Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add OpCooperativeMatrixApplyFunctionINTEL instruction #2214

Merged
merged 6 commits into from
Nov 17, 2023
Merged

Add OpCooperativeMatrixApplyFunctionINTEL instruction #2214

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
1e33c8b
adding OpCooperativeMatrixApplyFunctionINTEL operation and Cooperativ…
VyacheslavLevytskyy Nov 14, 2023
d039fdd
Add test
MrSidims Nov 15, 2023
7682dd4
Remove some unnecessary things
MrSidims Nov 15, 2023
4983844
implement requirement: implicitly declares CooperativeMatrixKHR
VyacheslavLevytskyy Nov 16, 2023
e7ad51a
change tokens order
VyacheslavLevytskyy Nov 17, 2023
a76e816
fix a wrong value of the Use parameter to match documented accepted r…
VyacheslavLevytskyy Nov 17, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions lib/SPIRV/libSPIRV/SPIRVEnum.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -217,6 +217,8 @@ template <> inline void SPIRVMap<SPIRVCapabilityKind, SPIRVCapVec>::init() {
{internal::CapabilityJointMatrixINTEL});
ADD_VEC_INIT(internal::CapabilityCooperativeMatrixPrefetchINTEL,
{CapabilityCooperativeMatrixKHR});
ADD_VEC_INIT(internal::CapabilityCooperativeMatrixInvocationInstructionsINTEL,
{CapabilityCooperativeMatrixKHR});
}

template <> inline void SPIRVMap<SPIRVExecutionModelKind, SPIRVCapVec>::init() {
Expand Down
20 changes: 20 additions & 0 deletions lib/SPIRV/libSPIRV/SPIRVInstruction.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3420,6 +3420,26 @@ class SPIRVCooperativeMatrixPrefetchINTELInstBase
_SPIRV_OP(CooperativeMatrixPrefetch, false, 8, true, 5)
#undef _SPIRV_OP

class SPIRVCooperativeMatrixInvocationInstructionsINTELInstBase
MrSidims marked this conversation as resolved.
Show resolved Hide resolved
: public SPIRVInstTemplateBase {
protected:
std::optional<ExtensionID> getRequiredExtension() const override {
return ExtensionID::SPV_INTEL_joint_matrix;
}
SPIRVCapVec getRequiredCapability() const override {
return getVec(
internal::CapabilityCooperativeMatrixInvocationInstructionsINTEL);
MrSidims marked this conversation as resolved.
Show resolved Hide resolved
}
};

#define _SPIRV_OP(x, ...) \
typedef SPIRVInstTemplate< \
SPIRVCooperativeMatrixInvocationInstructionsINTELInstBase, \
internal::Op##x##INTEL, __VA_ARGS__> \
SPIRV##x##INTEL;
_SPIRV_OP(CooperativeMatrixApplyFunction, true, 5)
MrSidims marked this conversation as resolved.
Show resolved Hide resolved
#undef _SPIRV_OP

class SPIRVCooperativeMatrixKHRInstBase : public SPIRVInstTemplateBase {
protected:
std::optional<ExtensionID> getRequiredExtension() const override {
Expand Down
2 changes: 2 additions & 0 deletions lib/SPIRV/libSPIRV/SPIRVNameMapEnum.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -657,6 +657,8 @@ template <> inline void SPIRVMap<Capability, std::string>::init() {
add(internal::CapabilityCacheControlsINTEL, "CacheControlsINTEL");
add(internal::CapabilityCooperativeMatrixPrefetchINTEL,
"CooperativeMatrixPrefetchINTEL");
add(internal::CapabilityCooperativeMatrixInvocationInstructionsINTEL,
"CooperativeMatrixInvocationInstructionsINTEL");
}
SPIRV_DEF_NAMEMAP(Capability, SPIRVCapabilityNameMap)

Expand Down
2 changes: 2 additions & 0 deletions lib/SPIRV/libSPIRV/SPIRVOpCodeEnumInternal.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,8 @@ _SPIRV_OP_INTERNAL(JointMatrixGetElementCoordINTEL,
internal::OpJointMatrixGetElementCoordINTEL)
_SPIRV_OP_INTERNAL(CooperativeMatrixPrefetchINTEL,
internal::OpCooperativeMatrixPrefetchINTEL)
_SPIRV_OP_INTERNAL(CooperativeMatrixApplyFunctionINTEL,
internal::OpCooperativeMatrixApplyFunctionINTEL)
_SPIRV_OP_INTERNAL(ComplexFMulINTEL, internal::ComplexFMulINTEL)
_SPIRV_OP_INTERNAL(ComplexFDivINTEL, internal::ComplexFDivINTEL)
_SPIRV_OP_INTERNAL(MaskedGatherINTEL, internal::OpMaskedGatherINTEL)
Expand Down
5 changes: 5 additions & 0 deletions lib/SPIRV/libSPIRV/spirv_internal.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,7 @@ enum InternalOp {
IOpMaskedGatherINTEL = 6428,
IOpMaskedScatterINTEL = 6429,
IOpJointMatrixGetElementCoordINTEL = 6440,
IOpCooperativeMatrixApplyFunctionINTEL = 6448,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: Reorder to make 6448 come first?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

fixed

IOpCooperativeMatrixPrefetchINTEL = 6449,
IOpPrev = OpMax - 2,
IOpForward
Expand Down Expand Up @@ -104,6 +105,7 @@ enum InternalCapability {
ICapabilityTensorFloat32RoundingINTEL = 6425,
ICapabilityMaskedGatherScatterINTEL = 6427,
ICapabilityJointMatrixWIInstructionsINTEL = 6435,
asudarsa marked this conversation as resolved.
Show resolved Hide resolved
ICapabilityCooperativeMatrixInvocationInstructionsINTEL = 6435,
asudarsa marked this conversation as resolved.
Show resolved Hide resolved
ICapabilityJointMatrixTF32ComponentTypeINTEL = 6436,
ICapabilityJointMatrixBF16ComponentTypeINTEL = 6437,
ICapabilityJointMatrixPackedInt2ComponentTypeINTEL = 6438,
Expand Down Expand Up @@ -178,6 +180,9 @@ _SPIRV_OP(Op, JointMatrixGetElementCoordINTEL)
_SPIRV_OP(Capability, CooperativeMatrixPrefetchINTEL)
_SPIRV_OP(Op, CooperativeMatrixPrefetchINTEL)

_SPIRV_OP(Capability, CooperativeMatrixInvocationInstructionsINTEL)
_SPIRV_OP(Op, CooperativeMatrixApplyFunctionINTEL)

_SPIRV_OP(Capability, HWThreadQueryINTEL)
_SPIRV_OP(BuiltIn, SubDeviceIDINTEL)
_SPIRV_OP(BuiltIn, GlobalHWThreadIDINTEL)
Expand Down
149 changes: 149 additions & 0 deletions test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_apply.ll
View file
Open in desktop
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Wouldn't it be good to minimize test case to actually containing only necessary's lines for a test?

Copy link
Contributor

@MrSidims MrSidims Nov 15, 2023 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In general I agree with this, and that is the way to add 90%+ of the tests. Yet ideally for language features we should have tests be almost end-to-end'ish (or Integration'ish), see test/transcoding/*.cl tests, unfortunately sycl is not yet fully unstreamed to llvm.org. So the closest thing until then is to generate IR from some sycl test and use it as an input to the translator.

Why I'm thinking, that language features should be tested like this? Practice showed up, that for example for matrices having just calls to __spirv_JointMatrixLoadINTEL(A), (B), (C) followed by __spirv_JointMatrixMadINTEL(A, B, C) tests almost nothing as the compiler will generate extra code during optimizations that will affect translation (prime example mem2reg pass inserting phi instructions selecting matrices). Counter argument to this is that such cases should be tested in the appropriate repository, where language APIs are being added, but it won't work for features under active development, as tests in those repositories won't be enabled until all components are ready. And trust me, you don't want to harry with fixes for issues coming from the real-world code, that a synthetic test doesn't cover amid release period, especially taking into the account the fact, that reverse translator changes should be adopted across consumers.

Copy link
Contributor

@MrSidims MrSidims Nov 15, 2023 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

And for this very example - I'm not trusting myself to write LLVM IR that have an external function (acting like a SPIR-V instruction), accepting a function object in a reference wrapper.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

And btw, the test is quite minimized, all metadata, attributes and other not needed stuff is removed

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In general I agree with this, and that is the way to add 90%+ of the tests. Yet ideally for language features we should have tests be almost end-to-end'ish (or Integration'ish), see test/transcoding/*.cl tests, unfortunately sycl is not yet fully unstreamed to llvm.org. So the closest thing until then is to generate IR from some sycl test and use it as an input to the translator.

And it's ok to use frontend to get IR. But the next steps from that point is to strip everything which is unneeded and blurs the essence of the test.
Minimization here have a simple goal - to separate things which are necessary to provide a functionality and nothing more. So no @llvm.assume or @llvm.lifetime.*, !0, func attr etc. - clear minimal IR needed for feature. Then, when you want to extend a feature or correct a bug one year from that point - on first sight you know what feature looked like when it was implemented and what's needed to fix.
The problem with using straight forward outputs from frontend is that after some time the implementation in frontend will change but the IR version of the test here assumes that it's unchanged - frozen in time. Recompilation of a test issue from frontend will be impossible (because file doesn't exist or is renamed) or it will yield different IR. Is it test still useful and correct or it's not? Clear to read? This is what happened with FPGA_memory_attributes tests - which were implemented few years ago.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We could run opt -S -passes=metarenamer < reduced.ll > reduced2.ll on it in order to get rid off of burdened names like %arrayidx.i29.i.i.i.i.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It will change, but it doesn't mean, that the current output should be untested. And at this point of time for this very feature the only place, where it can be tested is llvm-spirv.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I just think that the feature test should be focused only on thing which is implemented feature - nothing more and nothing less. And implemented code should be tested thoroughly. IMO. If the whole sequence of features need to be tested and for singular frontend then it's some kind of integration test and should be treated differently than typical feature test.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

opt -S -passes=metarenamer < reduced.ll > reduced2.ll

this actually renames also useful things, like functions and structures

Original file line number Diff line number Diff line change
@@ -0,0 +1,149 @@
;; compiled from joint_matrix_apply_bf16.cpp from intel/llvm with some modifications

; RUN: llvm-as < %s -o %t.bc
; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix,+SPV_INTEL_joint_matrix -o %t.spv
MrSidims marked this conversation as resolved.
Show resolved Hide resolved
; RUN: llvm-spirv %t.spv -to-text -o %t.spt
; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV

; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM

; CHECK-SPIRV-DAG: Capability CooperativeMatrixKHR
; CHECK-SPIRV-DAG: Capability CooperativeMatrixInvocationInstructionsINTEL
; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"
; CHECK-SPIRV-DAG: Extension "SPV_KHR_cooperative_matrix"
; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy:]]
; CHECK-SPIRV: CompositeConstruct [[#MatTy]] [[#Mat:]]
; CHECK-SPIRV: PtrCastToGeneric [[#]] [[#Ptr:]] [[#]]
; CHECK-SPIRV: CooperativeMatrixApplyFunctionINTEL [[#MatTy]] [[#Apply:]] [[#Ptr]] [[#Mat]]
; CHECK-SPIRV: CooperativeMatrixStoreKHR [[#]] [[#Apply]]

; CHECK-LLVM: %[[Mat:[%0-9a-z.]+]] = call spir_func target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) @"_Z26__spirv_CompositeConstructP38class.sycl::_V1::ext::oneapi::bfloat16"
; CHECK-LLVM: %[[Apply:[%0-9a-z.]+]] = call spir_func target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) @"_Z43__spirv_CooperativeMatrixApplyFunctionINTELPU3AS477class.sycl::_V1::ext::oneapi::experimental::matrix::helper::reference_wrapperPU3AS144__spirv_CooperativeMatrixKHR__short_8_16_0_0"(ptr addrspace(4) %ref.tmp.ascast.i21, target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) %[[Mat]])
; CHECK-LLVM: call spir_func void @"_Z33__spirv_CooperativeMatrixStoreKHRPU3AS138class.sycl::_V1::ext::oneapi::bfloat16PU3AS144__spirv_CooperativeMatrixKHR__short_8_16_0_0liii"(ptr addrspace(1) %{{.*}}, target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) %[[Apply]], i64 32, i32 0, i32 3, i32 0)

; ModuleID = 'matrix_apply.bc'
source_filename = "../llvm/sycl/test-e2e/Matrix/joint_matrix_apply_bf16.cpp"
target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
target triple = "spir64-unknown-unknown"

%"class.sycl::_V1::range" = type { %"class.sycl::_V1::detail::array" }
%"class.sycl::_V1::detail::array" = type { [2 x i64] }
%"class.sycl::_V1::id" = type { %"class.sycl::_V1::detail::array" }
%"class.sycl::_V1::ext::oneapi::experimental::matrix::helper::reference_wrapper" = type { ptr addrspace(4) }
%"class.sycl::_V1::ext::oneapi::bfloat16" = type { i16 }
%class.anon.0 = type <{ %"class.sycl::_V1::accessor", %class.anon, [7 x i8] }>
%"class.sycl::_V1::accessor" = type { %"class.sycl::_V1::detail::AccessorImplDevice", %union.anon }
%"class.sycl::_V1::detail::AccessorImplDevice" = type { %"class.sycl::_V1::id", %"class.sycl::_V1::range", %"class.sycl::_V1::range" }
%union.anon = type { ptr addrspace(1) }
%class.anon = type { i8 }

$_ZTSZZ17matrix_verify_addIN4sycl3_V13ext6oneapi8bfloat16ELm16ELm32EZ4mainEUlRS4_E_EvNS1_5queueER10big_matrixIT_XT0_EXT1_EERNS1_8nd_rangeILi2EEEfOT2_ENKUlRNS1_7handlerEE_clESI_EUlNS1_7nd_itemILi2EEEE_ = comdat any

@__spirv_BuiltInGlobalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
@__spirv_BuiltInLocalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32

; Function Attrs: convergent norecurse nounwind
define weak_odr dso_local spir_kernel void @_ZTSZZ17matrix_verify_addIN4sycl3_V13ext6oneapi8bfloat16ELm16ELm32EZ4mainEUlRS4_E_EvNS1_5queueER10big_matrixIT_XT0_EXT1_EERNS1_8nd_rangeILi2EEEfOT2_ENKUlRNS1_7handlerEE_clESI_EUlNS1_7nd_itemILi2EEEE_(ptr addrspace(1) noundef align 2 %_arg_accA, ptr noundef byval(%"class.sycl::_V1::range") align 8 %_arg_accA1, ptr noundef byval(%"class.sycl::_V1::range") align 8 %_arg_accA2, ptr noundef byval(%"class.sycl::_V1::id") align 8 %_arg_accA3) local_unnamed_addr {
entry:
%ref.tmp.i20 = alloca %"class.sycl::_V1::ext::oneapi::experimental::matrix::helper::reference_wrapper", align 8
%agg.tmp.i17 = alloca %"class.sycl::_V1::ext::oneapi::bfloat16", align 2
%ref.tmp6.i = alloca float, align 4
%__SYCLKernel = alloca %class.anon.0, align 8
%__SYCLKernel.ascast = addrspacecast ptr %__SYCLKernel to ptr addrspace(4)
call void @llvm.lifetime.start.p0(i64 64, ptr nonnull %__SYCLKernel)
%agg.tmp.sroa.0.sroa.0.0.copyload = load i64, ptr %_arg_accA1, align 8
%agg.tmp.sroa.0.sroa.2.0._arg_accA1.ascast.sroa_idx = getelementptr inbounds i8, ptr %_arg_accA1, i64 8
%agg.tmp.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp.sroa.0.sroa.2.0._arg_accA1.ascast.sroa_idx, align 8
%agg.tmp5.sroa.0.sroa.0.0.copyload = load i64, ptr %_arg_accA2, align 8
%agg.tmp5.sroa.0.sroa.2.0._arg_accA2.ascast.sroa_idx = getelementptr inbounds i8, ptr %_arg_accA2, i64 8
%agg.tmp5.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp5.sroa.0.sroa.2.0._arg_accA2.ascast.sroa_idx, align 8
%agg.tmp6.sroa.0.sroa.0.0.copyload = load i64, ptr %_arg_accA3, align 8
%agg.tmp6.sroa.0.sroa.2.0._arg_accA3.ascast.sroa_idx = getelementptr inbounds i8, ptr %_arg_accA3, i64 8
%agg.tmp6.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp6.sroa.0.sroa.2.0._arg_accA3.ascast.sroa_idx, align 8
%0 = getelementptr inbounds %"class.sycl::_V1::accessor", ptr %__SYCLKernel, i64 0, i32 1
store i64 %agg.tmp6.sroa.0.sroa.0.0.copyload, ptr %__SYCLKernel, align 8
%AccessRange.i.i.i.i.i = getelementptr inbounds %"class.sycl::_V1::detail::AccessorImplDevice", ptr %__SYCLKernel, i64 0, i32 1
store i64 %agg.tmp.sroa.0.sroa.0.0.copyload, ptr %AccessRange.i.i.i.i.i, align 8
%MemRange.i.i.i.i.i = getelementptr inbounds %"class.sycl::_V1::detail::AccessorImplDevice", ptr %__SYCLKernel, i64 0, i32 2
store i64 %agg.tmp5.sroa.0.sroa.0.0.copyload, ptr %MemRange.i.i.i.i.i, align 8
%arrayidx.i21.i.i.i.i = getelementptr inbounds [2 x i64], ptr %__SYCLKernel, i64 0, i64 1
store i64 %agg.tmp6.sroa.0.sroa.2.0.copyload, ptr %arrayidx.i21.i.i.i.i, align 8
%arrayidx.i25.i.i.i.i = getelementptr inbounds %"class.sycl::_V1::detail::AccessorImplDevice", ptr %__SYCLKernel, i64 0, i32 1, i32 0, i32 0, i64 1
store i64 %agg.tmp.sroa.0.sroa.2.0.copyload, ptr %arrayidx.i25.i.i.i.i, align 8
%arrayidx.i29.i.i.i.i = getelementptr inbounds %"class.sycl::_V1::detail::AccessorImplDevice", ptr %__SYCLKernel, i64 0, i32 2, i32 0, i32 0, i64 1
store i64 %agg.tmp5.sroa.0.sroa.2.0.copyload, ptr %arrayidx.i29.i.i.i.i, align 8
%mul.i6.i.i.i.i = mul i64 %agg.tmp6.sroa.0.sroa.0.0.copyload, %agg.tmp5.sroa.0.sroa.2.0.copyload
%1 = getelementptr %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %_arg_accA, i64 %mul.i6.i.i.i.i
%add.ptr.i = getelementptr %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %1, i64 %agg.tmp6.sroa.0.sroa.2.0.copyload
store ptr addrspace(1) %add.ptr.i, ptr %0, align 8
%2 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, i64 8), align 8
%3 = load i64, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, align 32
%4 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, i64 8), align 8
%5 = load i64, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, align 32
%ref.tmp6.ascast.i = addrspacecast ptr %ref.tmp6.i to ptr addrspace(4)
%cmp.i11 = icmp ult i64 %2, 2147483648
%cmp.i = icmp ult i64 %3, 2147483648
%cmp.i15 = icmp ult i64 %4, 2147483648
%sub.i = sub nsw i64 %2, %4
%cmp.i12 = icmp ult i64 %5, 2147483648
%sub5.i = sub nsw i64 %3, %5
call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %ref.tmp6.i)
store float 5.000000e+00, ptr %ref.tmp6.i, align 4
%call.i.i = call spir_func noundef zeroext i16 @__devicelib_ConvertFToBF16INTEL(ptr addrspace(4) noundef align 4 dereferenceable(4) %ref.tmp6.ascast.i)
call void @llvm.lifetime.start.p0(i64 2, ptr nonnull %agg.tmp.i17)
store i16 %call.i.i, ptr %agg.tmp.i17, align 2
%call.i18 = call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) @_Z26__spirv_CompositeConstruct(ptr noundef nonnull byval(%"class.sycl::_V1::ext::oneapi::bfloat16") align 2 %agg.tmp.i17)
call void @llvm.lifetime.end.p0(i64 2, ptr nonnull %agg.tmp.i17)
call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %ref.tmp6.i)
%lambda.i = getelementptr inbounds %class.anon.0, ptr addrspace(4) %__SYCLKernel.ascast, i64 0, i32 1
%ref.tmp.ascast.i21 = addrspacecast ptr %ref.tmp.i20 to ptr addrspace(4)
call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %ref.tmp.i20)
store ptr addrspace(4) %lambda.i, ptr %ref.tmp.i20, align 8
%call.i22 = call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) @_Z43__spirv_CooperativeMatrixApplyFunctionINTEL(ptr addrspace(4) noundef align 8 dereferenceable(8) %ref.tmp.ascast.i21, target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) noundef %call.i18)
call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %ref.tmp.i20)
%6 = load ptr addrspace(1), ptr %0, align 8
%7 = load i64, ptr %__SYCLKernel, align 8
%8 = load i64, ptr %arrayidx.i29.i.i.i.i, align 8
%mul.i6.i.i.i.i.i = mul i64 %7, %8
%9 = load i64, ptr %arrayidx.i21.i.i.i.i, align 8
%add.i7.i.i.i.i.i = add i64 %mul.i6.i.i.i.i.i, %9
%idx.neg.i.i = sub i64 0, %add.i7.i.i.i.i.i
%add.ptr.i.i = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %6, i64 %idx.neg.i.i
%mul12.i = shl nsw i64 %sub.i, 8
%add.ptr.i43 = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %add.ptr.i.i, i64 %mul12.i
%div14.i = and i64 %sub5.i, -16
%add.ptr.i44 = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %add.ptr.i43, i64 %div14.i
call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHRPU3AS4iPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3ili(ptr addrspace(1) noundef %add.ptr.i44, target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) noundef %call.i22, i64 noundef 32, i32 noundef 0, i32 noundef 3, i32 noundef 0)
call void @llvm.lifetime.end.p0(i64 64, ptr nonnull %__SYCLKernel)
ret void
}

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture)

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture)

; Function Attrs: convergent nounwind
declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) @_Z26__spirv_CompositeConstruct(ptr noundef byval(%"class.sycl::_V1::ext::oneapi::bfloat16") align 2) local_unnamed_addr

; Function Attrs: convergent nounwind
declare dso_local spir_func zeroext i16 @__devicelib_ConvertFToBF16INTEL(ptr addrspace(4) noundef align 4 dereferenceable(4)) local_unnamed_addr

; Function Attrs: convergent nounwind
declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) @_Z43__spirv_CooperativeMatrixApplyFunctionINTEL(ptr addrspace(4) noundef align 8 dereferenceable(8), target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) noundef) local_unnamed_addr

; Function Attrs: convergent nounwind
declare dso_local spir_func void @_Z33__spirv_CooperativeMatrixStoreKHRPU3AS4iPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3ili(ptr addrspace(1) noundef, target("spirv.CooperativeMatrixKHR", i16, 8, 16, 0, 0) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr

!llvm.module.flags = !{!0, !1}
!opencl.spir.version = !{!2}
!spirv.Source = !{!3}
!llvm.ident = !{!4}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{i32 7, !"frame-pointer", i32 2}
!2 = !{i32 1, i32 2}
!3 = !{i32 4, i32 100000}
!4 = !{!"clang version 18.0.0 (https://github.com/intel/llvm.git)"}
Loading