[FEAT] Add custom CUDA tinygemm unpacker (#415)

* add unpack cuda

* add tests

* fix tests

* refactor tinygemm unpacking kernel

* add dequant

* add additional dequant check

* update tinygemm dequantize test

* correct dequant kernel logic

* clean up kernel

* update dequantize kernel tests

* rename kernel ops to tensor_core_tiled_layout

* add renamed kernel source

* add back test_aot_dispatch opcheck

* rename innerKTiles to inner_k_tiles

* add unpack and dequant test

* additional numerical checks for unpack then dequant

* rebase test_ops on main

* remove commented out code

* skip dynamic opcheck unless torch>=2.5

test/test_ops.py

@@ -1,3 +1,7 @@
+import itertools
+
+import torchao
+
 import torch
 from torch.testing._internal.common_utils import (
     TestCase,
@@ -6,7 +10,7 @@
     run_tests,
 )
 from torch.testing._internal.optests import opcheck
-from torchao.utils import is_fbcode
+from torchao.utils import is_fbcode, TORCH_VERSION_AFTER_2_5
 from torchao.prototype.quant_llm import from_scaled_tc_fpx
 import pytest
 
@@ -18,6 +22,14 @@
 except RuntimeError:
     pytest.skip("torchao.ops not available")
 
+from torchao.quantization.utils import (
+    get_groupwise_affine_qparams,
+    groupwise_affine_dequantize_tensor_from_qparams,
+    groupwise_affine_quantize_tensor_from_qparams,
+    pack_tinygemm_scales_and_zeros,
+    unpack_tinygemm_scales_and_zeros,
+)
+
 
 class TestOps(TestCase):
     def _create_fpx_inputs(self, ebits: int, mbits: int, BS: int, OC: int, IC: int, device):
@@ -61,9 +73,218 @@ def test_quant_llm_linear_correctness(self, ebits, mbits, BS, OC, IC, splitK):
         relative_error = error / gt
         assert relative_error < 1e-3
 
-
 instantiate_parametrized_tests(TestOps)
 
 
+## Tests for `tensor_core_layout`
+kTileSizeN = 8
+kTileSizeK = 16
+
+SHAPES = [
+    (4096, 4096),
+    # Llama 2 GEMM shapes
+    (4096, 11008),
+    (11008, 4096),
+    # Llama 3 GEMM shapes
+    (4096, 14336),
+    (14336, 4096),
+]
+INNERKTILES = [2, 4, 8]
+QGROUP_SIZES = [32, 64, 128, 256]
+TEST_CONFIGS_UNPACK = list(itertools.product(SHAPES, INNERKTILES))
+TEST_CONFIGS_DEQUANT = list(itertools.product(SHAPES, INNERKTILES, QGROUP_SIZES))
+
+@pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")
+@pytest.mark.parametrize("shape, inner_k_tiles", TEST_CONFIGS_UNPACK, ids=str)
+def test_unpack_tensor_core_tiled_layout_correctness(shape, inner_k_tiles):
+    N, K = shape
+    assert K % (inner_k_tiles * kTileSizeK) == 0 and N % kTileSizeN == 0
+
+    t = torch.randint(0, 16, dtype=torch.int, size=shape, device="cuda")
+    packed_w = torch.ops.aten._convert_weight_to_int4pack(t, inner_k_tiles)
+    unpacked = torchao.ops.unpack_tensor_core_tiled_layout(packed_w, inner_k_tiles)
+    assert torch.equal(t, unpacked)
+
+# TODO: Fix "test_aot_dispatch_dynamic" test failure
+@pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")
+@pytest.mark.parametrize("shape, inner_k_tiles", TEST_CONFIGS_UNPACK , ids=str)
+def test_unpack_tensor_core_tiled_layout_op(shape, inner_k_tiles):
+    test_utils = [
+        "test_schema",
+        "test_autograd_registration",
+        "test_faketensor",
+    ]
+
+    # TODO: Figure out why test fails unless torch >= 2.5
+    if TORCH_VERSION_AFTER_2_5:
+        test_utils.append("test_aot_dispatch_dynamic")
+
+    t = torch.randint(0, 16, dtype=torch.int, size=shape, device="cuda")
+    packed_w = torch.ops.aten._convert_weight_to_int4pack(t, inner_k_tiles)
+
+    opcheck(
+        torch.ops.torchao.unpack_tensor_core_tiled_layout,
+        (packed_w, inner_k_tiles),
+        test_utils=test_utils,
+    )
+
+def dequant_ref(q, scales, zeros, group_size, nbits=4, dtype=torch.bfloat16):
+    n, k = q.shape
+    assert q.dtype == torch.int
+
+    n_groups = k // group_size
+    assert scales.shape[0] == n and scales.shape[1] == n_groups
+    assert scales.shape == zeros.shape
+
+    midpoint = 2 ** (nbits - 1)
+
+    #Convert fron u4 -> s4 and upcast to bfloat16
+    q = q.sub(midpoint).to(dtype)
+
+    # Dequantize
+    q = q.reshape(-1, group_size)
+    dq = q * scales.reshape(-1, 1) + zeros.reshape(-1, 1)
+
+    return dq.reshape(n, k)
+
+
+@pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")
+@pytest.mark.parametrize("shape, inner_k_tiles, group_size", TEST_CONFIGS_DEQUANT, ids=str)
+def test_dequantize_tensor_core_tiled_layout_correctness_quant_dequant(shape, inner_k_tiles, group_size):
+    n, k = shape
+    dtype = torch.bfloat16    
+
+    device = "cuda"
+
+    t = torch.randn(n, k, dtype=dtype, device=device)
+    scales, zeros = get_groupwise_affine_qparams(t, n_bit=4, groupsize=group_size, dtype=dtype)
+
+    # Quantize
+    q = groupwise_affine_quantize_tensor_from_qparams(
+        t, scales, zeros, n_bit=4, groupsize=group_size
+    )
+
+    # Pack to tensor core layout
+    packed = torch.ops.aten._convert_weight_to_int4pack(q, inner_k_tiles)
+    scales_and_zeros = pack_tinygemm_scales_and_zeros(scales, zeros)
+    q_groups = k // group_size
+    assert scales_and_zeros.shape == torch.Size([q_groups, n, 2])
+
+    # Dequantize 'ao' ref
+    dq_ao = groupwise_affine_dequantize_tensor_from_qparams(
+        q, scales, zeros, n_bit=4, groupsize=group_size
+    )
+
+    # Dequantize by passing in an identity matrix as the activation
+    a_eye = torch.eye(k, device=device, dtype=dtype)
+    dq_id = torch.ops.aten._weight_int4pack_mm(
+        a_eye,
+        packed,
+        group_size,
+        scales_and_zeros,
+    ).t()
+
+    # Actual operation to test
+    dq_op = torchao.ops.dequantize_tensor_core_tiled_layout(packed, scales_and_zeros, group_size, inner_k_tiles)
+
+    # Compare results
+    diff_ao_id = (dq_id - dq_ao).abs().max()
+    diff_op_id = (dq_op - dq_id).abs().max()
+    diff_op_ao = (dq_op - dq_ao).abs().max()
+
+    # There are slight numerical differences when dequantizing with an identity matrix when compared to `groupwise_affine_dequantize`
+    # Since the `dequantize_tensor_core_layout` kernel relies on the same underlying bit twiddling tricks for fast
+    # conversion from u4 -> s4 -> bf16, the identity matrix dequant hack and `dequantize_tensor_core_layout` are
+    # expected to give same results, while both will have similar numerical differences to `groupwise_affine_dequantize`.
+
+    # Test that the `dequant` kernel gives same results as identity matrix-based dequant 
+    assert diff_op_id == 0
+
+    # Test that the `dequant` kernel gives same numerical diffs as the `groupwise_affine_dequantize` when compared against the identity matrix
+    assert diff_op_ao == diff_ao_id
+
+    assert diff_op_ao < 1e-1
+
+# This test differs from one above in that it uses `unpack_tensor_core_tiled_layout` to unpack then dequantize
+@pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")
+@pytest.mark.parametrize("shape, inner_k_tiles, group_size", TEST_CONFIGS_DEQUANT, ids=str)
+def test_dequantize_tensor_core_tiled_layout_correctness_unpack_and_dequant(shape, inner_k_tiles, group_size):
+    n, k = shape
+    dtype = torch.bfloat16    
+    device = "cuda"
+
+    # Quantize and pack
+    t = torch.randn(n, k, dtype=dtype, device=device)
+    scales, zeros = get_groupwise_affine_qparams(t, n_bit=4, groupsize=group_size, dtype=dtype)
+    q = groupwise_affine_quantize_tensor_from_qparams(
+        t, scales, zeros, n_bit=4, groupsize=group_size
+    )
+
+    packed = torch.ops.aten._convert_weight_to_int4pack(q, inner_k_tiles)
+    scales_and_zeros = pack_tinygemm_scales_and_zeros(scales, zeros)
+
+    # Unpack and dequantize
+    unpacked = torchao.ops.unpack_tensor_core_tiled_layout(packed, inner_k_tiles)
+    dq_ao = groupwise_affine_dequantize_tensor_from_qparams(
+        unpacked, scales, zeros, n_bit=4, groupsize=group_size
+    )
+
+    # Dequantize by passing in an identity matrix as the activation
+    a_eye = torch.eye(k, device=device, dtype=dtype)
+    dq_id = torch.ops.aten._weight_int4pack_mm(
+        a_eye,
+        packed,
+        group_size,
+        scales_and_zeros,
+    ).t()
+
+    # Actual operation to test
+    dq_op = torchao.ops.dequantize_tensor_core_tiled_layout(packed, scales_and_zeros, group_size, inner_k_tiles)
+
+    # Compare results
+    diff_ao_id = (dq_id - dq_ao).abs().max()
+    diff_op_id = (dq_op - dq_id).abs().max()
+    diff_op_ao = (dq_op - dq_ao).abs().max()
+
+    # There are slight numerical differences when dequantizing with an identity matrix when compared to `groupwise_affine_dequantize`
+    # Since the `dequantize_tensor_core_layout` kernel relies on the same underlying bit twiddling tricks for fast
+    # conversion from u4 -> s4 -> bf16, the identity matrix dequant hack and `dequantize_tensor_core_layout` are
+    # expected to give same results, while both will have similar numerical differences to `groupwise_affine_dequantize`.
+
+    # Test that the `dequant` kernel gives same results as identity matrix-based dequant 
+    assert diff_op_id == 0
+
+    # Test that the `dequant` kernel gives same numerical diffs as the `groupwise_affine_dequantize` when compared against the identity matrix
+    assert diff_op_ao == diff_ao_id
+
+    assert diff_op_ao < 1e-1
+
+@pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")
+@pytest.mark.parametrize("shape, inner_k_tiles, group_size", TEST_CONFIGS_DEQUANT, ids=str)
+def test_dequantize_tensor_core_tiled_layout_op(shape, inner_k_tiles, group_size):
+    n, k = shape
+    device = "cuda"
+
+    q = torch.randint(0, 16, shape, dtype=torch.int, device=device)
+    packed_w = torch._convert_weight_to_int4pack(q, inner_k_tiles)
+    q_groups = k // group_size
+    scales = torch.randn(n, q_groups, dtype=torch.bfloat16, device=device)
+    zeros = torch.randn_like(scales)
+    scales_and_zeros = pack_tinygemm_scales_and_zeros(scales, zeros)
+
+    test_utils = [
+    "test_schema",
+    "test_autograd_registration",
+    "test_faketensor",
+    ]
+    # TODO: Figure out why test fails unless torch >= 2.5
+    if TORCH_VERSION_AFTER_2_5:
+        test_utils.append("test_aot_dispatch_dynamic")
+    opcheck(
+        torch.ops.torchao.dequantize_tensor_core_tiled_layout,
+        (packed_w, scales_and_zeros, group_size, inner_k_tiles),
+        test_utils=test_utils,
+    )
+
 if __name__ == "__main__":
-    run_tests()
+    run_tests()