xehpg: jit: gemm: reduce register pressure in 32x24 FHS TN kernel

oneapi-src · Nov 14, 2024 · 08e15c6 · 08e15c6
1 parent ae9018f
commit 08e15c6
Showing 1 changed file with 1 addition 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
@@ -306,7 +306,7 @@ auto _CATALOG_ = kcatalog::toFlatCatalog({
 {{'E', "gemm", {"F", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, 2, -1}, {-1, 8, -1}, {-1, 2, -1}, {-1, 8, -1}, {4, 4, 1}, "xyI"}, "sB64 sS16 aS wg 2x1x16 ikr af vav sr bk0 bm0 sys pab grf256", {8, (LoopType) 0, 256, {(LoopType) 0, (LoopType) 1, (LoopType) 2}, {16777216, 8192, 16777216}, {8192, 8192, 16777216}, {16, 8, 64}, {2, 1, 16}, 1, (WGType) 0, 4357, 0, 1024, {4, 4, 4}, {false, false, true}}, {'W', 1, {128}}},
 {{'E', "gemm", {"F", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, 1, -1}, {-1, 1, -1}, {-1, 1, -1}, {-1, 1, -1}, {4, 4, 1}, "hxyI"}, "sB64 sS16 aS wg 2x1x16 ikr af vav sr bk0 bm0 sys pab grf256", {8, (LoopType) 0, 256, {(LoopType) 0, (LoopType) 1, (LoopType) 2}, {16777216, 8192, 16777216}, {8192, 8192, 16777216}, {16, 8, 64}, {2, 1, 16}, 1, (WGType) 0, 4357, 0, 1024, {4, 4, 4}, {false, false, true}}, {'W', 1, {128}}},
 {{'E', "gemm", {"F", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, -1, -1}, {-1, -1, -1}, {1, 1, 1}, "ixy"}, "sB4 sB4 aB wg 4x8 kc4 cab4 ks8 nse bo sr bk0 sm sn l4 pab", {8, (LoopType) 0, 128, {(LoopType) 128, (LoopType) 255, (LoopType) 255}, {524288, 262144, 16777216}, {524288, 262144, 16777216}, {32, 16, 8}, {4, 8, 1}, 1, (WGType) 1, 257, 32768, 0, {1, 2, 4}, {false, false, true}}, {'W', 1, {512}}},
-{{'E', "gemm", {"F", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, 33, -1}, {-1, -1, -1}, {4, 4, 1}, "xyIs"}, "sB32 sB32 aB wg 4x8 cab3x2 ks32 xaf st dw vav bo sr bk0 dm grf256 sys pab", {8, (LoopType) 0, 256, {(LoopType) 128, (LoopType) 255, (LoopType) 255}, {524288, 393216, 16777216}, {524288, 393216, 16777216}, {32, 24, 32}, {4, 8, 1}, 1, (WGType) 1, 257, 61440, 0, {4, 4, 4}, {false, false, true}}, {'E', 17, {1.01302e+06, 570829, 0, 0, 0, 0, 3.67307, 6.66635, 6.86396, 18.2302, 0.0202076, 0.0155595, 0.00597746, 1, 1.56109, 1.12816, 4.46535e-12}}},
+{{'E', "gemm", {"F", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, 33, -1}, {-1, -1, -1}, {4, 4, 1}, "xyIs"}, "sB16 sB32 aB wg 4x8 cab3x2 ks32 xaf st dw vav bo sr bk0 dm grf256 sys pab", {8, (LoopType) 0, 256, {(LoopType) 128, (LoopType) 255, (LoopType) 255}, {524288, 393216, 16777216}, {524288, 393216, 16777216}, {32, 24, 32}, {4, 8, 1}, 1, (WGType) 1, 257, 61440, 0, {4, 4, 4}, {false, false, true}}, {'E', 17, {1.01302e+06, 570829, 0, 0, 0, 0, 3.67307, 6.66635, 6.86396, 18.2302, 0.0202076, 0.0155595, 0.00597746, 1, 1.56109, 1.12816, 4.46535e-12}}},
 {{'E', "gemm", {"F", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, 33, -1}, {-1, -1, -1}, {4, 4, 1}, "xyI"}, "sB32 sB16 aB wg 8x4 cab4x2 ks32 xaf st dw vav bo sr bk0 dm grf256 sys pab", {8, (LoopType) 0, 256, {(LoopType) 128, (LoopType) 255, (LoopType) 255}, {262144, 524288, 16777216}, {262144, 524288, 16777216}, {16, 32, 32}, {8, 4, 1}, 1, (WGType) 1, 257, 65536, 0, {4, 4, 4}, {false, false, true}}, {'E', 17, {970381, 499059, 0, 0, 0, 0, 2.81407, 6.26965, 6.95402, 17.4255, 0.0209222, 0.0155763, 0.00737804, 1, 1.69814, 1.09799, 5.06192e-12}}},
 {{'E', "gemm", {"F", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, -1, -1}, {-1, -1, -1}, {-1, 33, -1}, {-1, -1, -1}, {4, 4, 1}, "xyI"}, "sB16 sB16 aB wg 8x4 cab4x2 ks16 xaf dw vav bo sr bk0 dm grf256 sys pab", {8, (LoopType) 0, 256, {(LoopType) 128, (LoopType) 255, (LoopType) 255}, {524288, 524288, 16777216}, {524288, 524288, 16777216}, {32, 32, 16}, {8, 4, 1}, 1, (WGType) 1, 257, 49152, 0, {4, 4, 4}, {false, false, true}}, {'E', 17, {979611, 558578, 0, 0, 0, 0, 2.44965, 6.25941, 6.99298, 18.8195, 0.0200391, 0.00951994, 0.00909253, 0.933181, 1.77877, 1.1097, 6.88098e-12}}},
 {{'E', "gemm", {"F", "H", "S"}, {"T", "N", "N"}}, {-1, -1, {-1, 1, -1}, {512, 1, -1}, {-1, 1, -1}, {512, 1, -1}, {4, 4, 1}, "H"}, "aB128x2 aB128x2 aB wg 2x1x8 ikr ar sb128 bk0 dot", {8, (LoopType) 0, 128, {(LoopType) 0, (LoopType) 1, (LoopType) 2}, {8192, 8192, 16777216}, {8192, 8192, 16777216}, {1, 1, 128}, {2, 1, 8}, 1, (WGType) 0, 4357, 0, 64, {4, 4, 4}, {true, true, true}}, {'W', 1, {1}}},