; i < lse_elems; ++i)
{
if(std::isfinite(lse_host[i]) && std::abs(lse_host[i]) < 100.0f)
++lse_reasonable;
max_lse_err =
std::max(max_lse_err, static_cast<double>(std::abs(lse_host[i] - lse_ref[i])));
}
std::cout << " Output max abs error: " << std::scientific << max_o_err << "\n";
std::cout << " Output errors: " << o_errors << " / " << qkv_elems << "\n";
std::cout << " LSE reasonable: " << lse_reasonable << " / " << lse_elems << "\n";
std::cout << " LSE max error: " << std::scientific << max_lse_err << "\n";
fwd_passed = (o_errors == 0) && (lse_reasonable == lse_elems);
}
// Step 5: Show backward API pattern
std::cout << "\nStep 5: Backward API Pattern (traits + args)\n";
std::cout << " bwd_traits.mask_type = mask_top_left\n";
std::cout << " bwd_traits.bias_type = no_bias\n";
std::cout << " bwd_traits.has_dropout = false\n";
std::cout << " bwd_traits.is_deterministic = false\n";
std::cout << " bwd_args.window_size_left = -1\n";
std::cout << " bwd_args.window_size_right = 0 (causal)\n";
std::cout << " bwd_args.mask_type = 1 (top_left)\n";
std::cout << " Backward plan resolves to " << bwd_plan.stages.size() << " stage(s)\n";
print_separator();
std::cout << "Status: " << (fwd_passed ? "PASS" : "FAIL") << "\n";
print_separator();
return fwd_passed ? 0 : 1;
}
Reference in New Issue View Git Blame Copy Permalink