Skip to content

Commit 08e06d3

Browse files
authored
[SYCL][DOC] Add BMG31 to the matrix combinations (#20554)
1 parent b288f5d commit 08e06d3

File tree

1 file changed

+34
-33
lines changed

1 file changed

+34
-33
lines changed

sycl/doc/extensions/experimental/sycl_ext_matrix/sycl_ext_oneapi_matrix.asciidoc

Lines changed: 34 additions & 33 deletions
Original file line numberDiff line numberDiff line change
@@ -1107,51 +1107,51 @@ is shown in a single column in the table below.
11071107
==== Intel XMX Supported Combinations
11081108
This is currently available in devices with the architecture
11091109
`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1110-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_dg2_g10`,
1111-
`architecture::intel_gpu_dg2_g11`, `architecture::intel_gpu_dg2_g12`,
1112-
`architecture::intel_gpu_arl_h`, `architecture::intel_gpu_ptl_h`, and
1113-
`architecture::intel_gpu_ptl_u`.
1110+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1111+
`architecture::intel_gpu_dg2_g10`, `architecture::intel_gpu_dg2_g11`,
1112+
`architecture::intel_gpu_dg2_g12`, `architecture::intel_gpu_arl_h`,
1113+
`architecture::intel_gpu_ptl_h`, and `architecture::intel_gpu_ptl_u`.
11141114

11151115
[frame="none",options="header"]
11161116
|======================
11171117
| A type | B type | C type | D type | M | N | K | device
11181118
.2+| `matrix_type::uint8` .2+| `matrix_type::uint8` .2+|
11191119
`matrix_type::sint32` .2+| `matrix_type::sint32` .2+| +<=+ 8 | 16 .2+| 32
11201120
|`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1121-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1122-
`architecture::intel_gpu_ptl_u`
1121+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1122+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11231123
|8|`architecture::intel_gpu_dg2_g10,
11241124
architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`,
11251125
`architecture::intel_gpu_arl_h`
11261126
.2+| `matrix_type::uint8` .2+| `matrix_type::sint8` .2+|
11271127
`matrix_type::sint32` .2+|`matrix_type::sint32` .2+| +<=+ 8 | 16 .2+| 32 |
11281128
`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1129-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1130-
`architecture::intel_gpu_ptl_u`
1129+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1130+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11311131
|8|`architecture::intel_gpu_dg2_g10,
11321132
architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`,
11331133
`architecture::intel_gpu_arl_h`
11341134
.2+| `matrix_type::sint8` .2+| `matrix_type::uint8` .2+|
11351135
`matrix_type::sint32` .2+|`matrix_type::sint32` .2+| +<=+ 8 | 16 .2+| 32 |
11361136
`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1137-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1138-
`architecture::intel_gpu_ptl_u`
1137+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1138+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11391139
|8|`architecture::intel_gpu_dg2_g10,
11401140
architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`,
11411141
`architecture::intel_gpu_arl_h`
11421142
.2+| `matrix_type::sint8` .2+| `matrix_type::sint8` .2+|
11431143
`matrix_type::sint32` .2+| `matrix_type::sint32` .2+| +<=+ 8 | 16 .2+| 32 |
11441144
`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1145-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1146-
`architecture::intel_gpu_ptl_u`
1145+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1146+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11471147
|8|`architecture::intel_gpu_dg2_g10,
11481148
architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`,
11491149
`architecture::intel_gpu_arl_h`
11501150
.8+|`matrix_type::fp16` .8+| `matrix_type::fp16` .8+|
11511151
`matrix_type::fp32` .8+|`matrix_type::fp32` .1+| 16 .1+| 16 | 16
11521152
.6+|`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1153-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1154-
`architecture::intel_gpu_ptl_u`
1153+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1154+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11551155
.2+| 1 .2+| 64 | 16 |32
11561156
.2+| 32 .2+| 64 | 16 |32
11571157
.2+| +<=+ 8 | 16 .2+| 16
@@ -1162,28 +1162,28 @@ architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`,
11621162
.6+|`matrix_type::fp16` .6+| `matrix_type::fp16` .6+|
11631163
`matrix_type::fp16` .6+|`matrix_type::fp32` .1+| +<=+ 8 | 16 .1+| 16
11641164
.6+| `architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1165-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1166-
`architecture::intel_gpu_ptl_u`
1165+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1166+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11671167
| 16 | 16 | 16 .2+| 1 .2+| 64 | 16 | 32
11681168
.2+| 32 .2+| 64 | 16 | 32
11691169
.6+|`matrix_type::fp16` .6+| `matrix_type::fp16` .6+|
11701170
`matrix_type::fp32` .6+|`matrix_type::fp16` .1+| +<=+ 8 | 16 .1+| 16
11711171
.6+|`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1172-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1173-
`architecture::intel_gpu_ptl_u`
1172+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1173+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11741174
| 16 | 16 | 16 .2+| 1 .2+| 64 | 16 | 32
11751175
.2+| 32 .2+| 64 |16 | 32
11761176
.6+|`matrix_type::fp16` .6+| `matrix_type::fp16` .6+|
11771177
`matrix_type::fp16` .6+|`matrix_type::fp16` .1+| +<=+ 8 | 16 .1+| 16
11781178
.6+|`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1179-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1180-
`architecture::intel_gpu_ptl_u`
1179+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1180+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11811181
| 16 | 16 | 16 .2+| 1 .2+| 64 | 16 |32 .2+| 32 .2+| 64 | 16 | 32
11821182
.8+| `matrix_type::bf16` .8+| `matrix_type::bf16` .8+|
11831183
`matrix_type::fp32` .8+| `matrix_type::fp32` | 16 | 16 | 16
11841184
.6+|`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1185-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1186-
`architecture::intel_gpu_ptl_u`
1185+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1186+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11871187
.2+| 1 .2+| 64 | 16 | 32
11881188
.2+| 32 .2+| 64 | 16 |32
11891189
.2+| +<=+ 8 | 16 .2+| 16
@@ -1194,34 +1194,35 @@ architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`,
11941194
.6+|`matrix_type::bf16` .6+| `matrix_type::bf16` .6+|
11951195
`matrix_type::bf16` .6+|`matrix_type::fp32` .1+| +<=+ 8 | 16 .1+| 16 .6+|
11961196
`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1197-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1198-
`architecture::intel_gpu_ptl_u`
1197+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1198+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
11991199
| 16 | 16 | 16 .2+| 1 .2+| 64 | 16 | 32
12001200
.2+| 32 .2+| 64 |16 | 32
12011201
.6+|`matrix_type::bf16` .6+| `matrix_type::bf16` .6+|
12021202
`matrix_type::fp32` .6+|`matrix_type::bf16` .1+| +<=+ 8 | 16 .1+| 16 .6+|
12031203
`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1204-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1205-
`architecture::intel_gpu_ptl_u`
1204+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1205+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
12061206
| 16 | 16 | 16 .2+| 1 .2+| 64 | 16 | 32
12071207
.2+| 32 .2+| 64 |16 | 32
12081208
.6+|`matrix_type::bf16` .6+| `matrix_type::bf16` .6+|
12091209
`matrix_type::bf16` .6+|`matrix_type::bf16` .1+| +<=+ 8 | 16 .1+| 16 .6+|
12101210
`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1211-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1212-
`architecture::intel_gpu_ptl_u`
1211+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1212+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
12131213
| 16 | 16 | 16 .2+| 1 .2+| 64 | 16 | 32
12141214
.2+| 32 .2+| 64 |16 | 32
12151215
| `matrix_type::tf32` | `matrix_type::tf32` |
12161216
`matrix_type::fp32` .2+| `matrix_type::fp32` | +<=+ 8 | 16 | 8 |
12171217
`architecture::intel_gpu_pvc`, `architecture::intel_gpu_bmg_g21`,
1218-
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1219-
`architecture::intel_gpu_ptl_u`
1218+
`architecture::intel_gpu_bmg_g31`, `architecture::intel_gpu_lnl_m`,
1219+
`architecture::intel_gpu_ptl_h`, `architecture::intel_gpu_ptl_u`
12201220
|======================
12211221

12221222
===== Restrictions on `architecture::intel_gpu_pvc`,
1223-
`architecture::intel_gpu_bmg_g21`, `architecture::intel_gpu_lnl_m`,
1224-
`architecture::intel_gpu_ptl_h`, and `architecture::intel_gpu_ptl_u`
1223+
`architecture::intel_gpu_bmg_g21`, `architecture::intel_gpu_bmg_g31`,
1224+
`architecture::intel_gpu_lnl_m`, `architecture::intel_gpu_ptl_h`,
1225+
and `architecture::intel_gpu_ptl_u`
12251226

12261227
- The `stride` parameter to `joint_matrix_load` and
12271228
`joint_matrix_store` has the following restrictions:
@@ -1363,4 +1364,4 @@ load/store overloads
13631364
|11 |2024-04-29 |Yury Plyakhin | Add 1x64x16 supported combination for
13641365
Intel XMX (intel_gpu_pvc)
13651366
|12 |2024-06-14 |Jack Kirk | Add note on sm version device matching issue.
1366-
|======================
1367+
|======================

0 commit comments

Comments
 (0)