update torch patches (#4193)

guangyey · web-flow · commit 2cbf5c660aeb · 2024-04-26T11:46:23.000+08:00
diff --git a/torch_patches/0001-Add-API-for-intel-XPU-to-register-new-tensor-type.patch b/torch_patches/0001-Add-API-for-intel-XPU-to-register-new-tensor-type.patch
@@ -1,7 +1,7 @@
 From 6bf65db4efdaee36d9019c61a3b566524f9a87db Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <guangye.yu@intel.com>
 Date: Mon, 22 May 2023 09:56:52 +0000
-Subject: [PATCH 01/29] Add API for intel XPU to register new tensor type
+Subject: [PATCH 01/30] Add API for intel XPU to register new tensor type
 
 (cherry picked from commit 6493115242ca9657826b50b21383cec922c253ae)
 (cherry picked from commit b7ad62ba8da4e34958ba1ed9d31d1785d508aea8)
diff --git a/torch_patches/0002-Add-xpu-legacy-profiler-163.patch b/torch_patches/0002-Add-xpu-legacy-profiler-163.patch
@@ -1,7 +1,7 @@
 From ce8a0310e85e07270ef9f006ef9e974a3c5f0000 Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <106960996+guangyey@users.noreply.github.com>
 Date: Fri, 15 Sep 2023 00:16:02 +0800
-Subject: [PATCH 02/29] Add xpu legacy profiler (#163)
+Subject: [PATCH 02/30] Add xpu legacy profiler (#163)
 
 * Add xpu legacy profiler
 
diff --git a/torch_patches/0003-add-custom-pass-hooks-in-post_grad_passes-164.patch b/torch_patches/0003-add-custom-pass-hooks-in-post_grad_passes-164.patch
@@ -1,7 +1,7 @@
 From e2c83750f5761cd527941e134ba791b21ee58ad6 Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <106960996+guangyey@users.noreply.github.com>
 Date: Fri, 15 Sep 2023 00:39:22 +0800
-Subject: [PATCH 03/29] add custom pass hooks in post_grad_passes (#164)
+Subject: [PATCH 03/30] add custom pass hooks in post_grad_passes (#164)
 
 ---
  test/inductor/test_custom_post_grad_passes.py | 172 ++++++++++++++++++
diff --git a/torch_patches/0004-Add-qweight-prepack-runtime-registration-linear-conv.patch b/torch_patches/0004-Add-qweight-prepack-runtime-registration-linear-conv.patch
@@ -1,7 +1,7 @@
 From e396a88d893335c8c1221d3bde057fe9ca1cbeb4 Mon Sep 17 00:00:00 2001
 From: Zhiwei <532707544@qq.com>
 Date: Wed, 13 Sep 2023 13:31:06 +0800
-Subject: [PATCH 04/29] Add qweight prepack runtime registration(linear, conv)
+Subject: [PATCH 04/30] Add qweight prepack runtime registration(linear, conv)
  & Add QXPU QEngine (#147)
 
 Add qweight prepack runtime registration(linear, conv) & Add QXPU QEngine
diff --git a/torch_patches/0005-fix-hpu-deserialization-bug-165.patch b/torch_patches/0005-fix-hpu-deserialization-bug-165.patch
@@ -1,7 +1,7 @@
 From 414928de5c0e00ea4d21fbef5297ddd90c887cd7 Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <106960996+guangyey@users.noreply.github.com>
 Date: Tue, 19 Sep 2023 09:39:36 +0800
-Subject: [PATCH 05/29] fix hpu deserialization bug (#165)
+Subject: [PATCH 05/30] fix hpu deserialization bug (#165)
 
 ---
  torch/serialization.py | 4 ++--
diff --git a/torch_patches/0006-C10d-Add-xpu-as-the-supported-device-in-device-backe.patch b/torch_patches/0006-C10d-Add-xpu-as-the-supported-device-in-device-backe.patch
@@ -1,7 +1,7 @@
 From 407a10305a5c104edc1e5bdd22cf7aa3544c7894 Mon Sep 17 00:00:00 2001
 From: zhuhong61 <95205772+zhuhong61@users.noreply.github.com>
 Date: Thu, 21 Sep 2023 13:14:52 +0800
-Subject: [PATCH 06/29] C10d: Add xpu as the supported device in device backend
+Subject: [PATCH 06/30] C10d: Add xpu as the supported device in device backend
  map (#166)
 
 ---
diff --git a/torch_patches/0007-add-path-for-DPC-SYCL-device-code-167.patch b/torch_patches/0007-add-path-for-DPC-SYCL-device-code-167.patch
@@ -1,7 +1,7 @@
 From adbcccaa2248aa881ff5282ddf34e4eede6cb25d Mon Sep 17 00:00:00 2001
 From: min-jean-cho <min.jean.cho@intel.com>
 Date: Fri, 22 Sep 2023 03:09:51 -0700
-Subject: [PATCH 07/29] add path for DPC++ SYCL device code (#167)
+Subject: [PATCH 07/30] add path for DPC++ SYCL device code (#167)
 
 ---
  c10/util/Float8_e4m3fn.h | 4 ++++
diff --git a/torch_patches/0008-generalize-inductor-triton-backend-device-agnostic-1.patch b/torch_patches/0008-generalize-inductor-triton-backend-device-agnostic-1.patch
@@ -1,7 +1,7 @@
 From 0d61b020088d65064072b1cb4d64d249120a88fe Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <106960996+guangyey@users.noreply.github.com>
 Date: Tue, 26 Sep 2023 20:49:36 +0800
-Subject: [PATCH 08/29] generalize inductor triton backend device agnostic
+Subject: [PATCH 08/30] generalize inductor triton backend device agnostic
  (#168)
 
 ---
diff --git a/torch_patches/0009-Integrate-xpu-into-torch.Generator-and-torch.seed-10.patch b/torch_patches/0009-Integrate-xpu-into-torch.Generator-and-torch.seed-10.patch
@@ -1,7 +1,7 @@
 From 6c5a85526546347ae39d66fca5beb31af09d1c16 Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <106960996+guangyey@users.noreply.github.com>
 Date: Thu, 28 Sep 2023 10:07:40 +0800
-Subject: [PATCH 09/29] Integrate xpu into torch.Generator and torch.seed
+Subject: [PATCH 09/30] Integrate xpu into torch.Generator and torch.seed
  (#109866) (#110220) (#110376) (#169)
 
 Integrate torch.xpu.Generator into torch.Generator
diff --git a/torch_patches/0010-Enable-xpu-backend-in-torchdynamo-benchmarks-171.patch b/torch_patches/0010-Enable-xpu-backend-in-torchdynamo-benchmarks-171.patch
@@ -1,7 +1,7 @@
 From 58392a58c207ab03600c7abd6e7b3baa21262c3a Mon Sep 17 00:00:00 2001
 From: "Wang, Chuanqi" <chuanqi.wang@intel.com>
 Date: Thu, 12 Oct 2023 14:29:35 +0800
-Subject: [PATCH 10/29] Enable xpu backend in torchdynamo benchmarks (#171)
+Subject: [PATCH 10/30] Enable xpu backend in torchdynamo benchmarks (#171)
 
 ---
  benchmarks/dynamo/common.py      | 29 ++++++++++++++++++++++-------
diff --git a/torch_patches/0011-support-channel-last-for-xpu-conv-in-inductor-layout.patch b/torch_patches/0011-support-channel-last-for-xpu-conv-in-inductor-layout.patch
@@ -1,7 +1,7 @@
 From 4400b4896a315a7c9009b141b6e8f9f737cfeb7e Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <106960996+guangyey@users.noreply.github.com>
 Date: Thu, 12 Oct 2023 23:40:03 +0800
-Subject: [PATCH 11/29] support channel last for xpu conv in inductor layout
+Subject: [PATCH 11/30] support channel last for xpu conv in inductor layout
  opt path (#174)
 
 ---
diff --git a/torch_patches/0012-Cherry-pick-https-github.com-pytorch-pytorch-pull-10.patch b/torch_patches/0012-Cherry-pick-https-github.com-pytorch-pytorch-pull-10.patch
@@ -1,7 +1,7 @@
 From 56978c02bd83816bc92f4dbc34c93e2f3169e5b7 Mon Sep 17 00:00:00 2001
 From: zejun <zejun.chen@intel.com>
 Date: Wed, 18 Oct 2023 15:19:50 +0800
-Subject: [PATCH 12/29] Cherry-pick:
+Subject: [PATCH 12/30] Cherry-pick:
  https://github.com/pytorch/pytorch/pull/108312 (#175)
 
 Signed-off-by: Chen, Zejun <zejun.chen@intel.com>
diff --git a/torch_patches/0013-Inductor-backend-specific-num-sm-for-XPU-backend-in-.patch b/torch_patches/0013-Inductor-backend-specific-num-sm-for-XPU-backend-in-.patch
@@ -1,7 +1,7 @@
 From 00f48b6ff7b342a322098dbeec0b9488f52823df Mon Sep 17 00:00:00 2001
 From: Feng Yuan <feng1.yuan@intel.com>
 Date: Thu, 19 Oct 2023 09:49:25 +0800
-Subject: [PATCH 13/29] [Inductor] backend specific num sm for XPU backend in
+Subject: [PATCH 13/30] [Inductor] backend specific num sm for XPU backend in
  Triton reduction kernel (#176)
 
 Signed-off-by: Feng Yuan <feng1.yuan@intel.com>
diff --git a/torch_patches/0014-ATen-scaled_dot_product_attention-follow-supposed-CU.patch b/torch_patches/0014-ATen-scaled_dot_product_attention-follow-supposed-CU.patch
@@ -1,7 +1,7 @@
 From a2a7717a2d25ab71869a33600a28b4728fb8b70d Mon Sep 17 00:00:00 2001
 From: majing <jing1.ma@intel.com>
 Date: Mon, 23 Oct 2023 11:33:05 +0800
-Subject: [PATCH 14/29] ATen scaled_dot_product_attention: follow supposed
+Subject: [PATCH 14/30] ATen scaled_dot_product_attention: follow supposed
  (CUDA/CPU) operators registration strategy (#177)
 
 * Enable XPU backend implementation for scaled_dot_product_attention
diff --git a/torch_patches/0015-Profiler-Kineto-Legacy-Enable-Profiler-with-Legacy-K.patch b/torch_patches/0015-Profiler-Kineto-Legacy-Enable-Profiler-with-Legacy-K.patch
@@ -1,7 +1,7 @@
 From 1af783e03f8e7c223b2fbb0e60117d1ea64001bd Mon Sep 17 00:00:00 2001
 From: "Xunsong, Huang" <xunsong.huang@intel.com>
 Date: Fri, 27 Oct 2023 21:07:29 +0800
-Subject: [PATCH 15/29] [Profiler] [Kineto+Legacy] Enable Profiler with Legacy
+Subject: [PATCH 15/30] [Profiler] [Kineto+Legacy] Enable Profiler with Legacy
  & Kineto Support Together (#179)
 
 * Enable XPU profiling with Kineto support
diff --git a/torch_patches/0016-Fix-dynamo-benchmark-random-seed-for-xpu-device-181.patch b/torch_patches/0016-Fix-dynamo-benchmark-random-seed-for-xpu-device-181.patch
@@ -1,7 +1,7 @@
 From 602734510615ec93a5e05364aa5995eb4e624ef9 Mon Sep 17 00:00:00 2001
 From: "Wang, Chuanqi" <chuanqi.wang@intel.com>
 Date: Fri, 10 Nov 2023 16:40:18 +0800
-Subject: [PATCH 16/29] Fix dynamo benchmark random seed for xpu device (#181)
+Subject: [PATCH 16/30] Fix dynamo benchmark random seed for xpu device (#181)
 
 * Fix dynamo benchmark random seed for xpu device
 
diff --git a/torch_patches/0017-remove-debug-message-182.patch b/torch_patches/0017-remove-debug-message-182.patch
@@ -1,7 +1,7 @@
 From dd9913ff2d889dd9dac05929c5a4aa7713dc741a Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <106960996+guangyey@users.noreply.github.com>
 Date: Fri, 10 Nov 2023 16:43:43 +0800
-Subject: [PATCH 17/29] remove debug message (#182)
+Subject: [PATCH 17/30] remove debug message (#182)
 
 ---
  torch/autograd/profiler_legacy.py | 2 --
diff --git a/torch_patches/0018-Nested-Tensor-Add-xpu-device-in-assertion-for-nested.patch b/torch_patches/0018-Nested-Tensor-Add-xpu-device-in-assertion-for-nested.patch
@@ -1,7 +1,7 @@
 From 5c5feeef2804f452bda93462633be85cff9c8bcd Mon Sep 17 00:00:00 2001
 From: zejun <zejun.chen@intel.com>
 Date: Wed, 29 Nov 2023 15:10:32 +0800
-Subject: [PATCH 18/29] [Nested Tensor] Add xpu device in assertion for nested
+Subject: [PATCH 18/30] [Nested Tensor] Add xpu device in assertion for nested
  tensor creation (#197)
 
 * [Nested Tensor] Add xpu device in assertion for nested tensor creation
diff --git a/torch_patches/0019-Cherry-pick-PR-to-enable-nested-cos-and-sin-198.patch b/torch_patches/0019-Cherry-pick-PR-to-enable-nested-cos-and-sin-198.patch
@@ -1,7 +1,7 @@
 From b5c6f139f0f7b3972d956e7246a88facee18e842 Mon Sep 17 00:00:00 2001
 From: zejun <zejun.chen@intel.com>
 Date: Wed, 29 Nov 2023 15:10:51 +0800
-Subject: [PATCH 19/29] Cherry pick PR to enable nested cos and sin (#198)
+Subject: [PATCH 19/30] Cherry pick PR to enable nested cos and sin (#198)
 
 Signed-off-by: Chen, Zejun <zejun.chen@intel.com>
 ---
diff --git a/torch_patches/0020-fix-deterministic-issue-199.patch b/torch_patches/0020-fix-deterministic-issue-199.patch
@@ -1,7 +1,7 @@
 From 5f22c2c503d7e424f2d2953fec333e0e8e901d4d Mon Sep 17 00:00:00 2001
 From: "Yu, Guangye" <106960996+guangyey@users.noreply.github.com>
 Date: Wed, 29 Nov 2023 18:08:57 +0800
-Subject: [PATCH 20/29] fix deterministic issue (#199)
+Subject: [PATCH 20/30] fix deterministic issue (#199)
 
 ---
  benchmarks/dynamo/common.py | 2 ++
diff --git a/torch_patches/0021-Add-sfdp-pattern-for-huggingface-Bert-fp16-inference.patch b/torch_patches/0021-Add-sfdp-pattern-for-huggingface-Bert-fp16-inference.patch
@@ -1,7 +1,7 @@
 From 44d1ed87a2675d0764e32b8839936371da8ea3a8 Mon Sep 17 00:00:00 2001
 From: xingyuan li <108672484+hoshibara@users.noreply.github.com>
 Date: Fri, 8 Dec 2023 14:48:45 +0800
-Subject: [PATCH 21/29] Add sfdp pattern for huggingface Bert fp16 inference
+Subject: [PATCH 21/30] Add sfdp pattern for huggingface Bert fp16 inference
  model (#195)
 
 * Add sfdp pattern for huggingface Bert fp16 inference model (to half version & xpu optimized version)
diff --git a/torch_patches/0022-Dynamo-handle-the-.tolist-method-of-np.arrays-in-dyn.patch b/torch_patches/0022-Dynamo-handle-the-.tolist-method-of-np.arrays-in-dyn.patch
@@ -1,7 +1,7 @@
 From 1aa638adaa6c4424faa5e5c7dfc8473118589198 Mon Sep 17 00:00:00 2001
 From: "Weishi.Deng" <weishi.deng@intel.com>
 Date: Fri, 12 Jan 2024 16:06:31 +0800
-Subject: [PATCH 22/29] Dynamo: handle the .tolist method of np.arrays in
+Subject: [PATCH 22/30] Dynamo: handle the .tolist method of np.arrays in
  dynamo (#201)
 
 ---
diff --git a/torch_patches/0023-expose-nested-related-header-202.patch b/torch_patches/0023-expose-nested-related-header-202.patch
@@ -1,7 +1,7 @@
 From 2f3017f60aa8b9363acb62c180b6d362c836b5d0 Mon Sep 17 00:00:00 2001
 From: Lei Zhenyuan <zhenyuan.lei@intel.com>
 Date: Fri, 26 Jan 2024 09:46:01 +0800
-Subject: [PATCH 23/29] expose nested related header (#202)
+Subject: [PATCH 23/30] expose nested related header (#202)
 
 ---
  aten/src/ATen/CMakeLists.txt | 3 ++-
diff --git a/torch_patches/0024-use-ONEMKL_LIBRARIES-for-mkl-libs-in-torch-to-not-ov.patch b/torch_patches/0024-use-ONEMKL_LIBRARIES-for-mkl-libs-in-torch-to-not-ov.patch
@@ -1,7 +1,7 @@
 From 730877fe5a89e11c36ef1873b72083bea9f35c72 Mon Sep 17 00:00:00 2001
 From: min-jean-cho <min.jean.cho@intel.com>
 Date: Wed, 28 Feb 2024 10:00:30 -0800
-Subject: [PATCH 24/29] use ONEMKL_LIBRARIES for mkl libs in torch to not
+Subject: [PATCH 24/30] use ONEMKL_LIBRARIES for mkl libs in torch to not
  overwritten by oneMKL MKL_LIBRARIES cmake config (#208)
 
 ---
diff --git a/torch_patches/0025-cherry-pick-https-github.com-pytorch-pytorch-pull-11.patch b/torch_patches/0025-cherry-pick-https-github.com-pytorch-pytorch-pull-11.patch
@@ -1,7 +1,7 @@
 From dc10c9d352ca60bd0cf6d9a746fe10b6460e7dbd Mon Sep 17 00:00:00 2001
 From: min-jean-cho <min.jean.cho@intel.com>
 Date: Wed, 28 Feb 2024 16:46:37 -0800
-Subject: [PATCH 25/29] cherry pick
+Subject: [PATCH 25/30] cherry pick
  https://github.com/pytorch/pytorch/pull/117497 (#207)
 
 ---
diff --git a/torch_patches/0026-expose-nested-header-to-wheel-206.patch b/torch_patches/0026-expose-nested-header-to-wheel-206.patch
@@ -1,7 +1,7 @@
 From f24e7798951208a4409cef909ff9615065577329 Mon Sep 17 00:00:00 2001
 From: Lei Zhenyuan <zhenyuan.lei@intel.com>
 Date: Fri, 1 Mar 2024 12:33:22 +0800
-Subject: [PATCH 26/29] expose nested header to wheel (#206)
+Subject: [PATCH 26/30] expose nested header to wheel (#206)
 
 ---
  setup.py | 1 +
diff --git a/torch_patches/0027-Hide-MKL-symbols-211.patch b/torch_patches/0027-Hide-MKL-symbols-211.patch
@@ -1,7 +1,7 @@
 From 7897dbe6b8868cfe318e52bab1e233acc07b40f3 Mon Sep 17 00:00:00 2001
 From: "Cui, Yifeng" <yifeng.cui@intel.com>
 Date: Thu, 7 Mar 2024 13:30:38 +0800
-Subject: [PATCH 27/29] Hide MKL symbols (#211)
+Subject: [PATCH 27/30] Hide MKL symbols (#211)
 
 ---
  cmake/public/mkl.cmake | 12 ++++++++++++
diff --git a/torch_patches/0028-Fix-Windows-Build-214.patch b/torch_patches/0028-Fix-Windows-Build-214.patch
@@ -1,7 +1,7 @@
 From 1b1c17bb765537fa099e8faea12c86f4254323bf Mon Sep 17 00:00:00 2001
 From: "Cui, Yifeng" <yifeng.cui@intel.com>
 Date: Mon, 11 Mar 2024 19:46:04 +0800
-Subject: [PATCH 28/29] Fix Windows Build (#214)
+Subject: [PATCH 28/30] Fix Windows Build (#214)
 
 ---
  cmake/public/mkl.cmake | 20 +++++++++++---------
diff --git a/torch_patches/0029-Move-observer-to-device-203.patch b/torch_patches/0029-Move-observer-to-device-203.patch
@@ -1,7 +1,7 @@
 From c61d29a68eaa4af9a5941e102a8ba59f989cf0d6 Mon Sep 17 00:00:00 2001
 From: Zhiwei <zhiwei.yan@intel.com>
 Date: Fri, 15 Mar 2024 15:18:03 +0800
-Subject: [PATCH 29/29] Move observer to device (#203)
+Subject: [PATCH 29/30] Move observer to device (#203)
 
 ---
  torch/ao/quantization/observer.py | 2 ++
diff --git a/torch_patches/0030-Allocate-a-temporary-output-tensor-for-linalg_eigval.patch b/torch_patches/0030-Allocate-a-temporary-output-tensor-for-linalg_eigval.patch
@@ -0,0 +1,27 @@
+From 1c122af900de226d2324b789a5c9864870640b5d Mon Sep 17 00:00:00 2001
+From: "Cui, Yifeng" <yifeng.cui@intel.com>
+Date: Thu, 11 Apr 2024 14:00:06 +0800
+Subject: [PATCH 30/30] Allocate a temporary output tensor for linalg_eigvals on
+ Non-CPU device (#226) (#230)
+
+Cherry pick from https://github.com/pytorch/pytorch/pull/116682
+---
+ aten/src/ATen/native/BatchLinearAlgebra.cpp | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/aten/src/ATen/native/BatchLinearAlgebra.cpp b/aten/src/ATen/native/BatchLinearAlgebra.cpp
+index 4d0c9fa8ad4..a760c85977b 100644
+--- a/aten/src/ATen/native/BatchLinearAlgebra.cpp
++++ b/aten/src/ATen/native/BatchLinearAlgebra.cpp
+@@ -3081,7 +3081,7 @@ Tensor& linalg_eigvals_out(const Tensor& input, Tensor& values) {
+ 
+   // because MAGMA's GEEV takes CPU inputs and returns CPU outputs
+   // 'values' tensor that is on GPU device can't be used directly
+-  values_tmp_needed |= values.is_cuda();
++  values_tmp_needed |= (!values.is_cpu());
+ 
+   // determine the appropriate scalar_type for the temporary tensors
+   ScalarType values_type = input.scalar_type();
+-- 
+2.34.1
+
