[SYCL] Use uniform group instructions in headers #5705
Conversation
|
There are changes in extension vendor & operations name. @MrSidims will we promote it as INTEL extension now and then change it to KHR? |
|
@aobolensk please update the instructions' names to the names used here: KhronosGroup/SPIRV-Registry#138 |
| @@ -49,3 +49,4 @@ EXT(SPV_INTEL_bfloat16_conversion) | |||
| EXT(SPV_INTEL_joint_matrix) | |||
| EXT(SPV_INTEL_hw_thread_queries) | |||
| EXT(SPV_INTEL_global_variable_decorations) | |||
| EXT(SPV_KHR_uniform_group_instructions) | |||
There was a problem hiding this comment.
I don't sure if we need it here, until the extension is ready and transfered to intel/llvm repository: KhronosGroup/SPIRV-LLVM-Translator#1253. The same about Clang.cpp. @MrSidims WDYT?
There was a problem hiding this comment.
@aobolensk please remove this change from the translator, let the translator's work be merged and pulled to intel/llvm first.
| @@ -49,3 +49,4 @@ EXT(SPV_INTEL_bfloat16_conversion) | |||
| EXT(SPV_INTEL_joint_matrix) | |||
| EXT(SPV_INTEL_hw_thread_queries) | |||
| EXT(SPV_INTEL_global_variable_decorations) | |||
| EXT(SPV_KHR_uniform_group_instructions) | |||
There was a problem hiding this comment.
@aobolensk please remove this change from the translator, let the translator's work be merged and pulled to intel/llvm first.
| @@ -9119,7 +9119,8 @@ void SPIRVTranslator::ConstructJob(Compilation &C, const JobAction &JA, | |||
| std::string DefaultExtArg = | |||
| ",+SPV_EXT_shader_atomic_float_add,+SPV_EXT_shader_atomic_float_min_max" | |||
| ",+SPV_KHR_no_integer_wrap_decoration,+SPV_KHR_float_controls" | |||
| ",+SPV_KHR_expect_assume,+SPV_KHR_linkonce_odr"; | |||
| ",+SPV_KHR_expect_assume,+SPV_KHR_linkonce_odr" | |||
| ",+SPV_KHR_uniform_group_instructions"; | |||
There was a problem hiding this comment.
Let's enable the extension on any target but FPGA H/W aka move this code to line 9154.
|
@aobolensk KhronosGroup/SPIRV-LLVM-Translator#1253 is in intel/llvm now |
aobolensk
force-pushed
the
uniform_group_instructions
branch
from
5a4bed5 to
121db8b
Compare
Thanks! Rebased this on top of merged change |
…nsions intel#5705 changed the use of NonUniform group binary operators to the KHR extension operators. These changes rename the implementation of these in libclc for both amdgcn and PTX to the new naming. Signed-off-by: Larsen, Steffen <[email protected]>
…nsions (#6279) #5705 changed the use of NonUniform group binary operators to the KHR extension operators. These changes rename the implementation of these in libclc for both amdgcn and PTX to the new naming. Signed-off-by: Larsen, Steffen <[email protected]>
No description provided.