[GlobalISel] Start using vectors in GISelKnownBits

[Petar.Avramovic]

Summary

For vectors we consider a bit as known if it is the same for all demanded
vector elements (all elements by default). KnownBits BitWidth for vector
type is size of vector element. Add support for G_BUILD_VECTOR.
This allows combines of urem_pow2_to_mask in pre-legalizer combiner.

[foad]

"we consider bit as knows if it same for all" -> "we consider a bit as known if it is the same for all"

Looks pretty good to me, but adding some more globalisel folks in case they have comments.

llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp

171

Don't pass in DemandedElts here.

[Petar.Avramovic]

llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp

171

We don't have such computeKnownBitsImpl in global-isel. Sdag computeKnownBits(SrcOp, Depth + 1) will call computeKnownBits(Op, 'allones', Depth).

[foad]

llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp

171

OK, well I guess it's harmless, since the operand will never have vector type, so I guess it will be ignored. I just found it a bit confusing.

[arsenm]

Some tests for the non-splat case would be nice

llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp

164

As a follow up should also try to handle G_BUILD_VECTOR_TRUNC

[Petar.Avramovic]

Adding non-splat test.
isKnownToBeAPowerOfTwo ends up checking known bits for build_vector and fails for non-splat values.

[paquette]

Can you add a case to unittests/CodeGen/GlobalISel/KnownBitsTest.cpp as well?

Other than that, this LGTM.

[dsanders]

I think lifting the if (DstTy.isVector()) without testing all the cases it affects is a bit risky. LGTM if that's confirmed safe or if the check is sunk to the cases you didn't mean to affect

llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp

143–144

Is removing this safe for the other cases of the switch (Opcode) below? Cases like G_ADD/G_SUB haven't been tested on vectors yet and this patch doesn't test them. It may be safer to sink this to each case and remove it as we test each one

[Petar.Avramovic]

Add a few unit tests and break for vector types on some opcodes.

[arsenm]

llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp

233–234

This is impossible

240–241

This is impossible

446–447

This is impossible

[Petar.Avramovic]

Removed isVector checks for non-Vector opcodes. Use Register instead of unsigned.

[thakis]

One of your 3 changes in https://github.com/llvm/llvm-project/compare/fe5c2c3ca682...bf5a58265047 broke check-llvm: http://45.33.8.238/linux/40834/step_12.txt

Please take a look for now, and revert if fixing takes a while.

Also, please spread out commits a bit more over time, so that when one breaks something, it's clearer which one it is :)

[thakis]

That failure was due to this change, so I reverted it in 4b1015361c16113e1b03c11af646d11522913705 . Please run tests before committing.

I've also reverted D96031 in 59beb1ef6d759eb63422507e8e8488502bcb62d8 since it depended on this change here.

[thakis]

...and D95432 in e68de60bc4f49e282756f1bc8072a196660e0fab since it also depended on this change if the AMDGPU target is enabled.

llvm/lib/CodeGen/GlobalISel/GISelKnownBits.cpp

void GISelKnownBits::computeKnownBitsImpl(Register R, KnownBits &Known,
   // type constraint (i.e. it has a register class constraint instead). This is
   // unlikely to occur except by looking through copies but it is possible for
   // the initial register being queried to be in this state.
   if (!DstTy.isValid()) {
     Known = KnownBits();
     return;
   }
 
-  unsigned BitWidth = DstTy.getSizeInBits();
+  unsigned BitWidth = DstTy.getScalarSizeInBits();
   auto CacheEntry = ComputeKnownBitsCache.find(R);
   if (CacheEntry != ComputeKnownBitsCache.end()) {
     Known = CacheEntry->second;
     LLVM_DEBUG(dbgs() << "Cache hit at ");
     LLVM_DEBUG(dumpResult(MI, Known, Depth));
     assert(Known.getBitWidth() == BitWidth && "Cache entry size doesn't match");
     return;
   }
   Known = KnownBits(BitWidth); // Don't know anything
 
-  if (DstTy.isVector())
-    return; // TODO: Handle vectors.
-
   // Depth may get bigger than max depth if it gets passed to a different
   // GISelKnownBits object.
   // This may happen when say a generic part uses a GISelKnownBits object
   // with some max depth, but then we hit TL.computeKnownBitsForTargetInstr
   // which creates a new GISelKnownBits object with a different and smaller
   // depth. If we just check for equality, we would never exit if the depth
   // that is passed down to the target specific GISelKnownBits object is
   // already bigger than its max depth.
   if (Depth >= getMaxDepth())
     return;
 
   if (!DemandedElts)
     return; // No demanded elts, better to assume we don't know anything.
 
   KnownBits Known2;
 
   switch (Opcode) {
   default:
     TL.computeKnownBitsForTargetInstr(*this, R, Known, DemandedElts, MRI,
                                       Depth);
     break;
+  case TargetOpcode::G_BUILD_VECTOR: {
+    // Collect the known bits that are shared by every demanded vector element.
+    Known.Zero.setAllBits(); Known.One.setAllBits();
+    for (unsigned i = 0, e = MI.getNumOperands() - 1; i < e; ++i) {
+      if (!DemandedElts[i])
+        continue;
+
+      computeKnownBitsImpl(MI.getOperand(i + 1).getReg(), Known2, DemandedElts,
+                           Depth + 1);
+
+      // Known bits are the values that are shared by every demanded element.
+      Known = KnownBits::commonBits(Known, Known2);
+
+      // If we don't know any bits, early out.
+      if (Known.isUnknown())
+        break;
+    }
+    break;
+  }
   case TargetOpcode::COPY:
   case TargetOpcode::G_PHI:
   case TargetOpcode::PHI: {
     Known.One = APInt::getAllOnesValue(BitWidth);
     Known.Zero = APInt::getAllOnesValue(BitWidth);
     // Destination registers should not have subregisters at this
     // point of the pipeline, otherwise the main live-range will be
     // defined more than once, which is against SSA.
for (unsigned Idx = 1; Idx < MI.getNumOperands(); Idx += 2) {
         // We know nothing.
         Known = KnownBits(BitWidth);
         break;
       }
     }
     break;
   }
   case TargetOpcode::G_CONSTANT: {
     auto CstVal = getConstantVRegVal(R, MRI);
     if (!CstVal)
       break;
     Known = KnownBits::makeConstant(*CstVal);
     break;
   }
   case TargetOpcode::G_FRAME_INDEX: {
     int FrameIdx = MI.getOperand(1).getIndex();
     TL.computeKnownBitsForFrameIndex(FrameIdx, Known, MF);
     break;
   }
   case TargetOpcode::G_SUB: {
     computeKnownBitsImpl(MI.getOperand(1).getReg(), Known, DemandedElts,
                          Depth + 1);
     computeKnownBitsImpl(MI.getOperand(2).getReg(), Known2, DemandedElts,
                          Depth + 1);
     Known = KnownBits::computeForAddSub(/*Add*/ false, /*NSW*/ false, Known,
case TargetOpcode::G_TRUNC: {
     Known = Known.zextOrTrunc(SrcBitWidth);
     computeKnownBitsImpl(SrcReg, Known, DemandedElts, Depth + 1);
     Known = Known.zextOrTrunc(BitWidth);
     if (BitWidth > SrcBitWidth)
       Known.Zero.setBitsFrom(SrcBitWidth);
     break;
   }
   case TargetOpcode::G_MERGE_VALUES: {
     unsigned NumOps = MI.getNumOperands();
     unsigned OpSize = MRI.getType(MI.getOperand(1).getReg()).getSizeInBits();
 
     for (unsigned I = 0; I != NumOps - 1; ++I) {
       KnownBits SrcOpKnown;
       computeKnownBitsImpl(MI.getOperand(I + 1).getReg(), SrcOpKnown,
                            DemandedElts, Depth + 1);
       Known.insertBits(SrcOpKnown, I * OpSize);
     }
     break;

llvm/test/CodeGen/AMDGPU/GlobalISel/combine-urem-pow-2.mir

bb.0:
     ; GCN: $vgpr0 = COPY %rem(<2 x s16>)
     %var:_(<2 x s16>) = COPY $vgpr0
     %shift_amt:_(<2 x s16>) = COPY $vgpr1
     %four:_(s32) = G_CONSTANT i32 4
     %four_vec:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC %four, %four
     %rem:_(<2 x s16>) = G_UREM %var, %four_vec
     $vgpr0 = COPY %rem
 ...
+
+---
+name: v_urem_v2i32_pow2k_denom
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; GCN-LABEL: name: v_urem_v2i32_pow2k_denom
+    ; GCN: liveins: $vgpr0_vgpr1
+    ; GCN: %var:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GCN: %pow2:_(s32) = G_CONSTANT i32 4096
+    ; GCN: %pow2_vec:_(<2 x s32>) = G_BUILD_VECTOR %pow2(s32), %pow2(s32)
+    ; GCN: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 -1
+    ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[C]](s32), [[C]](s32)
+    ; GCN: [[ADD:%[0-9]+]]:_(<2 x s32>) = G_ADD %pow2_vec, [[BUILD_VECTOR]]
+    ; GCN: %rem:_(<2 x s32>) = G_AND %var, [[ADD]]
+    ; GCN: $vgpr0_vgpr1 = COPY %rem(<2 x s32>)
+    %var:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %pow2:_(s32) = G_CONSTANT i32 4096
+    %pow2_vec:_(<2 x s32>) = G_BUILD_VECTOR %pow2(s32), %pow2(s32)
+    %rem:_(<2 x s32>) = G_UREM %var, %pow2_vec
+    $vgpr0_vgpr1 = COPY %rem
+...
+
+---
+name: v_urem_v2i32_pow2k_not_splat_denom
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; GCN-LABEL: name: v_urem_v2i32_pow2k_not_splat_denom
+    ; GCN: liveins: $vgpr0_vgpr1
+    ; GCN: %var:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GCN: %pow2_1:_(s32) = G_CONSTANT i32 4096
+    ; GCN: %pow2_2:_(s32) = G_CONSTANT i32 2048
+    ; GCN: %pow2_vec:_(<2 x s32>) = G_BUILD_VECTOR %pow2_1(s32), %pow2_2(s32)
+    ; GCN: %rem:_(<2 x s32>) = G_UREM %var, %pow2_vec
+    ; GCN: $vgpr0_vgpr1 = COPY %rem(<2 x s32>)
+    %var:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %pow2_1:_(s32) = G_CONSTANT i32 4096
+    %pow2_2:_(s32) = G_CONSTANT i32 2048
+    %pow2_vec:_(<2 x s32>) = G_BUILD_VECTOR %pow2_1(s32), %pow2_2(s32)
+    %rem:_(<2 x s32>) = G_UREM %var, %pow2_vec
+    $vgpr0_vgpr1 = COPY %rem
+...
+
+---
+name: v_urem_v2i64_pow2k_denom
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+
+    ; GCN-LABEL: name: v_urem_v2i64_pow2k_denom
+    ; GCN: liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GCN: %var:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; GCN: %pow2:_(s64) = G_CONSTANT i64 4096
+    ; GCN: %pow2_vec:_(<2 x s64>) = G_BUILD_VECTOR %pow2(s64), %pow2(s64)
+    ; GCN: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 -1
+    ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[C]](s64), [[C]](s64)
+    ; GCN: [[ADD:%[0-9]+]]:_(<2 x s64>) = G_ADD %pow2_vec, [[BUILD_VECTOR]]
+    ; GCN: %rem:_(<2 x s64>) = G_AND %var, [[ADD]]
+    ; GCN: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %rem(<2 x s64>)
+    %var:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %pow2:_(s64) = G_CONSTANT i64 4096
+    %pow2_vec:_(<2 x s64>) = G_BUILD_VECTOR %pow2(s64), %pow2(s64)
+    %rem:_(<2 x s64>) = G_UREM %var, %pow2_vec
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %rem
+...

llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i32.ll

 ; CHECK-NEXT:    s_add_i32 s4, 0x1000, -1
 ; CHECK-NEXT:    v_and_b32_e32 v0, s4, v0
 ; CHECK-NEXT:    s_setpc_b64 s[30:31]
   %result = urem i32 %num, 4096
   ret i32 %result
 }
 
 define <2 x i32> @v_urem_v2i32_pow2k_denom(<2 x i32> %num) {
-; GISEL-LABEL: v_urem_v2i32_pow2k_denom:
-; GISEL:       ; %bb.0:
-; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT:    s_movk_i32 s4, 0x1000
-; GISEL-NEXT:    v_cvt_f32_u32_e32 v2, s4
-; GISEL-NEXT:    s_sub_i32 s5, 0, s4
-; GISEL-NEXT:    v_rcp_iflag_f32_e32 v2, v2
-; GISEL-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
-; GISEL-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GISEL-NEXT:    v_mul_lo_u32 v3, s5, v2
-; GISEL-NEXT:    v_mul_hi_u32 v3, v2, v3
-; GISEL-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
-; GISEL-NEXT:    v_mul_hi_u32 v3, v0, v2
-; GISEL-NEXT:    v_mul_hi_u32 v2, v1, v2
-; GISEL-NEXT:    v_lshlrev_b32_e32 v3, 12, v3
-; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 12, v2
-; GISEL-NEXT:    v_sub_i32_e32 v0, vcc, v0, v3
-; GISEL-NEXT:    v_sub_i32_e32 v1, vcc, v1, v2
-; GISEL-NEXT:    v_subrev_i32_e32 v2, vcc, s4, v0
-; GISEL-NEXT:    v_subrev_i32_e32 v3, vcc, s4, v1
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, s4, v0
-; GISEL-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, s4, v1
-; GISEL-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GISEL-NEXT:    v_subrev_i32_e32 v2, vcc, s4, v0
-; GISEL-NEXT:    v_subrev_i32_e32 v3, vcc, s4, v1
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, s4, v0
-; GISEL-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, s4, v1
-; GISEL-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GISEL-NEXT:    s_setpc_b64 s[30:31]
-;
-; CGP-LABEL: v_urem_v2i32_pow2k_denom:
-; CGP:       ; %bb.0:
-; CGP-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CGP-NEXT:    s_add_i32 s4, 0x1000, -1
-; CGP-NEXT:    v_and_b32_e32 v0, s4, v0
-; CGP-NEXT:    v_and_b32_e32 v1, s4, v1
-; CGP-NEXT:    s_setpc_b64 s[30:31]
+; CHECK-LABEL: v_urem_v2i32_pow2k_denom:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    s_add_i32 s4, 0x1000, -1
+; CHECK-NEXT:    v_and_b32_e32 v0, s4, v0
+; CHECK-NEXT:    v_and_b32_e32 v1, s4, v1
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
   %result = urem <2 x i32> %num, <i32 4096, i32 4096>
   ret <2 x i32> %result
 }
 
 define i32 @v_urem_i32_oddk_denom(i32 %num) {
 ; CHECK-LABEL: v_urem_i32_oddk_denom:
 ; CHECK:       ; %bb.0:
 ; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)

llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll

 ; CHECK-NEXT:    v_and_b32_e32 v0, s4, v0
 ; CHECK-NEXT:    v_and_b32_e32 v1, s5, v1
 ; CHECK-NEXT:    s_setpc_b64 s[30:31]
   %result = urem i64 %num, 4096
   ret i64 %result
 }
 
 define <2 x i64> @v_urem_v2i64_pow2k_denom(<2 x i64> %num) {
-; GISEL-LABEL: v_urem_v2i64_pow2k_denom:
-; GISEL:       ; %bb.0:
-; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT:    s_movk_i32 s10, 0x1000
-; GISEL-NEXT:    v_cvt_f32_u32_e32 v4, s10
-; GISEL-NEXT:    s_sub_u32 s8, 0, s10
-; GISEL-NEXT:    s_cselect_b32 s4, 1, 0
-; GISEL-NEXT:    v_cvt_f32_ubyte0_e32 v5, 0
-; GISEL-NEXT:    v_mov_b32_e32 v6, v4
-; GISEL-NEXT:    s_and_b32 s4, s4, 1
-; GISEL-NEXT:    v_mac_f32_e32 v4, 0x4f800000, v5
-; GISEL-NEXT:    v_mac_f32_e32 v6, 0x4f800000, v5
-; GISEL-NEXT:    v_rcp_iflag_f32_e32 v4, v4
-; GISEL-NEXT:    v_rcp_iflag_f32_e32 v5, v6
-; GISEL-NEXT:    s_cmp_lg_u32 s4, 0
-; GISEL-NEXT:    s_subb_u32 s9, 0, 0
-; GISEL-NEXT:    v_mul_f32_e32 v4, 0x5f7ffffc, v4
-; GISEL-NEXT:    v_mul_f32_e32 v5, 0x5f7ffffc, v5
-; GISEL-NEXT:    v_mul_f32_e32 v6, 0x2f800000, v4
-; GISEL-NEXT:    s_sub_u32 s11, 0, s10
-; GISEL-NEXT:    s_cselect_b32 s4, 1, 0
-; GISEL-NEXT:    v_mul_f32_e32 v7, 0x2f800000, v5
-; GISEL-NEXT:    v_trunc_f32_e32 v6, v6
-; GISEL-NEXT:    s_and_b32 s4, s4, 1
-; GISEL-NEXT:    v_trunc_f32_e32 v7, v7
-; GISEL-NEXT:    v_mac_f32_e32 v4, 0xcf800000, v6
-; GISEL-NEXT:    v_cvt_u32_f32_e32 v6, v6
-; GISEL-NEXT:    v_mac_f32_e32 v5, 0xcf800000, v7
-; GISEL-NEXT:    v_cvt_u32_f32_e32 v7, v7
-; GISEL-NEXT:    v_cvt_u32_f32_e32 v4, v4
-; GISEL-NEXT:    s_cmp_lg_u32 s4, 0
-; GISEL-NEXT:    s_subb_u32 s6, 0, 0
-; GISEL-NEXT:    v_mul_lo_u32 v8, s11, v6
-; GISEL-NEXT:    v_cvt_u32_f32_e32 v5, v5
-; GISEL-NEXT:    v_mul_lo_u32 v9, s8, v7
-; GISEL-NEXT:    v_mul_lo_u32 v10, s11, v4
-; GISEL-NEXT:    v_mul_lo_u32 v11, s6, v4
-; GISEL-NEXT:    v_mul_hi_u32 v12, s11, v4
-; GISEL-NEXT:    v_mul_lo_u32 v13, s8, v5
-; GISEL-NEXT:    v_mul_lo_u32 v14, s9, v5
-; GISEL-NEXT:    v_mul_hi_u32 v15, s8, v5
-; GISEL-NEXT:    v_add_i32_e32 v8, vcc, v11, v8
-; GISEL-NEXT:    v_mul_lo_u32 v11, v6, v10
-; GISEL-NEXT:    v_mul_hi_u32 v16, v4, v10
-; GISEL-NEXT:    v_mul_hi_u32 v10, v6, v10
-; GISEL-NEXT:    v_add_i32_e32 v9, vcc, v14, v9
-; GISEL-NEXT:    v_mul_lo_u32 v14, v7, v13
-; GISEL-NEXT:    v_mul_hi_u32 v17, v5, v13
-; GISEL-NEXT:    v_mul_hi_u32 v13, v7, v13
-; GISEL-NEXT:    v_add_i32_e32 v8, vcc, v8, v12
-; GISEL-NEXT:    v_add_i32_e32 v9, vcc, v9, v15
-; GISEL-NEXT:    v_mul_lo_u32 v12, v4, v8
-; GISEL-NEXT:    v_mul_lo_u32 v15, v6, v8
-; GISEL-NEXT:    v_mul_hi_u32 v18, v4, v8
-; GISEL-NEXT:    v_mul_hi_u32 v8, v6, v8
-; GISEL-NEXT:    v_mul_lo_u32 v19, v5, v9
-; GISEL-NEXT:    v_add_i32_e32 v14, vcc, v14, v19
-; GISEL-NEXT:    v_cndmask_b32_e64 v19, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v14, vcc, v14, v17
-; GISEL-NEXT:    v_mul_lo_u32 v14, v7, v9
-; GISEL-NEXT:    v_mul_hi_u32 v17, v5, v9
-; GISEL-NEXT:    v_mul_hi_u32 v9, v7, v9
-; GISEL-NEXT:    v_add_i32_e64 v11, s[4:5], v11, v12
-; GISEL-NEXT:    v_cndmask_b32_e64 v12, 0, 1, s[4:5]
-; GISEL-NEXT:    v_add_i32_e64 v10, s[4:5], v15, v10
-; GISEL-NEXT:    v_cndmask_b32_e64 v15, 0, 1, s[4:5]
-; GISEL-NEXT:    v_add_i32_e64 v13, s[4:5], v14, v13
-; GISEL-NEXT:    v_cndmask_b32_e64 v14, 0, 1, s[4:5]
-; GISEL-NEXT:    v_add_i32_e64 v11, s[4:5], v11, v16
-; GISEL-NEXT:    v_cndmask_b32_e64 v11, 0, 1, s[4:5]
-; GISEL-NEXT:    v_add_i32_e64 v10, s[4:5], v10, v18
-; GISEL-NEXT:    v_cndmask_b32_e64 v16, 0, 1, s[4:5]
-; GISEL-NEXT:    v_cndmask_b32_e64 v18, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v13, vcc, v13, v17
-; GISEL-NEXT:    v_cndmask_b32_e64 v17, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v11, vcc, v12, v11
-; GISEL-NEXT:    v_add_i32_e32 v12, vcc, v15, v16
-; GISEL-NEXT:    v_add_i32_e32 v15, vcc, v19, v18
-; GISEL-NEXT:    v_add_i32_e32 v14, vcc, v14, v17
-; GISEL-NEXT:    v_add_i32_e32 v10, vcc, v10, v11
-; GISEL-NEXT:    v_cndmask_b32_e64 v11, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v13, vcc, v13, v15
-; GISEL-NEXT:    v_cndmask_b32_e64 v15, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v11, vcc, v12, v11
-; GISEL-NEXT:    v_add_i32_e32 v12, vcc, v14, v15
-; GISEL-NEXT:    v_add_i32_e32 v8, vcc, v8, v11
-; GISEL-NEXT:    v_add_i32_e32 v9, vcc, v9, v12
-; GISEL-NEXT:    v_add_i32_e32 v4, vcc, v4, v10
-; GISEL-NEXT:    v_addc_u32_e64 v10, s[4:5], v6, v8, vcc
-; GISEL-NEXT:    v_add_i32_e64 v6, s[4:5], v6, v8
-; GISEL-NEXT:    v_mul_lo_u32 v8, s11, v4
-; GISEL-NEXT:    v_mul_lo_u32 v11, s6, v4
-; GISEL-NEXT:    v_mul_hi_u32 v12, s11, v4
-; GISEL-NEXT:    v_add_i32_e64 v5, s[4:5], v5, v13
-; GISEL-NEXT:    v_addc_u32_e64 v13, s[6:7], v7, v9, s[4:5]
-; GISEL-NEXT:    v_add_i32_e64 v7, s[6:7], v7, v9
-; GISEL-NEXT:    v_mul_lo_u32 v9, s8, v5
-; GISEL-NEXT:    v_mul_lo_u32 v14, s9, v5
-; GISEL-NEXT:    v_mul_hi_u32 v15, s8, v5
-; GISEL-NEXT:    v_mul_lo_u32 v16, s11, v10
-; GISEL-NEXT:    v_mul_lo_u32 v17, v10, v8
-; GISEL-NEXT:    v_mul_hi_u32 v18, v4, v8
-; GISEL-NEXT:    v_mul_hi_u32 v8, v10, v8
-; GISEL-NEXT:    v_mul_lo_u32 v19, s8, v13
-; GISEL-NEXT:    v_add_i32_e64 v11, s[6:7], v11, v16
-; GISEL-NEXT:    v_mul_lo_u32 v16, v13, v9
-; GISEL-NEXT:    v_add_i32_e64 v14, s[6:7], v14, v19
-; GISEL-NEXT:    v_mul_hi_u32 v19, v5, v9
-; GISEL-NEXT:    v_mul_hi_u32 v9, v13, v9
-; GISEL-NEXT:    v_add_i32_e64 v11, s[6:7], v11, v12
-; GISEL-NEXT:    v_add_i32_e64 v12, s[6:7], v14, v15
-; GISEL-NEXT:    v_mul_lo_u32 v14, v4, v11
-; GISEL-NEXT:    v_mul_lo_u32 v15, v5, v12
-; GISEL-NEXT:    v_add_i32_e64 v15, s[6:7], v16, v15
-; GISEL-NEXT:    v_cndmask_b32_e64 v16, 0, 1, s[6:7]
-; GISEL-NEXT:    v_add_i32_e64 v15, s[6:7], v15, v19
-; GISEL-NEXT:    v_mul_lo_u32 v15, v10, v11
-; GISEL-NEXT:    v_mul_hi_u32 v19, v4, v11
-; GISEL-NEXT:    v_mul_hi_u32 v10, v10, v11
-; GISEL-NEXT:    v_mul_lo_u32 v11, v13, v12
-; GISEL-NEXT:    v_mul_hi_u32 v13, v13, v12
-; GISEL-NEXT:    v_mul_hi_u32 v12, v5, v12
-; GISEL-NEXT:    v_add_i32_e64 v14, s[8:9], v17, v14
-; GISEL-NEXT:    v_cndmask_b32_e64 v17, 0, 1, s[8:9]
-; GISEL-NEXT:    v_add_i32_e64 v8, s[8:9], v15, v8
-; GISEL-NEXT:    v_cndmask_b32_e64 v15, 0, 1, s[8:9]
-; GISEL-NEXT:    v_add_i32_e64 v9, s[8:9], v11, v9
-; GISEL-NEXT:    v_cndmask_b32_e64 v11, 0, 1, s[8:9]
-; GISEL-NEXT:    v_add_i32_e64 v14, s[8:9], v14, v18
-; GISEL-NEXT:    v_cndmask_b32_e64 v14, 0, 1, s[8:9]
-; GISEL-NEXT:    v_add_i32_e64 v8, s[8:9], v8, v19
-; GISEL-NEXT:    v_cndmask_b32_e64 v18, 0, 1, s[8:9]
-; GISEL-NEXT:    v_cndmask_b32_e64 v19, 0, 1, s[6:7]
-; GISEL-NEXT:    v_add_i32_e64 v9, s[6:7], v9, v12
-; GISEL-NEXT:    v_cndmask_b32_e64 v12, 0, 1, s[6:7]
-; GISEL-NEXT:    v_add_i32_e64 v14, s[6:7], v17, v14
-; GISEL-NEXT:    v_add_i32_e64 v15, s[6:7], v15, v18
-; GISEL-NEXT:    v_add_i32_e64 v16, s[6:7], v16, v19
-; GISEL-NEXT:    v_add_i32_e64 v11, s[6:7], v11, v12
-; GISEL-NEXT:    v_add_i32_e64 v8, s[6:7], v8, v14
-; GISEL-NEXT:    v_cndmask_b32_e64 v12, 0, 1, s[6:7]
-; GISEL-NEXT:    v_add_i32_e64 v9, s[6:7], v9, v16
-; GISEL-NEXT:    v_cndmask_b32_e64 v14, 0, 1, s[6:7]
-; GISEL-NEXT:    v_add_i32_e64 v12, s[6:7], v15, v12
-; GISEL-NEXT:    v_add_i32_e64 v11, s[6:7], v11, v14
-; GISEL-NEXT:    v_add_i32_e64 v10, s[6:7], v10, v12
-; GISEL-NEXT:    v_add_i32_e64 v11, s[6:7], v13, v11
-; GISEL-NEXT:    v_addc_u32_e32 v6, vcc, v6, v10, vcc
-; GISEL-NEXT:    v_addc_u32_e64 v7, vcc, v7, v11, s[4:5]
-; GISEL-NEXT:    v_add_i32_e32 v4, vcc, v4, v8
-; GISEL-NEXT:    v_addc_u32_e32 v6, vcc, 0, v6, vcc
-; GISEL-NEXT:    v_mul_lo_u32 v8, v3, v4
-; GISEL-NEXT:    v_mul_hi_u32 v10, v2, v4
-; GISEL-NEXT:    v_mul_hi_u32 v4, v3, v4
-; GISEL-NEXT:    v_add_i32_e32 v5, vcc, v5, v9
-; GISEL-NEXT:    v_addc_u32_e32 v7, vcc, 0, v7, vcc
-; GISEL-NEXT:    v_mul_lo_u32 v9, v1, v5
-; GISEL-NEXT:    v_mul_hi_u32 v11, v0, v5
-; GISEL-NEXT:    v_mul_hi_u32 v5, v1, v5
-; GISEL-NEXT:    v_mul_lo_u32 v12, v2, v6
-; GISEL-NEXT:    v_mul_lo_u32 v13, v3, v6
-; GISEL-NEXT:    v_mul_hi_u32 v14, v2, v6
-; GISEL-NEXT:    v_mul_hi_u32 v6, v3, v6
-; GISEL-NEXT:    v_mul_lo_u32 v15, v0, v7
-; GISEL-NEXT:    v_mul_lo_u32 v16, v1, v7
-; GISEL-NEXT:    v_mul_hi_u32 v17, v0, v7
-; GISEL-NEXT:    v_mul_hi_u32 v7, v1, v7
-; GISEL-NEXT:    v_add_i32_e32 v8, vcc, v8, v12
-; GISEL-NEXT:    v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v4, vcc, v13, v4
-; GISEL-NEXT:    v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v9, vcc, v9, v15
-; GISEL-NEXT:    v_cndmask_b32_e64 v15, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v5, vcc, v16, v5
-; GISEL-NEXT:    v_cndmask_b32_e64 v16, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v8, vcc, v8, v10
-; GISEL-NEXT:    v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v4, vcc, v4, v14
-; GISEL-NEXT:    v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v9, vcc, v9, v11
-; GISEL-NEXT:    v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v5, vcc, v5, v17
-; GISEL-NEXT:    v_cndmask_b32_e64 v11, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v8, vcc, v12, v8
-; GISEL-NEXT:    v_add_i32_e32 v10, vcc, v13, v10
-; GISEL-NEXT:    v_add_i32_e32 v9, vcc, v15, v9
-; GISEL-NEXT:    v_add_i32_e32 v11, vcc, v16, v11
-; GISEL-NEXT:    v_add_i32_e32 v4, vcc, v4, v8
-; GISEL-NEXT:    v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v5, vcc, v5, v9
-; GISEL-NEXT:    v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT:    v_add_i32_e32 v8, vcc, v10, v8
-; GISEL-NEXT:    v_mul_lo_u32 v10, s10, v4
-; GISEL-NEXT:    v_mul_lo_u32 v12, 0, v4
-; GISEL-NEXT:    v_mul_hi_u32 v4, s10, v4
-; GISEL-NEXT:    v_add_i32_e32 v9, vcc, v11, v9
-; GISEL-NEXT:    v_mul_lo_u32 v11, s10, v5
-; GISEL-NEXT:    v_mul_lo_u32 v13, 0, v5
-; GISEL-NEXT:    v_mul_hi_u32 v5, s10, v5
-; GISEL-NEXT:    v_add_i32_e32 v6, vcc, v6, v8
-; GISEL-NEXT:    v_add_i32_e32 v7, vcc, v7, v9
-; GISEL-NEXT:    v_mul_lo_u32 v6, s10, v6
-; GISEL-NEXT:    v_mul_lo_u32 v7, s10, v7
-; GISEL-NEXT:    v_add_i32_e32 v6, vcc, v12, v6
-; GISEL-NEXT:    v_add_i32_e32 v7, vcc, v13, v7
-; GISEL-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
-; GISEL-NEXT:    v_add_i32_e32 v5, vcc, v7, v5
-; GISEL-NEXT:    v_sub_i32_e32 v2, vcc, v2, v10
-; GISEL-NEXT:    v_subb_u32_e64 v6, s[4:5], v3, v4, vcc
-; GISEL-NEXT:    v_sub_i32_e64 v3, s[4:5], v3, v4
-; GISEL-NEXT:    v_cmp_le_u32_e64 s[4:5], s10, v2
-; GISEL-NEXT:    v_cndmask_b32_e64 v4, 0, -1, s[4:5]
-; GISEL-NEXT:    v_sub_i32_e64 v0, s[4:5], v0, v11
-; GISEL-NEXT:    v_subb_u32_e64 v7, s[6:7], v1, v5, s[4:5]
-; GISEL-NEXT:    v_sub_i32_e64 v1, s[6:7], v1, v5
-; GISEL-NEXT:    v_cmp_le_u32_e64 s[6:7], s10, v0
-; GISEL-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[6:7]
-; GISEL-NEXT:    v_cmp_le_u32_e64 s[6:7], 0, v6
-; GISEL-NEXT:    v_cndmask_b32_e64 v8, 0, -1, s[6:7]
-; GISEL-NEXT:    v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, 0, v7
-; GISEL-NEXT:    v_cndmask_b32_e64 v9, 0, -1, vcc
-; GISEL-NEXT:    v_subbrev_u32_e64 v1, vcc, 0, v1, s[4:5]
-; GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
-; GISEL-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GISEL-NEXT:    v_subrev_i32_e32 v8, vcc, s10, v2
-; GISEL-NEXT:    v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, s10, v8
-; GISEL-NEXT:    v_cndmask_b32_e64 v10, 0, -1, vcc
-; GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GISEL-NEXT:    v_cndmask_b32_e32 v5, v9, v5, vcc
-; GISEL-NEXT:    v_subrev_i32_e32 v9, vcc, s10, v0
-; GISEL-NEXT:    v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, s10, v9
-; GISEL-NEXT:    v_cndmask_b32_e64 v11, 0, -1, vcc
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, 0, v3
-; GISEL-NEXT:    v_cndmask_b32_e64 v12, 0, -1, vcc
-; GISEL-NEXT:    v_subrev_i32_e32 v13, vcc, s10, v8
-; GISEL-NEXT:    v_subbrev_u32_e32 v14, vcc, 0, v3, vcc
-; GISEL-NEXT:    v_cmp_le_u32_e32 vcc, 0, v1
-; GISEL-NEXT:    v_cndmask_b32_e64 v15, 0, -1, vcc
-; GISEL-NEXT:    v_subrev_i32_e32 v16, vcc, s10, v9
-; GISEL-NEXT:    v_subbrev_u32_e32 v17, vcc, 0, v1, vcc
-; GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; GISEL-NEXT:    v_cndmask_b32_e32 v10, v12, v10, vcc
-; GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
-; GISEL-NEXT:    v_cndmask_b32_e32 v11, v15, v11, vcc
-; GISEL-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v10
-; GISEL-NEXT:    v_cndmask_b32_e32 v8, v8, v13, vcc
-; GISEL-NEXT:    v_cmp_ne_u32_e64 s[4:5], 0, v11
-; GISEL-NEXT:    v_cndmask_b32_e64 v9, v9, v16, s[4:5]
-; GISEL-NEXT:    v_cndmask_b32_e32 v3, v3, v14, vcc
-; GISEL-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
-; GISEL-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
-; GISEL-NEXT:    v_cndmask_b32_e64 v1, v1, v17, s[4:5]
-; GISEL-NEXT:    v_cmp_ne_u32_e64 s[4:5], 0, v5
-; GISEL-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[4:5]
-; GISEL-NEXT:    v_cndmask_b32_e64 v1, v7, v1, s[4:5]
-; GISEL-NEXT:    v_cndmask_b32_e32 v3, v6, v3, vcc
-; GISEL-NEXT:    s_setpc_b64 s[30:31]
-;
-; CGP-LABEL: v_urem_v2i64_pow2k_denom:
-; CGP:       ; %bb.0:
-; CGP-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CGP-NEXT:    s_movk_i32 s4, 0x1000
-; CGP-NEXT:    s_add_u32 s5, s4, -1
-; CGP-NEXT:    s_cselect_b32 s6, 1, 0
-; CGP-NEXT:    s_and_b32 s6, s6, 1
-; CGP-NEXT:    s_cmp_lg_u32 s6, 0
-; CGP-NEXT:    s_addc_u32 s6, 0, -1
-; CGP-NEXT:    s_add_u32 s4, s4, -1
-; CGP-NEXT:    s_cselect_b32 s7, 1, 0
-; CGP-NEXT:    v_and_b32_e32 v0, s5, v0
-; CGP-NEXT:    s_and_b32 s5, s7, 1
-; CGP-NEXT:    v_and_b32_e32 v1, s6, v1
-; CGP-NEXT:    s_cmp_lg_u32 s5, 0
-; CGP-NEXT:    s_addc_u32 s5, 0, -1
-; CGP-NEXT:    v_and_b32_e32 v2, s4, v2
-; CGP-NEXT:    v_and_b32_e32 v3, s5, v3
-; CGP-NEXT:    s_setpc_b64 s[30:31]
+; CHECK-LABEL: v_urem_v2i64_pow2k_denom:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    s_movk_i32 s4, 0x1000
+; CHECK-NEXT:    s_add_u32 s5, s4, -1
+; CHECK-NEXT:    s_cselect_b32 s6, 1, 0
+; CHECK-NEXT:    s_and_b32 s6, s6, 1
+; CHECK-NEXT:    s_cmp_lg_u32 s6, 0
+; CHECK-NEXT:    s_addc_u32 s6, 0, -1
+; CHECK-NEXT:    s_add_u32 s4, s4, -1
+; CHECK-NEXT:    s_cselect_b32 s7, 1, 0
+; CHECK-NEXT:    v_and_b32_e32 v0, s5, v0
+; CHECK-NEXT:    s_and_b32 s5, s7, 1
+; CHECK-NEXT:    v_and_b32_e32 v1, s6, v1
+; CHECK-NEXT:    s_cmp_lg_u32 s5, 0
+; CHECK-NEXT:    s_addc_u32 s5, 0, -1
+; CHECK-NEXT:    v_and_b32_e32 v2, s4, v2
+; CHECK-NEXT:    v_and_b32_e32 v3, s5, v3
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
   %result = urem <2 x i64> %num, <i64 4096, i64 4096>
   ret <2 x i64> %result
 }
 
 define i64 @v_urem_i64_oddk_denom(i64 %num) {
 ; CHECK-LABEL: v_urem_i64_oddk_denom:
 ; CHECK:       ; %bb.0:
 ; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)