[mux] Remove degenerate subgroup support.

Older versions of oneAPI Construction Kit used an implementation of
subgroups where the subgroup size was always equal to the workgroup
size. This implementation is no longer used by any targets, but the code
to support it was still in place. This commit removes it.

CHANGELOG.md

@@ -1,5 +1,12 @@
 # ComputeAorta Changes
 
+## TBD
+
+Upgrade guidance:
+
+* Support for degenerate subgroups has been removed. No in-tree target or
+  template was using this, but custom targets may need to be updated.
+
 ## Version 4.0.0
 
 Upgrade guidance:

doc/modules/mux/changes.rst

@@ -11,6 +11,11 @@ version increases mean backward compatible bug fixes have been applied.
    Versions prior to 1.0.0 may contain breaking changes in minor
    versions as the API is still under development.
 
+0.81.0
+------
+
+* Removed ``mux-degenerate-subgroups``.
+
 0.80.0
 ------
 

doc/specifications/mux-compiler-spec.rst

@@ -1,7 +1,7 @@
 ComputeMux Compiler Specification
 =================================
 
-   This is version 0.80.0 of the specification.
+   This is version 0.81.0 of the specification.
 
 ComputeMux is Codeplay’s proprietary API for executing compute workloads across
 heterogeneous devices. ComputeMux is an extremely lightweight,
@@ -1432,9 +1432,6 @@ different stages of the pipeline:
        by the use of known mux sub-group builtins). If a pass introduces the
        explicit use of sub-groups to a function, it should remove this
        attribute.
-   * - ``"mux-degenerate-subgroups"``
-     - Marks the function has using degenerate sub-groups (i.e. one sub-group
-       for the entire local work-group).
 
 ``mux-kernel`` attribute
 ~~~~~~~~~~~~~~~~~~~~~~~~

doc/specifications/mux-runtime-spec.rst

@@ -1,7 +1,7 @@
 ComputeMux Runtime Specification
 ================================
 
-   This is version 0.80.0 of the specification.
+   This is version 0.81.0 of the specification.
 
 ComputeMux is Codeplay’s proprietary API for executing compute workloads across
 heterogeneous devices. ComputeMux is an extremely lightweight,

modules/compiler/compiler_pipeline/CMakeLists.txt

@@ -27,7 +27,6 @@ add_ca_library(compiler-pipeline STATIC
   ${CMAKE_CURRENT_SOURCE_DIR}/include/compiler/utils/compute_local_memory_usage_pass.h
   ${CMAKE_CURRENT_SOURCE_DIR}/include/compiler/utils/define_mux_builtins_pass.h
   ${CMAKE_CURRENT_SOURCE_DIR}/include/compiler/utils/define_mux_dma_pass.h
-  ${CMAKE_CURRENT_SOURCE_DIR}/include/compiler/utils/degenerate_sub_group_pass.h
   ${CMAKE_CURRENT_SOURCE_DIR}/include/compiler/utils/device_info.h
   ${CMAKE_CURRENT_SOURCE_DIR}/include/compiler/utils/dma.h
   ${CMAKE_CURRENT_SOURCE_DIR}/include/compiler/utils/encode_builtin_range_metadata_pass.h
@@ -79,7 +78,6 @@ add_ca_library(compiler-pipeline STATIC
   ${CMAKE_CURRENT_SOURCE_DIR}/source/compute_local_memory_usage_pass.cpp
   ${CMAKE_CURRENT_SOURCE_DIR}/source/define_mux_builtins_pass.cpp
   ${CMAKE_CURRENT_SOURCE_DIR}/source/define_mux_dma_pass.cpp
-  ${CMAKE_CURRENT_SOURCE_DIR}/source/degenerate_sub_group_pass.cpp
   ${CMAKE_CURRENT_SOURCE_DIR}/source/dma.cpp
   ${CMAKE_CURRENT_SOURCE_DIR}/source/encode_builtin_range_metadata_pass.cpp
   ${CMAKE_CURRENT_SOURCE_DIR}/source/encode_kernel_metadata_pass.cpp

modules/compiler/compiler_pipeline/include/compiler/utils/attributes.h

@@ -162,16 +162,6 @@ void setBarrierSchedule(llvm::CallInst &CI, BarrierSchedule Sched);
 /// @return the execution schedule for this barrier
 BarrierSchedule getBarrierSchedule(const llvm::CallInst &CI);
 
-/// @brief Marks a kernel's subgroups as degenerate
-///
-/// @param[in] F Function in which to encode the information.
-void setHasDegenerateSubgroups(llvm::Function &F);
-
-/// @brief Returns whether the kernel has degenerate subgroups.
-///
-/// @param[in] F Function to check.
-bool hasDegenerateSubgroups(const llvm::Function &F);
-
 /// @brief Marks a function as not explicitly using subgroups
 ///
 /// May be set even with unresolved external functions, assuming those don't

modules/compiler/compiler_pipeline/source/attributes.cpp

@@ -186,18 +186,6 @@ BarrierSchedule getBarrierSchedule(const CallInst &CI) {
   return BarrierSchedule::Unordered;
 }
 
-static constexpr const char *MuxDegenerateSubgroupsAttrName =
-    "mux-degenerate-subgroups";
-
-void setHasDegenerateSubgroups(Function &F) {
-  F.addFnAttr(MuxDegenerateSubgroupsAttrName);
-}
-
-bool hasDegenerateSubgroups(const Function &F) {
-  const Attribute Attr = F.getFnAttribute(MuxDegenerateSubgroupsAttrName);
-  return Attr.isValid();
-}
-
 static constexpr const char *MuxNoSubgroupsAttrName = "mux-no-subgroups";
 
 void setHasNoExplicitSubgroups(Function &F) {