You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2022/08/31 10:06:48 UTC
[GitHub] [tvm] ibsidorenko opened a new pull request, #12659: [Hexagon] Implement fixed_point_multiply op through intrinsics.
ibsidorenko opened a new pull request, #12659:
URL: https://github.com/apache/tvm/pull/12659
This commit adds high-performance implementation of `fixed_point_multiply`
operation based on Hexagon intrinsics for vmpye/vmpyo instructions.
Benchmarking of `fixed_point_multiply` op with (1,8,56,56,32) input
tensor on Qualcomm SM8350:
* default implementation: **10.06 ms**
* optimized implementation: **1.42 ms**
* speedup: **7x** times (!!!)
Please note that this is introducing a small round-up error for some
corner cases with negative shift argument (The same as for ARM CPU, see
[PR#5980](https://github.com/apache/tvm/pull/5980)). This is because we are rounding twice instead than only once:
* original q_multiply_shift: round(x*y*2^-s)
* hexagon q_multiply_shift: round(round(x*y)*2^-s)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] ibsidorenko commented on a diff in pull request #12659: [Hexagon] Implement fixed_point_multiply op through intrinsics.
Posted by GitBox <gi...@apache.org>.
ibsidorenko commented on code in PR #12659:
URL: https://github.com/apache/tvm/pull/12659#discussion_r960256617
##########
python/tvm/topi/hexagon/tensor_intrin.py:
##########
@@ -0,0 +1,71 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+"""Optimized implementation of q_multiply_shift based on LLVM intrinsics"""
+
+import tvm
+from tvm.ir import register_intrin_lowering
+
+
+def _q_multiply_shift_hexagon(op):
+ """
+ Implementation of q_multiply_shift through hexagon intrinsics vmpyewuh and vmpyowh when q == 31.
+
+ Please note that this is introducing a small round-up error for some corner cases with negative
+ shift argument. This is because we are rounding twice instead than only once. I.e.:
+
+ * original q_multiply_shift: round(x*y*2^-s)
+ * hexagon q_multiply_shift: round(round(x*y)*2^-s)
+ """
+ x = op.args[0]
+ y = op.args[1]
+ fractional_bits = op.args[2]
+ shift = op.args[3]
+
+ # Don't use this intrinsic if we don't have a int32x32 vector
+ # or if we are not multiplying q31 numbers
+ if x.dtype != "int32x32" or fractional_bits.value != 31:
+ return op
+
+ # Case 1, shift is negative
+ mul_e_1 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyewuh.128B", tvm.tir.const(2, "uint32"), x, y
+ )
+ mul_o_1 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyowh.rnd.sacc.128B", tvm.tir.const(3, "uint32"), mul_e_1, x, y
+ )
+ fixup = mul_o_1 & (-shift)
+ round_mul = mul_o_1 + fixup
+ out_negative_shift = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vaslwv.128B", tvm.tir.const(2, "uint32"), round_mul, shift
+ )
+
+ # Case 2, shift is positive
+ x = x * (1 << (shift))
+ mul_e_2 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyewuh.128B", tvm.tir.const(2, "uint32"), x, y
+ )
+ mul_o_2 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyowh.rnd.sacc.128B", tvm.tir.const(3, "uint32"), mul_e_2, x, y
+ )
+
+ # Select depending on the shift
+ return tvm.tir.Select(shift < 0, out_negative_shift, mul_o_2)
Review Comment:
hm... Maybe I misunderstood the question, but I put shift separately before mul_o_2:
`x = x * (1 << (shift))`
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] tmoreau89 commented on pull request #12659: [Hexagon] Implement fixed_point_multiply op through intrinsics.
Posted by GitBox <gi...@apache.org>.
tmoreau89 commented on PR #12659:
URL: https://github.com/apache/tvm/pull/12659#issuecomment-1233289408
Excellent, thank you for the contribution @ibsidorenko !
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] kparzysz-quic merged pull request #12659: [Hexagon] Implement fixed_point_multiply op through intrinsics.
Posted by GitBox <gi...@apache.org>.
kparzysz-quic merged PR #12659:
URL: https://github.com/apache/tvm/pull/12659
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] kparzysz-quic commented on a diff in pull request #12659: [Hexagon] Implement fixed_point_multiply op through intrinsics.
Posted by GitBox <gi...@apache.org>.
kparzysz-quic commented on code in PR #12659:
URL: https://github.com/apache/tvm/pull/12659#discussion_r960653521
##########
python/tvm/topi/hexagon/tensor_intrin.py:
##########
@@ -0,0 +1,71 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+"""Optimized implementation of q_multiply_shift based on LLVM intrinsics"""
+
+import tvm
+from tvm.ir import register_intrin_lowering
+
+
+def _q_multiply_shift_hexagon(op):
+ """
+ Implementation of q_multiply_shift through hexagon intrinsics vmpyewuh and vmpyowh when q == 31.
+
+ Please note that this is introducing a small round-up error for some corner cases with negative
+ shift argument. This is because we are rounding twice instead than only once. I.e.:
+
+ * original q_multiply_shift: round(x*y*2^-s)
+ * hexagon q_multiply_shift: round(round(x*y)*2^-s)
+ """
+ x = op.args[0]
+ y = op.args[1]
+ fractional_bits = op.args[2]
+ shift = op.args[3]
+
+ # Don't use this intrinsic if we don't have a int32x32 vector
+ # or if we are not multiplying q31 numbers
+ if x.dtype != "int32x32" or fractional_bits.value != 31:
+ return op
+
+ # Case 1, shift is negative
+ mul_e_1 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyewuh.128B", tvm.tir.const(2, "uint32"), x, y
+ )
+ mul_o_1 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyowh.rnd.sacc.128B", tvm.tir.const(3, "uint32"), mul_e_1, x, y
+ )
+ fixup = mul_o_1 & (-shift)
+ round_mul = mul_o_1 + fixup
+ out_negative_shift = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vaslwv.128B", tvm.tir.const(2, "uint32"), round_mul, shift
+ )
+
+ # Case 2, shift is positive
+ x = x * (1 << (shift))
+ mul_e_2 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyewuh.128B", tvm.tir.const(2, "uint32"), x, y
+ )
+ mul_o_2 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyowh.rnd.sacc.128B", tvm.tir.const(3, "uint32"), mul_e_2, x, y
+ )
+
+ # Select depending on the shift
+ return tvm.tir.Select(shift < 0, out_negative_shift, mul_o_2)
Review Comment:
Sorry, I misread it as a part of the comment...
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] masahi commented on pull request #12659: [Hexagon] Implement fixed_point_multiply op through intrinsics.
Posted by GitBox <gi...@apache.org>.
masahi commented on PR #12659:
URL: https://github.com/apache/tvm/pull/12659#issuecomment-1232772601
cc @kparzysz-quic
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] kparzysz-quic commented on a diff in pull request #12659: [Hexagon] Implement fixed_point_multiply op through intrinsics.
Posted by GitBox <gi...@apache.org>.
kparzysz-quic commented on code in PR #12659:
URL: https://github.com/apache/tvm/pull/12659#discussion_r959988311
##########
python/tvm/topi/hexagon/tensor_intrin.py:
##########
@@ -0,0 +1,71 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+"""Optimized implementation of q_multiply_shift based on LLVM intrinsics"""
+
+import tvm
+from tvm.ir import register_intrin_lowering
+
+
+def _q_multiply_shift_hexagon(op):
+ """
+ Implementation of q_multiply_shift through hexagon intrinsics vmpyewuh and vmpyowh when q == 31.
+
+ Please note that this is introducing a small round-up error for some corner cases with negative
+ shift argument. This is because we are rounding twice instead than only once. I.e.:
+
+ * original q_multiply_shift: round(x*y*2^-s)
+ * hexagon q_multiply_shift: round(round(x*y)*2^-s)
+ """
+ x = op.args[0]
+ y = op.args[1]
+ fractional_bits = op.args[2]
+ shift = op.args[3]
+
+ # Don't use this intrinsic if we don't have a int32x32 vector
+ # or if we are not multiplying q31 numbers
+ if x.dtype != "int32x32" or fractional_bits.value != 31:
+ return op
+
+ # Case 1, shift is negative
+ mul_e_1 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyewuh.128B", tvm.tir.const(2, "uint32"), x, y
+ )
+ mul_o_1 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyowh.rnd.sacc.128B", tvm.tir.const(3, "uint32"), mul_e_1, x, y
+ )
+ fixup = mul_o_1 & (-shift)
+ round_mul = mul_o_1 + fixup
+ out_negative_shift = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vaslwv.128B", tvm.tir.const(2, "uint32"), round_mul, shift
+ )
+
+ # Case 2, shift is positive
+ x = x * (1 << (shift))
+ mul_e_2 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyewuh.128B", tvm.tir.const(2, "uint32"), x, y
+ )
+ mul_o_2 = tvm.tir.call_llvm_intrin(
+ op.dtype, "llvm.hexagon.V6.vmpyowh.rnd.sacc.128B", tvm.tir.const(3, "uint32"), mul_e_2, x, y
+ )
+
+ # Select depending on the shift
+ return tvm.tir.Select(shift < 0, out_negative_shift, mul_o_2)
Review Comment:
The `mul_o_2` is just `round(x*y)`. There is no shift in it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org