Update on "[ET-VK] Moving repeat functionality from copy_packed_dim_offset into a separate repeat shader."

This diff includes changes to the Vulkan backend for Executorch, specifically in the implementation of the Copy and Repeat ops. The changes add a new GLSL shader for repeat op, which allows for repeated copying from an input tensor.

Differential Revision: [D71511822](https://our.internmc.facebook.com/intern/diff/D71511822/)

[ghstack-poisoned]

Author: trivedivivek

.ci/scripts/test_qnn_static_llama.sh

@@ -5,7 +5,7 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-set -exu
+set -euxo pipefail
 
 source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
 
@@ -56,4 +56,3 @@ if [ $exit_code1 -ne 0 ] || [ $exit_code2 -ne 0 ]; then
 else
     exit 0
 fi
-set -e

.ci/scripts/utils.sh

@@ -32,7 +32,7 @@ install_executorch() {
   which pip
   # Install executorch, this assumes that Executorch is checked out in the
   # current directory.
-  ./install_executorch.sh --pybind xnnpack "$@"
+  ./install_executorch.sh "$@"
   # Just print out the list of packages for debugging
   pip list
 }

.github/workflows/apple.yml

@@ -37,7 +37,7 @@ jobs:
         id: set_version
         shell: bash
         run: |
-          VERSION="0.5.0.$(TZ='PST8PDT' date +%Y%m%d)"
+          VERSION="0.7.0.$(TZ='PST8PDT' date +%Y%m%d)"
           echo "version=$VERSION" >> "$GITHUB_OUTPUT"
 
   build-demo-ios:

.github/workflows/pull.yml

@@ -573,7 +573,6 @@ jobs:
 
         BUILD_TOOL="cmake"
 
-        ./install_requirements.sh --use-pt-pinned-commit
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-qnn-deps.sh
         PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
 

backends/arm/ethosu_backend.py

@@ -23,7 +23,6 @@
 
 # debug functionality
 logger = logging.getLogger(__name__)
-logger.setLevel(logging.WARNING)
 
 
 @final

backends/arm/process_node.py

@@ -14,11 +14,7 @@
 from executorch.backends.arm.operators.node_visitor import NodeVisitor
 from executorch.backends.arm.tosa_mapping import TosaArg
 from executorch.backends.arm.tosa_specification import TosaSpecification
-from executorch.backends.arm.tosa_utils import (
-    get_node_debug_info,
-    getNodeArgs,
-    tosa_shape,
-)
+from executorch.backends.arm.tosa_utils import getNodeArgs, tosa_shape
 from torch.export.exported_program import ExportedProgram
 
 
@@ -36,7 +32,7 @@ def process_call_function(
         output = TosaArg(node)
     except ValueError as e:
         raise ValueError(
-            f"Failed processing call_function:\n{get_node_debug_info(node)}"
+            f"Failed processing call_function: {node.name}. "
             "Is the original torch function supported?"
         ) from e
     tosa_graph.currRegion.currBasicBlock.addTensor(
@@ -74,7 +70,7 @@ def process_inputs(
         tosa_arg = TosaArg(node)
     except ValueError as e:
         raise ValueError(
-            f"Failed processing input placeholder:\n{get_node_debug_info(node)}"
+            f"Failed processing input placeholder: {node.name}. "
             "Is the original torch function supported?"
         ) from e
     input_shape = tosa_arg.shape
@@ -100,7 +96,7 @@ def process_inputs_to_parameters(
         tosa_arg = TosaArg(node)
     except ValueError as e:
         raise ValueError(
-            f"Failed processing parameter placeholder:\n{get_node_debug_info(node)}"
+            f"Failed processing parameter placeholder: {node.name}. "
             "Is the original torch function supported?"
         ) from e
     parameter_name = edge_program.graph_signature.inputs_to_parameters[tosa_arg.name]
@@ -129,7 +125,7 @@ def process_inputs_to_buffers(
         tosa_arg = TosaArg(node)
     except ValueError as e:
         raise ValueError(
-            f"Failed processing buffer placeholder:\n{get_node_debug_info(node)}"
+            f"Failed processing buffer placeholder: {node.name}. "
             "Is the original torch function supported?"
         ) from e
     buffer_name = edge_program.graph_signature.inputs_to_buffers[node.name]
@@ -157,7 +153,7 @@ def process_inputs_to_lifted_tensor_constants(
         tosa_arg = TosaArg(node)
     except ValueError as e:
         raise ValueError(
-            f"Failed processing lifted tensor constant placeholder:\n{get_node_debug_info(node)}"
+            f"Failed processing lifted tensor constant placeholder: {node.name}. "
             "Is the original torch function supported?"
         ) from e
     tensor_name = edge_program.graph_signature.inputs_to_lifted_tensor_constants[

backends/arm/test/ops/test_any.py

@@ -6,6 +6,7 @@
 
 from typing import List, Tuple
 
+import pytest
 import torch
 from executorch.backends.arm.test import common
 from executorch.backends.arm.test.tester.test_pipeline import (
@@ -117,25 +118,6 @@ def forward(self, x: torch.Tensor):
 }
 
 
-fvp_xfails = {
-    "any_rank1": "MLETORCH-706 Support ScalarType::Bool in EthosUBackend.",
-    "any_rank1_squeeze": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank2": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank2_squeeze": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank2_dims": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank2_dims_squeeze": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank3_dims_squeeze": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank4": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank4_squeeze": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank4_dims": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank4_dims_squeeze": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank1_reduce_all": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank2_reduce_all": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank3_reduce_all": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-    "any_rank4_reduce_all": "MLETORCH-706: Support ScalarType::Bool in EthosUBackend.",
-}
-
-
 @common.parametrize("test_data", test_data)
 def test_any_tosa_MI(test_data: input_t1):
     op, test_input = test_data
@@ -147,13 +129,13 @@ def test_any_tosa_MI(test_data: input_t1):
 def test_any_tosa_BI(test_data: input_t1):
     op, test_input = test_data
     pipeline = TosaPipelineBI[input_t1](op, test_input, op.aten_op, op.exir_op)
-    pipeline.pop_stage(pipeline.find_pos("quantize") + 1)
     pipeline.pop_stage("quantize")
+    pipeline.pop_stage("check.quant_nodes")
     pipeline.run()
 
 
 @common.parametrize("test_data", test_data)
-def test_logical_u55_BI(test_data: input_t1):
+def test_any_u55_BI(test_data: input_t1):
     # Tests that we don't delegate these ops since they are not supported on U55.
     op, test_input = test_data
     pipeline = OpNotSupportedPipeline[input_t1](
@@ -163,23 +145,13 @@ def test_logical_u55_BI(test_data: input_t1):
 
 
 @common.parametrize("test_data", test_data)
-def test_floor_u85_BI(test_data: input_t1):
-    op, test_input = test_data
-    pipeline = EthosU85PipelineBI[input_t1](
-        op, test_input, op.aten_op, op.exir_op, run_on_fvp=False
-    )
-    pipeline.pop_stage(pipeline.find_pos("quantize") + 1)
-    pipeline.pop_stage("quantize")
-    pipeline.run()
-
-
-@common.parametrize("test_data", test_data, fvp_xfails)
-@common.SkipIfNoCorstone320
-def test_floor_u85_BI_on_fvp(test_data: input_t1):
+@pytest.mark.xfail(reason="MLETORCH-706: Support ScalarType::Bool in EthosUBackend.")
+@common.XfailIfNoCorstone320
+def test_any_u85_BI(test_data: input_t1):
     op, test_input = test_data
     pipeline = EthosU85PipelineBI[input_t1](
         op, test_input, op.aten_op, op.exir_op, run_on_fvp=True
     )
-    pipeline.pop_stage(pipeline.find_pos("quantize") + 1)
     pipeline.pop_stage("quantize")
+    pipeline.pop_stage("check.quant_nodes")
     pipeline.run()