Files
composable_kernel/example/ck_tile/49_sageattention/mask.hpp
ltqin de0a61e5c2 [rocm-libraries] ROCm/rocm-libraries#6574 (commit b3db057)
[CK_TILE] Add SageAttention v2 forward kernel with
 multi-granularity quantization (#6574)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## Summary

Add a CK_TILE forward kernel implementing [SageAttention
v2](https://arxiv.org/abs/2411.10958) — an attention algorithm that
applies multi-granularity quantization to Q/K/V before computing
attention, trading minimal accuracy loss for higher throughput on
low-precision hardware.

### Quantization design

| Tensor | Supported data types | Scale granularity options |
|--------|---------------------|--------------------------|
| Q | fp8 / int8 / int4 | per-tensor, per-block (128 tokens), per-warp
(32 tokens), per-thread (4 tokens) |
| K | fp8 / int8 / int4 | per-tensor, per-block (128 tokens), per-warp
(64 tokens), per-thread (16 tokens) |
| V | fp8 | per-channel (always) |
| O | bf16 | — |

Three precision combinations are supported: `fp8/bf16` (QKV fp8, O
bf16), `i8/fp8/bf16` (QK int8, V fp8, O bf16), and `i4/fp8/bf16` (QK
int4, V fp8, O bf16).

### Architecture support

- **gfx9** (CDNA2/3, e.g. gfx90a, gfx942) — full tile set
- **gfx950** (CDNA4) — restricted tile set (N-per-block capped at 64 for
fp8-family dtypes)

### Implementation

- Two pipeline variants: `QRKSVS` (synchronous) and `QRKSVS_ASYNC`
(async copy)
- Masking support: no mask, causal (top-left / bottom-right), and
generic windowed
- Batch and group (variable-length) modes
- Head dimension: d=128, d_v=128
- Python codegen under `example/ck_tile/49_sageattention/codegen/`
generates kernel instances per target/dtype/tile combination
- Smoke tests included via `tile_example_sageattn_fwd`

### Test commands

\`\`\`bash
# fp8 QKV
./build/bin/tile_example_sageattn_fwd -v=1 -b=16 -h=8 -s=1024 -d=128
-kname=1 -prec=fp8bf16 -qscale=3 -init=3

# int8 QK, fp8 V
./build/bin/tile_example_sageattn_fwd -v=1 -b=16 -h=8 -s=1024 -d=128
-kname=1 -prec=i8fp8bf16 -qscale=3 -init=3
\`\`\`

\`-qscale\` values: 1=per-tensor, 2=per-block, 3=per-warp, 4=per-thread
2026-04-30 18:33:36 +00:00

170 lines
5.6 KiB
C++

// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#pragma once
#include <ostream>
#include <string>
#include "ck_tile/core.hpp"
#include "ck_tile/ops/fmha/block/block_masking.hpp"
// keep this in sync with ck_tile::GenericAttentionMaskEnum
enum class mask_enum
{
no_mask = 0,
mask_top_left,
mask_bottom_right,
window_generic,
};
struct mask_info
{
mask_enum type;
ck_tile::index_t seqlen_q;
ck_tile::index_t seqlen_k;
ck_tile::index_t y, x;
ck_tile::index_t left, right; // FA style SWA left/right
void serialize(std::ostream& os) const
{
if(type == mask_enum::no_mask)
os << "n";
else if(type == mask_enum::mask_top_left)
os << "t(" << left << ":" << right << ")";
else if(type == mask_enum::mask_bottom_right)
os << "b(" << left << ":" << right << ")";
else
{
os << "g(" << y << ":" << x << ")";
}
}
static mask_info decode(std::string str, ck_tile::index_t seqlen_q, ck_tile::index_t seqlen_k)
{
ck_tile::index_t x_total = seqlen_k;
ck_tile::index_t y_total = seqlen_q;
mask_info tmp;
tmp.seqlen_q = seqlen_q;
tmp.seqlen_k = seqlen_k;
auto found_0 = str.find(':');
if(found_0 != std::string::npos)
{
std::string t = str.substr(0, found_0);
std::string v = str.substr(found_0 + 1);
if(t == "xt" || t == "xb")
{
// xformer style sliding window attn from top-left
ck_tile::index_t window_size = std::stoi(v);
ck_tile::index_t left_size = -1;
ck_tile::index_t right_size = 0;
if(window_size > 0)
{
left_size = window_size / 2;
right_size = window_size - 1 - left_size;
}
auto r = ck_tile::make_generic_attention_mask_coordinates_from_lr_window(
left_size, right_size, 0, y_total, x_total, t == "xt");
tmp.type = t == "xt" ? mask_enum::mask_top_left : mask_enum::mask_bottom_right;
tmp.y = r.at(ck_tile::number<0>{});
tmp.x = r.at(ck_tile::number<1>{});
tmp.left = left_size;
tmp.right = right_size;
}
else if(t == "t" || t == "b" || t == "g")
{
auto found_1 = v.find(",");
if(found_1 == std::string::npos)
{
throw std::invalid_argument("invalid mask value: " + str);
}
tmp.type = mask_enum::window_generic;
ck_tile::index_t v0 = atoi(v.substr(0, found_1).c_str());
ck_tile::index_t v1 = atoi(v.substr(found_1 + 1).c_str());
// TODO: some validation
if(t == "t")
{
tmp.type = mask_enum::mask_top_left;
auto r = ck_tile::make_generic_attention_mask_coordinates_from_lr_window(
v0, v1, 0, y_total, x_total, true);
tmp.y = r.at(ck_tile::number<0>{});
tmp.x = r.at(ck_tile::number<1>{});
tmp.left = v0;
tmp.right = v1;
}
else if(t == "b")
{
tmp.type = mask_enum::mask_bottom_right;
auto r = ck_tile::make_generic_attention_mask_coordinates_from_lr_window(
v0, v1, 0, y_total, x_total, false);
tmp.y = r.at(ck_tile::number<0>{});
tmp.x = r.at(ck_tile::number<1>{});
tmp.left = v0;
tmp.right = v1;
}
else if(t == "g")
{
tmp.type = mask_enum::window_generic;
tmp.y = v0;
tmp.x = v1;
tmp.left = v0; // TODO: don't use this?
tmp.right = v1;
}
}
else
{
throw std::invalid_argument("invalid mask value: " + str);
}
}
else if(str == "0")
{
tmp.type = mask_enum::no_mask;
}
else if(str == "1" || str == "t")
{
tmp.type = mask_enum::mask_top_left;
tmp.y = seqlen_q;
tmp.x = 1;
tmp.left = -1;
tmp.right = 0;
}
else if(str == "2" || str == "b")
{
tmp.type = mask_enum::mask_bottom_right;
tmp.y = seqlen_q;
tmp.x = seqlen_k - seqlen_q + 1;
tmp.left = -1;
tmp.right = 0;
}
else
{
throw std::invalid_argument("invalid mask value: " + str);
}
return tmp;
}
std::size_t get_unmaskarea() const
{
if(type == mask_enum::no_mask)
return static_cast<std::size_t>(seqlen_q) * seqlen_k;
std::size_t area = 0;
for(ck_tile::index_t i_y = 0; i_y < seqlen_q; ++i_y)
{
ck_tile::index_t x_start = std::max(-y + i_y + 1, static_cast<ck_tile::index_t>(0));
ck_tile::index_t x_end = std::min(i_y + x, seqlen_k);
if(x_end > x_start)
{
area += (x_end - x_start);
}
}
return area;
}
friend std::ostream& operator<<(std::ostream& os, const mask_info& mi)
{
mi.serialize(os);
return os;
}
};