xe: jit: gemm: add dynamic quantization kernel

oneapi-src · Jan 16, 2025 · 283cf37 · 283cf37
1 parent 9e746e7
commit 283cf37
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/src/gpu/intel/jit/gemm/selector/db/kernel.db b/src/gpu/intel/jit/gemm/selector/db/kernel.db
@@ -1064,7 +1064,8 @@ auto _CATALOG_ = kcatalog::toFlatCatalog({
 {{'G', "gemm", {"O", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, -1, -1}, {-1, 8, -1}, {16, 16, 1}, "IAB"}, "at32x2+m64@16 am64x2 aB wg 4x1x4 ikr af vav ar sb64 bm0 bk0 sys nmk np", {16, (LoopType) 255, 128, {(LoopType) 0, (LoopType) 1, (LoopType) 2}, {8192, 8192, 16777216}, {8192, 8192, 16777216}, {16, 8, 64}, {4, 1, 4}, 1, (WGType) 1, 261, 0, 2048, {16, 16, 4}, {true, true, true}}, {'W', 1, {128}}},
 {{'G', "gemm", {"O", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, 9, -1}, {-1, 64, -1}, {16, 16, 1}, "IAB"}, "at32x2+m64@16 am64x2 aB wg 4x1x4 ikr af vav ar sb64 bm0 bk0 sys nmk grf256 np", {16, (LoopType) 255, 128, {(LoopType) 0, (LoopType) 1, (LoopType) 2}, {8192, 8192, 16777216}, {8192, 8192, 16777216}, {16, 16, 64}, {4, 1, 4}, 1, (WGType) 1, 261, 0, 2048, {16, 16, 4}, {true, true, true}}, {'W', 1, {128}}},
 {{'G', "gemm", {"[FO]", "O", "S"}, {"N", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, -1, -1}, {-1, -1, -1}, {16, 16, 1}, "ABIs"}, "am32+m32@112 am32x2+m32@112 aB wg 4x8 ca4 ks32 af st rr vav hi pt sr br bk0 sn nb 4x0 grf256 sys acb cr16", {16, (LoopType) 255, 256, {(LoopType) 208, (LoopType) 255, (LoopType) 255}, {1048576, 458752, 16777216}, {1048576, 458752, 16777216}, {64, 28, 32}, {4, 8, 1}, 1, (WGType) 1, 257, 32768, 0, {16, 16, 4}, {true, true, true}}, {'W', 1, {1e+06}}},
-{{'G', "gemm", {"[FO]", "O", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, -1, -1}, {-1, -1, -1}, {16, 16, 1}, "ABI"}, "at32+m32@64 am32+m64@64 aB wg 4x8 ca3 ks32 xaf st vav hi pt sr br bk0 sm sn nb 4x0 grf256 sys acb cr16", {16, (LoopType) 255, 256, {(LoopType) 208, (LoopType) 255, (LoopType) 255}, {1048576, 524288, 16777216}, {1048576, 524288, 16777216}, {64, 32, 32}, {4, 8, 1}, 1, (WGType) 1, 257, 24576, 0, {16, 16, 4}, {true, true, true}}, {'W', 1, {1e+06}}},
+{{'G', "gemm", {"[FO]", "O", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, 512, -1}, {-1, -1, -1}, {16, 16, 1}, "ABI"}, "at32+m32@64 am32+m64@64 aB wg 4x8 ca3 ks32 xaf st vav hi pt sr br bk0 sm sn nb 4x0 grf256 sys acb cr16", {16, (LoopType) 255, 256, {(LoopType) 208, (LoopType) 255, (LoopType) 255}, {1048576, 524288, 16777216}, {1048576, 524288, 16777216}, {64, 32, 32}, {4, 8, 1}, 1, (WGType) 1, 257, 24576, 0, {16, 16, 4}, {true, true, true}}, {'W', 1, {1e+06}}},
+{{'G', "gemm", {"[FO]", "O", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, -1, -1}, {-1, 511, -1}, {16, 16, 1}, "ABI"}, "at32+m32@32 am32+m64@32 aB wg 4x4 ca3 ks32 xaf st vav hi pt sr br bk0 sm sn nb 4x0 grf256 sys acb cr0 ", {16, (LoopType) 255, 256, {(LoopType) 208, (LoopType) 255, (LoopType) 255}, {1048576, 524288, 16777216}, {1048576, 524288, 16777216}, {32, 32, 128}, {4, 8, 1}, 1, (WGType) 1, 257, 24576, 0, {16, 16, 4}, {true, true, true}}, {'W', 1, {1e+06}}},
 {{'G', "gemm", {"[FO]", "O", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, 8, -1}, {-1, -1, -1}, {-1, 8, -1}, {16, 16, 1}, "ABIpq"}, "at32 am128 aB wg 2x1x8 ikr xaf st acb hi pt ar sb128 bk0 bm0 nmk sys", {16, (LoopType) 255, 128, {(LoopType) 209, (LoopType) 255, (LoopType) 2}, {16777216, 131072, 16777216}, {262144, 131072, 16777216}, {16, 8, 128}, {2, 1, 8}, 1, (WGType) 0, 4357, 0, 1024, {16, 16, 4}, {true, true, true}}, {'W', 1, {128}}},
 {{'G', "gemm", {"[FO]", "O", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, 9, -1}, {-1, 32, -1}, {-1, 9, -1}, {-1, 32, -1}, {16, 16, 1}, "ABIpq"}, "at32+m128@80 am32+m128@80 aB wg 8x1x4 ikr wx2 xaf acb hi pt ar sb128 bk0 sm sn bm0 nmk sys", {16, (LoopType) 255, 128, {(LoopType) 209, (LoopType) 255, (LoopType) 2}, {16777216, 262144, 16777216}, {262144, 262144, 16777216}, {16, 16, 128}, {8, 1, 4}, 2, (WGType) 1, 4357, 0, 8192, {16, 16, 4}, {true, true, true}}, {'W', 1, {256}}},
 {{'G', "gemm", {"[FO]", "O", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, 8, -1}, {-1, -1, -1}, {-1, 8, -1}, {16, 16, 1}, "IAB"}, "at32 am128 aB wg 2x1x8 ikr xaf st acb li ar sb128 bk0 bm0 nmk sys", {16, (LoopType) 255, 128, {(LoopType) 161, (LoopType) 255, (LoopType) 2}, {16777216, 131072, 16777216}, {262144, 131072, 16777216}, {16, 8, 128}, {2, 1, 8}, 1, (WGType) 0, 4357, 0, 1024, {16, 16, 4}, {true, true, true}}, {'W', 1, {128}}},