Andriy Roshchenko 7106976a72 MX GEMM - New GEMM pipeline for MX data types (#2059) ...

* Allow selection of mfma_scale instructions

* Read B tensor from LDS to VGPR in chunks of 16 in MFMA order

* Add constexpr and synchronize return type for `get_exponent_value`

* Pass scales by reference and add comments to `mfma_scale_f32_32x32x64`

* Add support for microscaling instructions in `XdlopsGemm`

* Fix `mfma_scale_f32_16x16x128f8f6f4` wrapper

* Remove software implementation of MX GEMM

* Make interface of `intrin_mfma_scale_f32_16x16x128f8f6f4<16, 16>` consistent with the other scale instruction

* Update README

* Updated CHANGELOG

* Remove unused static methods

2025-04-15 17:17:07 -06:00

..

amd_address_space.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

amd_buffer_addressing_builtins.hpp

Replace buffer load/store intrinsics with builtins (#1876)

2025-03-05 14:33:28 -08:00

amd_buffer_addressing.hpp

Rebase the PR #1520 to ROCm repo. (#1574)

2025-02-20 18:58:14 -08:00

amd_ck_fp8.hpp

fix codegen issues (#2052)

2025-04-07 07:08:39 -07:00

amd_gemm_dpp.hpp

Redesign the DPP8 GEMM kernel to use warp-wise component (#863)

2023-09-06 11:44:09 -05:00

amd_inline_asm.hpp

Split up data_type header. (#1996)

2025-03-24 15:08:54 -07:00

amd_lds.hpp

Implement direct loads split-K GEMM kernel (#1137)

2024-02-07 01:08:34 +01:00

amd_smfmac.hpp

Added structural sparsity blockwise gemm (#1435)

2024-09-11 15:19:42 +02:00

amd_wave_read_first_lane.hpp

Rebase the PR #1520 to ROCm repo. (#1574)

2025-02-20 18:58:14 -08:00

amd_wmma.hpp

Add bf16 and int8 wmma gemms for Navi3x and Navi4x. (#1671)

2024-11-18 14:07:04 -08:00

amd_xdlops.hpp

MX GEMM - New GEMM pipeline for MX data types (#2059)

2025-04-15 17:17:07 -06:00

array_multi_index.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

array.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

blkgemmpipe_scheduler.hpp

Fix on the error (#1956)

2025-03-07 13:43:52 -08:00

c_style_pointer_cast.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

common_header.hpp

use old instrinsics with staging compiler (#1970)

2025-03-12 07:29:09 -07:00

container_element_picker.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

container_helper.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

data_type.hpp

Add MX FP4 device conversion tests (#1889)

2025-03-26 19:23:01 -05:00

debug.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

dtype_fp64.hpp

Split up data_type header. (#1996)

2025-03-24 15:08:54 -07:00

dtype_vector.hpp

Fix Windows build. (#2012)

2025-04-01 12:22:10 -07:00

dynamic_buffer.hpp

use old instrinsics with staging compiler (#1970)

2025-03-12 07:29:09 -07:00

e8m0.hpp

MX GEMM - New GEMM pipeline for MX data types (#2059)

2025-04-15 17:17:07 -06:00

enable_if.hpp

Rebase the PR #1520 to ROCm repo. (#1574)

2025-02-20 18:58:14 -08:00

env.hpp

Split env.hpp header from the ck.hpp header. (#2049)

2025-04-03 15:30:21 -07:00

f8_utils.hpp

Split up data_type header. (#1996)

2025-03-24 15:08:54 -07:00

flush_icache.hpp

Universal gemm flush cache (#1251)

2024-04-25 15:07:14 -05:00

functional2.hpp

add a fast compilation path for static for (0..N) (#2005)

2025-04-01 12:06:25 -07:00

functional3.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

functional4.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

functional.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

generic_memory_space_atomic.hpp

Split up data_type header. (#1996)

2025-03-24 15:08:54 -07:00

get_id.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

get_shift.hpp

Batchnorm splitk single kernel (#771)

2023-07-06 10:58:55 -05:00

ignore.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

inner_product_dpp8.hpp

Redesign the DPP8 GEMM kernel to use warp-wise component (#863)

2023-09-06 11:44:09 -05:00

inner_product.hpp

Fix the DL kernel issues on Navi3x. (#998)

2023-10-19 09:34:39 -07:00

integral_constant.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

is_detected.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

is_known_at_compile_time.hpp

Add optimized copy to ck wrapper (#1126)

2024-01-19 11:29:00 +01:00

loop_scheduler.hpp

Rebase the PR #1520 to ROCm repo. (#1574)

2025-02-20 18:58:14 -08:00

magic_division.hpp

Split up data_type header. (#1996)

2025-03-24 15:08:54 -07:00

math_v2.hpp

Rebase the PR #1520 to ROCm repo. (#1574)

2025-02-20 18:58:14 -08:00

math.hpp

Extend available elementwise operations with conv examples (#995)

2023-10-19 17:23:19 +02:00

multi_index.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

mxf4_utils.hpp

Add MX FP4 device conversion tests (#1889)

2025-03-26 19:23:01 -05:00

mxf6_utils.hpp

Split up data_type header. (#1996)

2025-03-24 15:08:54 -07:00

mxf8_utils.hpp

Add FP16/BF16<->FP8/BF8 conversions (#2035)

2025-04-03 12:42:03 -05:00

mxfp_utils.hpp

MX GEMM - New GEMM pipeline for MX data types (#2059)

2025-04-15 17:17:07 -06:00

number.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

numeric_limits.hpp

Split up data_type header. (#1996)

2025-03-24 15:08:54 -07:00

numeric_utils.hpp

Split up data_type header. (#1996)

2025-03-24 15:08:54 -07:00

random_gen.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

reduction_common.hpp

Batchnorm splitk single kernel (#771)

2023-07-06 10:58:55 -05:00

reduction_enums.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

reduction_functions_accumulate.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

reduction_operator.hpp

Remove unsupported (fp8) type from Add memory operation. (#1521)

2024-09-20 09:40:45 +02:00

scaled_type_convert.hpp

Add FP16/BF16<->FP8/BF8 conversions (#2035)

2025-04-03 12:42:03 -05:00

sequence_helper.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

sequence.hpp

simplify generate_tuple (#2043)

2025-04-08 09:00:51 -07:00

span.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

static_buffer.hpp

Jing's contribution: prototype of mixed precision gemm FP16/BF16xint4 GEMM (#1762)

2025-01-02 11:48:06 +08:00

statically_indexed_array_multi_index.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

statically_indexed_array.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

synchronization.hpp

Merging the gfx12 code into public repo. (#1362)

2024-06-27 00:33:34 -07:00

thread_group.hpp

update copyright headers (#726)

2023-05-31 18:46:57 -05:00

transpose_vectors.hpp

[GEMM] Gemm universal device operation (#1154)

2024-04-13 21:03:18 -05:00

tuple_helper.hpp

simplify generate_tuple (#2043)

2025-04-08 09:00:51 -07:00

tuple.hpp

Codegen hipRTC compilation (#1579)

2025-01-31 09:48:39 -08:00

type_convert.hpp

Add FP16/BF16<->FP8/BF8 conversions (#2035)

2025-04-03 12:42:03 -05:00

type.hpp

Rebase the PR #1520 to ROCm repo. (#1574)

2025-02-20 18:58:14 -08:00

workgroup_barrier.hpp

initial stream-k implementation with example (#699)

2023-07-26 14:18:15 -05:00

workgroup_synchronization.hpp

Batchnorm splitk single kernel (#771)

2023-07-06 10:58:55 -05:00