[AMDGPU][CodeGen] To improve CGEMM performance: combine LDS reads.

[alex-t]

Summary

Change explores the fact that LDS reads may be reordered even if access the same location.

Prior the change, algorithm immediately stops as soon as any memory access encountered between loads that are expected to be merged together. Although, Read-After-Read conflict cannot affect execution correctness.

Improves hcBLAS CGEMM manually loop-unrolled kernels performance by 44%. Also improvement expected on any massive sequences of reads from LDS.

[alex-t]

cgemm_loopunroll_nofix.isa13 KBDownload

cgemm_loopunroll_fix.isa14 KBDownload

These 2 files are here to illustrate the ISA before and after the fix. See BB10 to view the effect.

[• tstellarAMD]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

155–156

I see this logic repeated in a few places, can we turn it into a helper function?

218–222

Need to use c++ style comments.

281–282

This doesn't seem correct. The comment mentions that we need to check that it was safe to move all the instruction in InstsToMove down past this instruction. Is this condition being met?

test/CodeGen/AMDGPU/cgemm_loopunroll_ds_combine.ll

1

This test case is too big, can you try to reduce it. Also you should run use opt -metarenamer to simplify the names. If you can't get a reduced IR test case, I would recommend writing an MIR test case.

[rampitec]

This is good as a fast workaround. In a long run direct selection of a vector operation from DAG seems more desirable.

[alex-t]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

281–282

Yes. If this 2 mem accesses may alias but neither is write access it is safe to reorder them. In this condition we check if 2 mem accesses may alias and if at least one may write we break out. Otherwise we are free to search down until the BB ends or we face the instruction that cannot be moved.

[alex-t]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

281–282

and yes, you are rigth.
I need to check if I users collected so far can be moved across MBBI before I go ahead looking for the merge point.

[alex-t]

Fixed according the reviewers comments.
Test case reducing in progress. Not trivial because smaller cases use less registers.
We need large unroll to expose the problem.

[alex-t]

Changed source code. New test is upcoming.

[• tstellarAMD]

In D25944#579909, @alex-t wrote:

Fixed according the reviewers comments.
Test case reducing in progress. Not trivial because smaller cases use less registers.

If it's not trivial you may want to look at writing a MIR testcase. This will be easier and also be a better test. For examples look in test/CodeGen/MIR/

We need large unroll to expose the problem.

[arsenm]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

149

We don't actually have AA enabled in the backend. We also need to add an address space alias analysis pass. If these were done, would that avoid the need to have this looser check?

[alex-t]

Small test case.

[alex-t]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

149

No. Even 2 reads from same segment (address space) and exactly same address are legal to be reordered as they both have no side effects. This is Read-After-Read (RAR) conflict. The check above is necessary to allow reordering Read-After-Write and Write-After-Write in case we can prove that 2 memory operations accesses different locations.
As for the address space layer in AA, I added such in HSAIL backend. It is trivial and I can port it to LC quickly if you feel it is necessary.

[• tstellarAMD]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

149

The alias analysis passes are always run so the information is available. There is a usesAA() subtarget query, but that's only used by a few places, mostly in the DAGCobminer.

[alex-t]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

149

As far as I understand Matt meant that we'd like to have custom AA layer that intercepts alias queries with 2 different address spaces. Given that:

• flat can alias any
• constant can alias nothing
• global group and private only can alias pointer in same address space

[rampitec]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

153

As far as I understand that is only legal to reorder two instructions if there are no stores in between of them which might go into the memory of instruction being moved. Also you need to make sure there are no barriers or fences in between. This function only checks two instructions passed but not anything in between. I.e. that is only legal if these two instructions are adjacent. I cannot immediately see it is only called for adjacent instructions, neither see a comment on such limitation.

[tony-tye]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

149

Would be good to consider @jlebar patches to support target-specific AA (D24441 and D12414) which add an NVPTX-specific AA.

[alex-t]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

153

In fact all the checks is done in a loop within findMatchingDSInst

if (MBBI->hasUnmodeledSideEffects())
    // We can't re-order this instruction with respect to other memory
    // opeations, so we fail both conditions mentioned above.
    return E;

  if (MBBI->mayLoadOrStore() &&
    !memAccessesCanBeReordered(*I, *MBBI, TII, AA)) {
    // We fail condition #1, but we may still be able to satisfy condition
    // #2.  Add this instruction to the move list and then we will check
    // if condition #2 holds once we have selected the matching instruction.
    InstsToMove.push_back(&*MBBI);
    addDefsToList(*MBBI, DefsToMove);
    continue;
  }

So any writes in between are in InstsToMove list and will be checked in canMoveInstsAcrossMemOp. Any instructions in between with side effects like barriers etc breaks search.
I had no intention to make a neat fix - just temporary workaround. So I preserved basic logic.
Indeed there is no need to put potential stores in the move list for later check. We can break out immediately faced mayAlias store in between.
If you insist I can change to this.

[rampitec]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

153

Thanks. LGTM as a w/a provided tests are passed.

[arsenm]

test/CodeGen/AMDGPU/ds_read2.ll

496–498 ↗

(On Diff #76023)

You should run instnamer on the test. Also positive checks are more useful

[alex-t]

Test case passed through the opt -instrenamer

[alex-t]

test/CodeGen/AMDGPU/ds_read2.ll

496–498 ↗

(On Diff #76023)

The goal of the test is to check that all LDS reads are combined.
So, any "ds_read_b32" in the output is the regression. In the original CGEMM inner loop there were 64 ds_read_b32 and all of them are combined with the patch. The test case is same but shorter excerpt. That's why it is negative.

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

static void addDefsToList(const MachineInstr &MI,
   }
 }
 
 static bool
 canMoveInstsAcrossMemOp(MachineInstr &MemOp,
                         ArrayRef<MachineInstr*> InstsToMove,
                         const SIInstrInfo *TII,
                         AliasAnalysis *AA) {
 
   assert(MemOp.mayLoadOrStore());
 
   for (MachineInstr *InstToMove : InstsToMove) {
     if (!InstToMove->mayLoadOrStore())
       continue;
-    if (!TII->areMemAccessesTriviallyDisjoint(MemOp, *InstToMove, AA))
+    if (!TII->areMemAccessesTriviallyDisjoint(MemOp, *InstToMove, AA)) {
+      if (MemOp.mayStore() || InstToMove->mayStore())
         return false;
     }
+  }
   return true;
 }
 
 bool SILoadStoreOptimizer::offsetsCanBeCombined(unsigned Offset0,
                                                 unsigned Offset1,
                                                 unsigned Size) {
   // XXX - Would the same offset be OK? Is there any reason this would happen or
   // be useful?
if (MBBI->getOpcode() != I->getOpcode()) {
       //    be merged into.
 
       if (MBBI->hasUnmodeledSideEffects())
         // We can't re-order this instruction with respect to other memory
         // opeations, so we fail both conditions mentioned above.
         return E;
 
       if (MBBI->mayLoadOrStore() &&
-          !TII->areMemAccessesTriviallyDisjoint(*I, *MBBI, AA)) {
+        /*
+           RAW or WAR - cannot reorder
+           WAW - cannot reorder
+           RAR - safe to reorder
+        */
+        if (MBBI->mayStore() || I->mayStore()) {
           // We fail condition #1, but we may still be able to satisfy condition
           // #2.  Add this instruction to the move list and then we will check
           // if condition #2 holds once we have selected the matching instruction.
           InstsToMove.push_back(&*MBBI);
           addDefsToList(*MBBI, DefsToMove);
+        }
         continue;
       }
 
       // When we match I with another DS instruction we will be moving I down
       // to the location of the matched instruction any uses of I will need to
       // be moved down as well.
       for (const MachineOperand *Def : DefsToMove) {
         bool ReadDef = MBBI->readsVirtualRegister(Def->getReg());
if (AddrReg0.getReg() == AddrReg1.getReg() &&
         return MBBI;
     }
 
     // We've found a load/store that we couldn't merge for some reason.
     // We could potentially keep looking, but we'd need to make sure that
     // it was safe to move I and also all the instruction in InstsToMove
     // down past this instruction.
     // FIXME: This is too conservative.
+    if (!TII->areMemAccessesTriviallyDisjoint(*I, *MBBI, AA) &&
+     (I->mayStore() || MBBI->mayStore()))
       break;
   }
   return E;
 }
 
 MachineBasicBlock::iterator  SILoadStoreOptimizer::mergeRead2Pair(
   MachineBasicBlock::iterator I,
   MachineBasicBlock::iterator Paired,
   unsigned EltSize,

test/CodeGen/AMDGPU/cgemm_loopunroll_ds_combine.ll

+; RUN: llc -O2 -mtriple amdgcn--amdhsa -mcpu=fiji -filetype=asm < %s | FileCheck -check-prefix=SI %s
+; SI-LABEL: _ZZ26cgemm_NoTransAB_loopunrollRN2
+; SI-NOT: ds_read_b32
+
+; ModuleID = 'fiji.opt.bc'
+source_filename = "llvm-link"
+target datalayout = "e-p:32:32-p1:64:64-p2:64:64-p3:32:32-p4:64:64-p5:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64"
+target triple = "amdgcn--amdhsa"
+
+; Function Attrs: nounwind
+declare void @llvm.amdgcn.s.waitcnt(i32) #2
+
+; Function Attrs: nounwind
+declare void @llvm.amdgcn.s.dcache.wb() #2
+
+; Function Attrs: convergent nounwind
+declare void @llvm.amdgcn.s.barrier() #3
+
+; Function Attrs: nounwind readnone
+declare float @llvm.fabs.f32(float) #4
+
+; Function Attrs: nounwind readnone
+declare i1 @llvm.amdgcn.class.f32(float, i32) #4
+
+; Function Attrs: nounwind readnone
+declare i1 @llvm.amdgcn.class.f64(double, i32) #4
+
+; Function Attrs: nounwind readnone
+declare i8 addrspace(2)* @llvm.amdgcn.dispatch.ptr() #4
+
+; Function Attrs: nounwind readnone
+declare i32 @llvm.amdgcn.workitem.id.x() #4
+
+; Function Attrs: nounwind readnone
+declare i32 @llvm.amdgcn.workgroup.id.x() #4
+
+; Function Attrs: nounwind readnone
+declare i8 addrspace(2)* @llvm.amdgcn.implicitarg.ptr() #4
+
+; Function Attrs: nounwind readnone
+declare i32 @llvm.amdgcn.workitem.id.y() #4
+
+; Function Attrs: nounwind readnone
+declare i32 @llvm.amdgcn.workgroup.id.y() #4
+
+; Function Attrs: nounwind readnone
+declare i32 @llvm.amdgcn.workgroup.id.z() #4
+
+%"class.hc::short_vector::float_2.0" = type { float, float }
+
+@"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2" = internal unnamed_addr addrspace(3) global [16 x [16 x float]] undef, align 4
+
+@"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3" = internal unnamed_addr addrspace(3) global [16 x [16 x float]] undef, align 4
+
+@"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4" = internal unnamed_addr addrspace(3) global [16 x [16 x float]] undef, align 4
+
+@"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5" = internal unnamed_addr addrspace(3) global [16 x [16 x float]] undef, align 4
+
+define amdgpu_kernel void @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff"(i32, i32, %"class.hc::short_vector::float_2.0" addrspace(1)* nocapture readonly, i64, i32, i32, %"class.hc::short_vector::float_2.0" addrspace(1)* nocapture readonly, i64, i32, %"class.hc::short_vector::float_2.0" addrspace(1)* nocapture, i64, i32, float, float, float, float) local_unnamed_addr #0 align 2 {
+  %17 = tail call i8 addrspace(2)* @llvm.amdgcn.dispatch.ptr() #4
+  %18 = tail call i32 @llvm.amdgcn.workitem.id.y() #4
+  %19 = tail call i32 @llvm.amdgcn.workgroup.id.y() #4
+  %20 = getelementptr inbounds i8, i8 addrspace(2)* %17, i64 6
+  %21 = bitcast i8 addrspace(2)* %20 to i16 addrspace(2)*
+  %22 = load i16, i16 addrspace(2)* %21, align 2, !tbaa !232
+  %23 = zext i16 %22 to i32
+  %24 = mul i32 %23, %19
+  %25 = add i32 %24, %18
+  %26 = tail call i8 addrspace(2)* @llvm.amdgcn.implicitarg.ptr() #4
+  %27 = getelementptr inbounds i8, i8 addrspace(2)* %26, i64 8
+  %28 = zext i32 %25 to i64
+  %29 = bitcast i8 addrspace(2)* %27 to i64 addrspace(2)*
+  %30 = load i64, i64 addrspace(2)* %29, align 8, !tbaa !241
+  %31 = add i64 %28, %30
+  %32 = tail call i32 @llvm.amdgcn.workitem.id.x() #4
+  %33 = tail call i32 @llvm.amdgcn.workgroup.id.x() #4
+  %34 = getelementptr inbounds i8, i8 addrspace(2)* %17, i64 4
+  %35 = bitcast i8 addrspace(2)* %34 to i16 addrspace(2)*
+  %36 = load i16, i16 addrspace(2)* %35, align 4, !tbaa !242
+  %37 = zext i16 %36 to i32
+  %38 = mul i32 %37, %33
+  %39 = add i32 %38, %32
+  %40 = zext i32 %39 to i64
+  %41 = bitcast i8 addrspace(2)* %26 to i64 addrspace(2)*
+  %42 = load i64, i64 addrspace(2)* %41, align 8, !tbaa !241
+  %conv.i.i.i = trunc i64 %31 to i32
+  %mul.i = shl nsw i32 %19, 4
+  %add.i = add nsw i32 %mul.i, %18
+  %mul5.i = shl nsw i32 %33, 4
+  %add8.i = add nsw i32 %mul5.i, %32
+  %cmp1281.i = icmp sgt i32 %0, 0
+  br i1 %cmp1281.i, label %.lr.ph.i, label %._crit_edge.i
+
+.lr.ph.i:                                         ; preds = %16
+  %conv781288.i = zext i32 %add8.i to i64
+  %cmp15.i = icmp slt i32 %add.i, %1
+  %mul16.i = mul nsw i32 %add.i, %4
+  %conv1115.i = zext i32 %mul16.i to i64
+  %add17.i = add i64 %conv1115.i, %3
+  %43 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 %32
+  %44 = bitcast float addrspace(3)* %43 to i32 addrspace(3)*
+  %45 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 %32
+  %cmp70.i = icmp slt i32 %add8.i, %5
+  %add77.i = add i64 %conv781288.i, %7
+  %46 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 %18, i32 %32
+  %47 = bitcast float addrspace(3)* %46 to i32 addrspace(3)*
+  %48 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 %18, i32 %32
+  %49 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 0
+  %50 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 0, i32 %32
+  %51 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 0
+  %52 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 0, i32 %32
+  %53 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 1
+  %54 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 1, i32 %32
+  %55 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 1
+  %56 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 1, i32 %32
+  %57 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 2
+  %58 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 2, i32 %32
+  %59 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 2
+  %60 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 2, i32 %32
+  %61 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 3
+  %62 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 3, i32 %32
+  %63 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 3
+  %64 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 3, i32 %32
+  %65 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 4
+  %66 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 4, i32 %32
+  %67 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 4
+  %68 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 4, i32 %32
+  %69 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 5
+  %70 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 5, i32 %32
+  %71 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 5
+  %72 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 5, i32 %32
+  %73 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 6
+  %74 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 6, i32 %32
+  %75 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 6
+  %76 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 6, i32 %32
+  %77 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 7
+  %78 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 7, i32 %32
+  %79 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 7
+  %80 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 7, i32 %32
+  %81 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 8
+  %82 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 8, i32 %32
+  %83 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 8
+  %84 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 8, i32 %32
+  %85 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 9
+  %86 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 9, i32 %32
+  %87 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 9
+  %88 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 9, i32 %32
+  %89 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 10
+  %90 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 10, i32 %32
+  %91 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 10
+  %92 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 10, i32 %32
+  %93 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 11
+  %94 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 11, i32 %32
+  %95 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 11
+  %96 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 11, i32 %32
+  %97 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 12
+  %98 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 12, i32 %32
+  %99 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 12
+  %100 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 12, i32 %32
+  %101 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 13
+  %102 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 13, i32 %32
+  %103 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 13
+  %104 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 13, i32 %32
+  %105 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 14
+  %106 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 14, i32 %32
+  %107 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 14
+  %108 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 14, i32 %32
+  %109 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Bsreal.2", i32 0, i32 %18, i32 15
+  %110 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE6Asreal.4", i32 0, i32 15, i32 %32
+  %111 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Bsimg.3", i32 0, i32 %18, i32 15
+  %112 = getelementptr [16 x [16 x float]], [16 x [16 x float]] addrspace(3)* @"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_EN3$_019__cxxamp_trampolineEiiS4_liiS4_liS4_liffff._ZZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEEE5Asimg.5", i32 0, i32 15, i32 %32
+  %add9.i = add nuw nsw i32 %0, 15
+  %div.i = sdiv i32 %add9.i, 16
+  br label %._crit_edge1286.i
+
+._crit_edge.i.loopexit:                           ; preds = %130
+  br label %._crit_edge.i
+
+._crit_edge.i:                                    ; preds = %._crit_edge.i.loopexit, %16
+  %CValue1.0.lcssa.i = phi float [ 0.000000e+00, %16 ], [ %add827.i, %._crit_edge.i.loopexit ]
+  %CValue.0.lcssa.i = phi float [ 0.000000e+00, %16 ], [ %add475.i, %._crit_edge.i.loopexit ]
+  %cmp829.i = icmp slt i32 %add.i, %1
+  %cmp830.i = icmp slt i32 %add8.i, %5
+  %or.cond34 = and i1 %cmp829.i, %cmp830.i
+  br i1 %or.cond34, label %195, label %"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEE.exit"
+
+._crit_edge1286.i:                                ; preds = %130, %.lr.ph.i
+  %k.01284.i = phi i32 [ 0, %.lr.ph.i ], [ %inc.i, %130 ]
+  %CValue.01283.i = phi float [ 0.000000e+00, %.lr.ph.i ], [ %add475.i, %130 ]
+  %CValue1.01282.i = phi float [ 0.000000e+00, %.lr.ph.i ], [ %add827.i, %130 ]
+  %mul10.i = shl nsw i32 %k.01284.i, 4
+  %add13.i = add nsw i32 %mul10.i, %32
+  %cmp14.i = icmp slt i32 %add13.i, %0
+  %or.cond = and i1 %cmp15.i, %cmp14.i
+  br i1 %or.cond, label %113, label %120
+
+; <label>:113:                                    ; preds = %._crit_edge1286.i
+  %conv221116.i = zext i32 %add13.i to i64
+  %add23.i = add i64 %add17.i, %conv221116.i
+  %sext4 = shl i64 %add23.i, 32
+  %114 = ashr exact i64 %sext4, 32
+  %115 = getelementptr %"class.hc::short_vector::float_2.0", %"class.hc::short_vector::float_2.0" addrspace(1)* %2, i64 %114, i32 0
+  %116 = bitcast float addrspace(1)* %115 to i32 addrspace(1)*
+  %117 = load i32, i32 addrspace(1)* %116, align 4, !tbaa !243
+  store i32 %117, i32 addrspace(3)* %44, align 4, !tbaa !248
+  %118 = getelementptr %"class.hc::short_vector::float_2.0", %"class.hc::short_vector::float_2.0" addrspace(1)* %2, i64 %114, i32 1
+  %119 = load float, float addrspace(1)* %118, align 4, !tbaa !249
+  br label %121
+
+; <label>:120:                                    ; preds = %._crit_edge1286.i
+  store float 0.000000e+00, float addrspace(3)* %43, align 4, !tbaa !248
+  br label %121
+
+; <label>:121:                                    ; preds = %120, %113
+  %.sink.i = phi float [ 0.000000e+00, %120 ], [ %119, %113 ]
+  store float %.sink.i, float addrspace(3)* %45, align 4, !tbaa !248
+  %add68.i = add nsw i32 %mul10.i, %18
+  %cmp69.i = icmp slt i32 %add68.i, %0
+  %or.cond33 = and i1 %cmp70.i, %cmp69.i
+  br i1 %or.cond33, label %122, label %129
+
+; <label>:122:                                    ; preds = %121
+  %mul75.i = mul nsw i32 %add68.i, %8
+  %conv761113.i = zext i32 %mul75.i to i64
+  %add79.i = add i64 %add77.i, %conv761113.i
+  %sext2 = shl i64 %add79.i, 32
+  %123 = ashr exact i64 %sext2, 32
+  %124 = getelementptr %"class.hc::short_vector::float_2.0", %"class.hc::short_vector::float_2.0" addrspace(1)* %6, i64 %123, i32 0
+  %125 = bitcast float addrspace(1)* %124 to i32 addrspace(1)*
+  %126 = load i32, i32 addrspace(1)* %125, align 4, !tbaa !243
+  store i32 %126, i32 addrspace(3)* %47, align 4, !tbaa !248
+  %127 = getelementptr %"class.hc::short_vector::float_2.0", %"class.hc::short_vector::float_2.0" addrspace(1)* %6, i64 %123, i32 1
+  %128 = load float, float addrspace(1)* %127, align 4, !tbaa !249
+  br label %130
+
+; <label>:129:                                    ; preds = %121
+  store float 0.000000e+00, float addrspace(3)* %46, align 4, !tbaa !248
+  br label %130
+
+; <label>:130:                                    ; preds = %129, %122
+  %.sink905.i = phi float [ 0.000000e+00, %129 ], [ %128, %122 ]
+  store float %.sink905.i, float addrspace(3)* %48, align 4, !tbaa !248
+  tail call void @llvm.amdgcn.s.waitcnt(i32 0) #2
+  tail call void @llvm.amdgcn.s.dcache.wb() #2
+  tail call void @llvm.amdgcn.s.barrier() #2
+  %131 = load float, float addrspace(3)* %49, align 4, !tbaa !248
+  %132 = load float, float addrspace(3)* %50, align 4, !tbaa !248
+  %mul134.i = fmul float %131, %132
+  %133 = load float, float addrspace(3)* %51, align 4, !tbaa !248
+  %134 = load float, float addrspace(3)* %52, align 4, !tbaa !248
+  %mul144.i = fmul float %133, %134
+  %sub.i = fsub float %mul134.i, %mul144.i
+  %135 = load float, float addrspace(3)* %53, align 4, !tbaa !248
+  %136 = load float, float addrspace(3)* %54, align 4, !tbaa !248
+  %mul154.i = fmul float %135, %136
+  %add155.i = fadd float %sub.i, %mul154.i
+  %137 = load float, float addrspace(3)* %55, align 4, !tbaa !248
+  %138 = load float, float addrspace(3)* %56, align 4, !tbaa !248
+  %mul165.i = fmul float %137, %138
+  %sub166.i = fsub float %add155.i, %mul165.i
+  %139 = load float, float addrspace(3)* %57, align 4, !tbaa !248
+  %140 = load float, float addrspace(3)* %58, align 4, !tbaa !248
+  %mul176.i = fmul float %139, %140
+  %add177.i = fadd float %sub166.i, %mul176.i
+  %141 = load float, float addrspace(3)* %59, align 4, !tbaa !248
+  %142 = load float, float addrspace(3)* %60, align 4, !tbaa !248
+  %mul187.i = fmul float %141, %142
+  %sub188.i = fsub float %add177.i, %mul187.i
+  %143 = load float, float addrspace(3)* %61, align 4, !tbaa !248
+  %144 = load float, float addrspace(3)* %62, align 4, !tbaa !248
+  %mul198.i = fmul float %143, %144
+  %add199.i = fadd float %sub188.i, %mul198.i
+  %145 = load float, float addrspace(3)* %63, align 4, !tbaa !248
+  %146 = load float, float addrspace(3)* %64, align 4, !tbaa !248
+  %mul209.i = fmul float %145, %146
+  %sub210.i = fsub float %add199.i, %mul209.i
+  %147 = load float, float addrspace(3)* %65, align 4, !tbaa !248
+  %148 = load float, float addrspace(3)* %66, align 4, !tbaa !248
+  %mul220.i = fmul float %147, %148
+  %add221.i = fadd float %sub210.i, %mul220.i
+  %149 = load float, float addrspace(3)* %67, align 4, !tbaa !248
+  %150 = load float, float addrspace(3)* %68, align 4, !tbaa !248
+  %mul231.i = fmul float %149, %150
+  %sub232.i = fsub float %add221.i, %mul231.i
+  %151 = load float, float addrspace(3)* %69, align 4, !tbaa !248
+  %152 = load float, float addrspace(3)* %70, align 4, !tbaa !248
+  %mul242.i = fmul float %151, %152
+  %add243.i = fadd float %sub232.i, %mul242.i
+  %153 = load float, float addrspace(3)* %71, align 4, !tbaa !248
+  %154 = load float, float addrspace(3)* %72, align 4, !tbaa !248
+  %mul253.i = fmul float %153, %154
+  %sub254.i = fsub float %add243.i, %mul253.i
+  %155 = load float, float addrspace(3)* %73, align 4, !tbaa !248
+  %156 = load float, float addrspace(3)* %74, align 4, !tbaa !248
+  %mul264.i = fmul float %155, %156
+  %add265.i = fadd float %sub254.i, %mul264.i
+  %157 = load float, float addrspace(3)* %75, align 4, !tbaa !248
+  %158 = load float, float addrspace(3)* %76, align 4, !tbaa !248
+  %mul275.i = fmul float %157, %158
+  %sub276.i = fsub float %add265.i, %mul275.i
+  %159 = load float, float addrspace(3)* %77, align 4, !tbaa !248
+  %160 = load float, float addrspace(3)* %78, align 4, !tbaa !248
+  %mul286.i = fmul float %159, %160
+  %add287.i = fadd float %sub276.i, %mul286.i
+  %161 = load float, float addrspace(3)* %79, align 4, !tbaa !248
+  %162 = load float, float addrspace(3)* %80, align 4, !tbaa !248
+  %mul297.i = fmul float %161, %162
+  %sub298.i = fsub float %add287.i, %mul297.i
+  %163 = load float, float addrspace(3)* %81, align 4, !tbaa !248
+  %164 = load float, float addrspace(3)* %82, align 4, !tbaa !248
+  %mul308.i = fmul float %163, %164
+  %add309.i = fadd float %sub298.i, %mul308.i
+  %165 = load float, float addrspace(3)* %83, align 4, !tbaa !248
+  %166 = load float, float addrspace(3)* %84, align 4, !tbaa !248
+  %mul319.i = fmul float %165, %166
+  %sub320.i = fsub float %add309.i, %mul319.i
+  %167 = load float, float addrspace(3)* %85, align 4, !tbaa !248
+  %168 = load float, float addrspace(3)* %86, align 4, !tbaa !248
+  %mul330.i = fmul float %167, %168
+  %add331.i = fadd float %sub320.i, %mul330.i
+  %169 = load float, float addrspace(3)* %87, align 4, !tbaa !248
+  %170 = load float, float addrspace(3)* %88, align 4, !tbaa !248
+  %mul341.i = fmul float %169, %170
+  %sub342.i = fsub float %add331.i, %mul341.i
+  %171 = load float, float addrspace(3)* %89, align 4, !tbaa !248
+  %172 = load float, float addrspace(3)* %90, align 4, !tbaa !248
+  %mul352.i = fmul float %171, %172
+  %add353.i = fadd float %sub342.i, %mul352.i
+  %173 = load float, float addrspace(3)* %91, align 4, !tbaa !248
+  %174 = load float, float addrspace(3)* %92, align 4, !tbaa !248
+  %mul363.i = fmul float %173, %174
+  %sub364.i = fsub float %add353.i, %mul363.i
+  %175 = load float, float addrspace(3)* %93, align 4, !tbaa !248
+  %176 = load float, float addrspace(3)* %94, align 4, !tbaa !248
+  %mul374.i = fmul float %175, %176
+  %add375.i = fadd float %sub364.i, %mul374.i
+  %177 = load float, float addrspace(3)* %95, align 4, !tbaa !248
+  %178 = load float, float addrspace(3)* %96, align 4, !tbaa !248
+  %mul385.i = fmul float %177, %178
+  %sub386.i = fsub float %add375.i, %mul385.i
+  %179 = load float, float addrspace(3)* %97, align 4, !tbaa !248
+  %180 = load float, float addrspace(3)* %98, align 4, !tbaa !248
+  %mul396.i = fmul float %179, %180
+  %add397.i = fadd float %sub386.i, %mul396.i
+  %181 = load float, float addrspace(3)* %99, align 4, !tbaa !248
+  %182 = load float, float addrspace(3)* %100, align 4, !tbaa !248
+  %mul407.i = fmul float %181, %182
+  %sub408.i = fsub float %add397.i, %mul407.i
+  %183 = load float, float addrspace(3)* %101, align 4, !tbaa !248
+  %184 = load float, float addrspace(3)* %102, align 4, !tbaa !248
+  %mul418.i = fmul float %183, %184
+  %add419.i = fadd float %sub408.i, %mul418.i
+  %185 = load float, float addrspace(3)* %103, align 4, !tbaa !248
+  %186 = load float, float addrspace(3)* %104, align 4, !tbaa !248
+  %mul429.i = fmul float %185, %186
+  %sub430.i = fsub float %add419.i, %mul429.i
+  %187 = load float, float addrspace(3)* %105, align 4, !tbaa !248
+  %188 = load float, float addrspace(3)* %106, align 4, !tbaa !248
+  %mul440.i = fmul float %187, %188
+  %add441.i = fadd float %sub430.i, %mul440.i
+  %189 = load float, float addrspace(3)* %107, align 4, !tbaa !248
+  %190 = load float, float addrspace(3)* %108, align 4, !tbaa !248
+  %mul451.i = fmul float %189, %190
+  %sub452.i = fsub float %add441.i, %mul451.i
+  %191 = load float, float addrspace(3)* %109, align 4, !tbaa !248
+  %192 = load float, float addrspace(3)* %110, align 4, !tbaa !248
+  %mul462.i = fmul float %191, %192
+  %add463.i = fadd float %sub452.i, %mul462.i
+  %193 = load float, float addrspace(3)* %111, align 4, !tbaa !248
+  %194 = load float, float addrspace(3)* %112, align 4, !tbaa !248
+  %mul473.i = fmul float %193, %194
+  %sub474.i = fsub float %add463.i, %mul473.i
+  %add475.i = fadd float %CValue.01283.i, %sub474.i
+  %mul485.i = fmul float %131, %134
+  %mul495.i = fmul float %132, %133
+  %add496.i = fadd float %mul495.i, %mul485.i
+  %mul506.i = fmul float %135, %138
+  %add507.i = fadd float %add496.i, %mul506.i
+  %mul517.i = fmul float %136, %137
+  %add518.i = fadd float %mul517.i, %add507.i
+  %mul528.i = fmul float %139, %142
+  %add529.i = fadd float %add518.i, %mul528.i
+  %mul539.i = fmul float %140, %141
+  %add540.i = fadd float %mul539.i, %add529.i
+  %mul550.i = fmul float %143, %146
+  %add551.i = fadd float %add540.i, %mul550.i
+  %mul561.i = fmul float %144, %145
+  %add562.i = fadd float %mul561.i, %add551.i
+  %mul572.i = fmul float %147, %150
+  %add573.i = fadd float %add562.i, %mul572.i
+  %mul583.i = fmul float %148, %149
+  %add584.i = fadd float %mul583.i, %add573.i
+  %mul594.i = fmul float %151, %154
+  %add595.i = fadd float %add584.i, %mul594.i
+  %mul605.i = fmul float %152, %153
+  %add606.i = fadd float %mul605.i, %add595.i
+  %mul616.i = fmul float %155, %158
+  %add617.i = fadd float %add606.i, %mul616.i
+  %mul627.i = fmul float %156, %157
+  %add628.i = fadd float %mul627.i, %add617.i
+  %mul638.i = fmul float %159, %162
+  %add639.i = fadd float %add628.i, %mul638.i
+  %mul649.i = fmul float %160, %161
+  %add650.i = fadd float %mul649.i, %add639.i
+  %mul660.i = fmul float %163, %166
+  %add661.i = fadd float %add650.i, %mul660.i
+  %mul671.i = fmul float %164, %165
+  %add672.i = fadd float %mul671.i, %add661.i
+  %mul682.i = fmul float %167, %170
+  %add683.i = fadd float %add672.i, %mul682.i
+  %mul693.i = fmul float %168, %169
+  %add694.i = fadd float %mul693.i, %add683.i
+  %mul704.i = fmul float %171, %174
+  %add705.i = fadd float %add694.i, %mul704.i
+  %mul715.i = fmul float %172, %173
+  %add716.i = fadd float %mul715.i, %add705.i
+  %mul726.i = fmul float %175, %178
+  %add727.i = fadd float %add716.i, %mul726.i
+  %mul737.i = fmul float %176, %177
+  %add738.i = fadd float %mul737.i, %add727.i
+  %mul748.i = fmul float %179, %182
+  %add749.i = fadd float %add738.i, %mul748.i
+  %mul759.i = fmul float %180, %181
+  %add760.i = fadd float %mul759.i, %add749.i
+  %mul770.i = fmul float %183, %186
+  %add771.i = fadd float %add760.i, %mul770.i
+  %mul781.i = fmul float %184, %185
+  %add782.i = fadd float %mul781.i, %add771.i
+  %mul792.i = fmul float %187, %190
+  %add793.i = fadd float %add782.i, %mul792.i
+  %mul803.i = fmul float %188, %189
+  %add804.i = fadd float %mul803.i, %add793.i
+  %mul814.i = fmul float %191, %194
+  %add815.i = fadd float %add804.i, %mul814.i
+  %mul825.i = fmul float %192, %193
+  %add826.i = fadd float %mul825.i, %add815.i
+  %add827.i = fadd float %CValue1.01282.i, %add826.i
+  tail call void @llvm.amdgcn.s.waitcnt(i32 0) #2
+  tail call void @llvm.amdgcn.s.dcache.wb() #2
+  tail call void @llvm.amdgcn.s.barrier() #2
+  %inc.i = add nuw nsw i32 %k.01284.i, 1
+  %cmp.i = icmp slt i32 %inc.i, %div.i
+  br i1 %cmp.i, label %._crit_edge1286.i, label %._crit_edge.i.loopexit
+
+; <label>:195:                                    ; preds = %._crit_edge.i
+  %mul832.i = mul nsw i32 %conv.i.i.i, %11
+  %conv8331105.i = zext i32 %mul832.i to i64
+  %196 = add i64 %42, %10
+  %add834.i = add i64 %196, %conv8331105.i
+  %add838.i = add i64 %add834.i, %40
+  %sext = shl i64 %add838.i, 32
+  %197 = ashr exact i64 %sext, 32
+  %198 = getelementptr %"class.hc::short_vector::float_2.0", %"class.hc::short_vector::float_2.0" addrspace(1)* %9, i64 %197, i32 0
+  %199 = load float, float addrspace(1)* %198, align 4, !tbaa !243
+  %200 = getelementptr %"class.hc::short_vector::float_2.0", %"class.hc::short_vector::float_2.0" addrspace(1)* %9, i64 %197, i32 1
+  %201 = load float, float addrspace(1)* %200, align 4, !tbaa !249
+  %202 = fcmp uno float %199, 0.000000e+00
+  br i1 %202, label %205, label %203
+
+; <label>:203:                                    ; preds = %195
+  %204 = tail call zeroext i1 @llvm.amdgcn.class.f32(float %199, i32 516) #4
+  %..i = select i1 %204, float 0.000000e+00, float %199
+  br label %205
+
+; <label>:205:                                    ; preds = %195, %203
+  %cond.i = phi float [ 0.000000e+00, %195 ], [ %..i, %203 ]
+  %206 = fcmp uno float %201, 0.000000e+00
+  br i1 %206, label %209, label %207
+
+; <label>:207:                                    ; preds = %205
+  %208 = tail call zeroext i1 @llvm.amdgcn.class.f32(float %201, i32 516) #4
+  %.907.i = select i1 %208, float 0.000000e+00, float %201
+  br label %209
+
+; <label>:209:                                    ; preds = %205, %207
+  %cond859.i = phi float [ 0.000000e+00, %205 ], [ %.907.i, %207 ]
+  %mul861.i = fmul float %cond.i, %12
+  %mul863.i = fmul float %cond859.i, %13
+  %sub864.i = fsub float %mul861.i, %mul863.i
+  %mul866.i = fmul float %cond.i, %13
+  %mul868.i = fmul float %cond859.i, %12
+  %add869.i = fadd float %mul866.i, %mul868.i
+  %mul871.i = fmul float %CValue.0.lcssa.i, %14
+  %mul873.i = fmul float %CValue1.0.lcssa.i, %15
+  %sub874.i = fsub float %mul871.i, %mul873.i
+  %add875.i = fadd float %sub874.i, %sub864.i
+  store float %add875.i, float addrspace(1)* %198, align 4, !tbaa !243
+  %mul888.i = fmul float %CValue.0.lcssa.i, %15
+  %mul890.i = fmul float %CValue1.0.lcssa.i, %14
+  %add891.i = fadd float %mul890.i, %mul888.i
+  %add892.i = fadd float %add891.i, %add869.i
+  store float %add892.i, float addrspace(1)* %200, align 4, !tbaa !249
+  br label %"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEE.exit"
+
+"_ZZ26cgemm_NoTransAB_loopunrollRN2hc16accelerator_viewEPNS_12short_vector7float_2ElS4_lS4_liiiiiiS3_S3_ENK3$_0clERNS_11tiled_indexILi2EEE.exit": ; preds = %209, %._crit_edge.i
+  ret void
+}
+
+attributes #0 = { nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="fiji" "target-features"="+fp64-denormals,-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { nounwind }
+attributes #3 = { convergent nounwind }
+attributes #4 = { nounwind readnone }
+
+
+!232 = !{!233, !234, i64 6}
+!233 = !{!"hsa_kernel_dispatch_packet_s", !234, i64 0, !234, i64 2, !234, i64 4, !234, i64 6, !234, i64 8, !234, i64 10, !237, i64 12, !237, i64 16, !237, i64 20, !237, i64 24, !237, i64 28, !238, i64 32, !239, i64 40, !237, i64 48, !238, i64 56, !240, i64 64}
+!234 = !{!"short", !235, i64 0}
+!235 = !{!"omnipotent char", !236, i64 0}
+!236 = !{!"Simple C/C++ TBAA"}
+!237 = !{!"int", !235, i64 0}
+!238 = !{!"long", !235, i64 0}
+!239 = !{!"any pointer", !235, i64 0}
+!240 = !{!"hsa_signal_s", !238, i64 0}
+!241 = !{!238, !238, i64 0}
+!242 = !{!233, !234, i64 4}
+!243 = !{!244, !245, i64 0}
+!244 = !{!"_ZTSN2hc12short_vector7float_2E", !245, i64 0, !245, i64 4}
+!245 = !{!"float", !246, i64 0}
+!246 = !{!"omnipotent char", !247, i64 0}
+!247 = !{!"Simple C++ TBAA"}
+!248 = !{!245, !245, i64 0}
+!249 = !{!244, !245, i64 4}
+!250 = !{!251, !251, i64 0}
+!251 = !{!"double", !246, i64 0}
+!252 = !{!233, !237, i64 16}
+!253 = !{!254, !254, i64 0}
+!254 = !{!"any pointer", !246, i64 0}
+!255 = !{!256, !256, i64 0}
+!256 = !{!"long", !246, i64 0}
+!257 = !{!258, !258, i64 0}
+!258 = !{!"int", !246, i64 0}
+!259 = !{!233, !237, i64 12}