diff --git a/llvm/lib/Transforms/Utils/IntegerDivision.cpp b/llvm/lib/Transforms/Utils/IntegerDivision.cpp
--- a/llvm/lib/Transforms/Utils/IntegerDivision.cpp
+++ b/llvm/lib/Transforms/Utils/IntegerDivision.cpp
@@ -53,6 +53,8 @@
   // ;   %urem         = urem i32 %dividend, %divisor
   // ;   %xored        = xor i32 %urem, %dividend_sgn
   // ;   %srem         = sub i32 %xored, %dividend_sgn
+  Dividend            = Builder.CreateFreeze(Dividend);
+  Divisor             = Builder.CreateFreeze(Divisor);
   Value *DividendSign = Builder.CreateAShr(Dividend, Shift);
   Value *DivisorSign  = Builder.CreateAShr(Divisor, Shift);
   Value *DvdXor       = Builder.CreateXor(Dividend, DividendSign);
@@ -84,6 +86,8 @@
   // ;   %quotient  = udiv i32 %dividend, %divisor
   // ;   %product   = mul i32 %divisor, %quotient
   // ;   %remainder = sub i32 %dividend, %product
+  Dividend         = Builder.CreateFreeze(Dividend);
+  Divisor          = Builder.CreateFreeze(Divisor);
   Value *Quotient  = Builder.CreateUDiv(Dividend, Divisor);
   Value *Product   = Builder.CreateMul(Divisor, Quotient);
   Value *Remainder = Builder.CreateSub(Dividend, Product);
@@ -126,6 +130,8 @@
   // ;   %q_mag  = udiv i32 %u_dvnd, %u_dvsr
   // ;   %tmp4   = xor i32 %q_mag, %q_sgn
   // ;   %q      = sub i32 %tmp4, %q_sgn
+  Dividend      = Builder.CreateFreeze(Dividend);
+  Divisor       = Builder.CreateFreeze(Divisor);
   Value *Tmp    = Builder.CreateAShr(Dividend, Shift);
   Value *Tmp1   = Builder.CreateAShr(Divisor, Shift);
   Value *Tmp2   = Builder.CreateXor(Tmp, Dividend);
@@ -236,28 +242,30 @@
   // ; special-cases:
   // ;   %ret0_1      = icmp eq i32 %divisor, 0
   // ;   %ret0_2      = icmp eq i32 %dividend, 0
-  // ;   %ret0_3      = or i1 %ret0_1, %ret0_2
+  // ;   %ret0_3      = select i1 %ret0_1, i1 1, i1 %ret0_2
   // ;   %tmp0        = tail call i32 @llvm.ctlz.i32(i32 %divisor, i1 true)
   // ;   %tmp1        = tail call i32 @llvm.ctlz.i32(i32 %dividend, i1 true)
   // ;   %sr          = sub nsw i32 %tmp0, %tmp1
   // ;   %ret0_4      = icmp ugt i32 %sr, 31
-  // ;   %ret0        = or i1 %ret0_3, %ret0_4
+  // ;   %ret0        = select i1 %ret0_3, i1 1, i1 %ret0_4
   // ;   %retDividend = icmp eq i32 %sr, 31
   // ;   %retVal      = select i1 %ret0, i32 0, i32 %dividend
-  // ;   %earlyRet    = or i1 %ret0, %retDividend
+  // ;   %earlyRet    = select i1 %ret0, i1 1, i1 %retDividend
   // ;   br i1 %earlyRet, label %end, label %bb1
   Builder.SetInsertPoint(SpecialCases);
+  Divisor            = Builder.CreateFreeze(Divisor);
+  Dividend           = Builder.CreateFreeze(Dividend);
   Value *Ret0_1      = Builder.CreateICmpEQ(Divisor, Zero);
   Value *Ret0_2      = Builder.CreateICmpEQ(Dividend, Zero);
-  Value *Ret0_3      = Builder.CreateOr(Ret0_1, Ret0_2);
+  Value *Ret0_3      = Builder.CreateLogicalOr(Ret0_1, Ret0_2);
   Value *Tmp0 = Builder.CreateCall(CTLZ, {Divisor, True});
   Value *Tmp1 = Builder.CreateCall(CTLZ, {Dividend, True});
   Value *SR          = Builder.CreateSub(Tmp0, Tmp1);
   Value *Ret0_4      = Builder.CreateICmpUGT(SR, MSB);
-  Value *Ret0        = Builder.CreateOr(Ret0_3, Ret0_4);
+  Value *Ret0        = Builder.CreateLogicalOr(Ret0_3, Ret0_4);
   Value *RetDividend = Builder.CreateICmpEQ(SR, MSB);
   Value *RetVal      = Builder.CreateSelect(Ret0, Zero, Dividend);
-  Value *EarlyRet    = Builder.CreateOr(Ret0, RetDividend);
+  Value *EarlyRet    = Builder.CreateLogicalOr(Ret0, RetDividend);
   Builder.CreateCondBr(EarlyRet, End, BB1);
 
   // ; bb1:                                             ; preds = %special-cases
diff --git a/llvm/test/CodeGen/AMDGPU/sdiv64.ll b/llvm/test/CodeGen/AMDGPU/sdiv64.ll
--- a/llvm/test/CodeGen/AMDGPU/sdiv64.ll
+++ b/llvm/test/CodeGen/AMDGPU/sdiv64.ll
@@ -168,12 +168,11 @@
 ; GCN-IR-NEXT:    s_sub_u32 s10, s14, s18
 ; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[20:21], s[10:11], 63
-; GCN-IR-NEXT:    s_mov_b32 s15, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[22:23], s[10:11], 63
 ; GCN-IR-NEXT:    s_or_b64 s[16:17], s[16:17], s[20:21]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[20:21], s[10:11], 63
-; GCN-IR-NEXT:    s_xor_b64 s[22:23], s[16:17], -1
-; GCN-IR-NEXT:    s_and_b64 s[20:21], s[22:23], s[20:21]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[20:21]
+; GCN-IR-NEXT:    s_or_b64 s[20:21], s[16:17], s[22:23]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[20:21]
+; GCN-IR-NEXT:    s_mov_b32 s15, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s16, s10, 1
@@ -524,7 +523,7 @@
 ; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
@@ -694,7 +693,7 @@
 ; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 31
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
@@ -761,7 +760,7 @@
 ; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 23
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
@@ -828,7 +827,7 @@
 ; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 25
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
@@ -910,7 +909,7 @@
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    s_ashr_i64 s[10:11], s[10:11], 40
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v2, s10
 ; GCN-IR-NEXT:    s_ashr_i64 s[6:7], s[6:7], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v3, s6
@@ -926,7 +925,7 @@
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v2, 0, v5, vcc
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-IR-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GCN-IR-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GCN-IR-NEXT:    v_bfe_i32 v2, v2, 0, 24
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v3, 31, v2
@@ -986,10 +985,10 @@
 ; GCN-IR-NEXT:    s_sext_i32_i16 s3, s3
 ; GCN-IR-NEXT:    s_sext_i32_i16 s5, s5
 ; GCN-IR-NEXT:    s_ashr_i64 s[6:7], s[2:3], 24
-; GCN-IR-NEXT:    s_ashr_i32 s2, s3, 31
-; GCN-IR-NEXT:    s_mov_b32 s3, s2
+; GCN-IR-NEXT:    s_ashr_i32 s2, s7, 31
 ; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[4:5], 24
-; GCN-IR-NEXT:    s_ashr_i32 s4, s5, 31
+; GCN-IR-NEXT:    s_mov_b32 s3, s2
+; GCN-IR-NEXT:    s_ashr_i32 s4, s9, 31
 ; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[2:3], s[6:7]
 ; GCN-IR-NEXT:    s_mov_b32 s5, s4
 ; GCN-IR-NEXT:    s_sub_u32 s12, s6, s2
@@ -1012,12 +1011,11 @@
 ; GCN-IR-NEXT:    s_sub_u32 s10, s14, s18
 ; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[20:21], s[10:11], 63
-; GCN-IR-NEXT:    s_mov_b32 s15, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[22:23], s[10:11], 63
 ; GCN-IR-NEXT:    s_or_b64 s[16:17], s[16:17], s[20:21]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[20:21], s[10:11], 63
-; GCN-IR-NEXT:    s_xor_b64 s[22:23], s[16:17], -1
-; GCN-IR-NEXT:    s_and_b64 s[20:21], s[22:23], s[20:21]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[20:21]
+; GCN-IR-NEXT:    s_or_b64 s[20:21], s[16:17], s[22:23]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[20:21]
+; GCN-IR-NEXT:    s_mov_b32 s15, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB9_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s16, s10, 1
@@ -1212,20 +1210,19 @@
 ; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[4:5], s[2:3]
 ; GCN-IR-NEXT:    s_sub_u32 s2, s2, s4
 ; GCN-IR-NEXT:    s_subb_u32 s3, s3, s4
-; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
-; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
-; GCN-IR-NEXT:    s_min_u32 s10, s6, s7
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s2
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s3
+; GCN-IR-NEXT:    s_min_u32 s10, s8, s9
 ; GCN-IR-NEXT:    s_add_u32 s8, s10, 0xffffffc5
 ; GCN-IR-NEXT:    s_addc_u32 s9, 0, -1
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[14:15], s[8:9], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[8:9], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[8:9], 63
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[6:7], s[12:13]
+; GCN-IR-NEXT:    s_or_b64 s[6:7], s[12:13], s[14:15]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[6:7]
 ; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[14:15]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[14:15], s[8:9], 63
-; GCN-IR-NEXT:    s_xor_b64 s[16:17], s[12:13], -1
-; GCN-IR-NEXT:    s_and_b64 s[14:15], s[16:17], s[14:15]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[14:15]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB10_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
@@ -1827,7 +1824,7 @@
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v1, v1
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
@@ -1884,7 +1881,7 @@
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s0
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, s8
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v2, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-IR-NEXT:    s_mov_b32 s5, s1
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
diff --git a/llvm/test/CodeGen/AMDGPU/srem64.ll b/llvm/test/CodeGen/AMDGPU/srem64.ll
--- a/llvm/test/CodeGen/AMDGPU/srem64.ll
+++ b/llvm/test/CodeGen/AMDGPU/srem64.ll
@@ -140,12 +140,11 @@
 ; GCN-IR-NEXT:    s_sub_u32 s8, s10, s14
 ; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    s_mov_b32 s11, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[8:9], 63
 ; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[16:17]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    s_xor_b64 s[18:19], s[12:13], -1
-; GCN-IR-NEXT:    s_and_b64 s[16:17], s[18:19], s[16:17]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[16:17]
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[18:19]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
+; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
@@ -202,8 +201,8 @@
 ; GCN-IR-NEXT:    v_mul_lo_u32 v3, s5, v0
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, s4, v0
 ; GCN-IR-NEXT:    s_mov_b32 s11, 0xf000
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
 ; GCN-IR-NEXT:    s_mov_b32 s10, -1
@@ -505,7 +504,7 @@
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
 ; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
@@ -576,7 +575,7 @@
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
 ; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
@@ -701,7 +700,7 @@
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
 ; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
@@ -839,7 +838,7 @@
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s3, v0
@@ -995,17 +994,17 @@
 ; GCN-IR-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[6:7], 31
-; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 31
-; GCN-IR-NEXT:    s_ashr_i32 s0, s7, 31
-; GCN-IR-NEXT:    s_ashr_i32 s6, s1, 31
+; GCN-IR-NEXT:    s_ashr_i64 s[6:7], s[0:1], 31
+; GCN-IR-NEXT:    s_ashr_i32 s0, s3, 31
+; GCN-IR-NEXT:    s_ashr_i32 s10, s7, 31
 ; GCN-IR-NEXT:    s_mov_b32 s1, s0
-; GCN-IR-NEXT:    s_mov_b32 s7, s6
+; GCN-IR-NEXT:    s_mov_b32 s11, s10
 ; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[2:3], s[0:1]
-; GCN-IR-NEXT:    s_xor_b64 s[8:9], s[8:9], s[6:7]
+; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[6:7], s[10:11]
 ; GCN-IR-NEXT:    s_sub_u32 s2, s2, s0
 ; GCN-IR-NEXT:    s_subb_u32 s3, s3, s0
-; GCN-IR-NEXT:    s_sub_u32 s8, s8, s6
-; GCN-IR-NEXT:    s_subb_u32 s9, s9, s6
+; GCN-IR-NEXT:    s_sub_u32 s8, s6, s10
+; GCN-IR-NEXT:    s_subb_u32 s9, s7, s10
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[8:9], 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[2:3], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
@@ -1021,12 +1020,11 @@
 ; GCN-IR-NEXT:    s_sub_u32 s10, s12, s16
 ; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[18:19], s[10:11], 63
-; GCN-IR-NEXT:    s_mov_b32 s13, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[10:11], 63
 ; GCN-IR-NEXT:    s_or_b64 s[14:15], s[14:15], s[18:19]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[18:19], s[10:11], 63
-; GCN-IR-NEXT:    s_xor_b64 s[20:21], s[14:15], -1
-; GCN-IR-NEXT:    s_and_b64 s[18:19], s[20:21], s[18:19]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[18:19]
+; GCN-IR-NEXT:    s_or_b64 s[18:19], s[14:15], s[20:21]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[18:19]
+; GCN-IR-NEXT:    s_mov_b32 s13, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB8_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s14, s10, 1
@@ -1152,17 +1150,17 @@
 ; GCN-IR-NEXT:    s_sext_i32_i16 s3, s3
 ; GCN-IR-NEXT:    s_sext_i32_i16 s5, s5
 ; GCN-IR-NEXT:    s_ashr_i64 s[6:7], s[2:3], 24
-; GCN-IR-NEXT:    s_ashr_i32 s2, s3, 31
-; GCN-IR-NEXT:    s_ashr_i32 s10, s5, 31
-; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[4:5], 24
+; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[4:5], 24
+; GCN-IR-NEXT:    s_ashr_i32 s2, s7, 31
+; GCN-IR-NEXT:    s_ashr_i32 s8, s5, 31
 ; GCN-IR-NEXT:    s_mov_b32 s3, s2
-; GCN-IR-NEXT:    s_mov_b32 s11, s10
-; GCN-IR-NEXT:    s_xor_b64 s[4:5], s[6:7], s[2:3]
-; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[8:9], s[10:11]
-; GCN-IR-NEXT:    s_sub_u32 s4, s4, s2
-; GCN-IR-NEXT:    s_subb_u32 s5, s5, s2
-; GCN-IR-NEXT:    s_sub_u32 s6, s6, s10
-; GCN-IR-NEXT:    s_subb_u32 s7, s7, s10
+; GCN-IR-NEXT:    s_mov_b32 s9, s8
+; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[6:7], s[2:3]
+; GCN-IR-NEXT:    s_xor_b64 s[10:11], s[4:5], s[8:9]
+; GCN-IR-NEXT:    s_sub_u32 s4, s6, s2
+; GCN-IR-NEXT:    s_subb_u32 s5, s7, s2
+; GCN-IR-NEXT:    s_sub_u32 s6, s10, s8
+; GCN-IR-NEXT:    s_subb_u32 s7, s11, s8
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[6:7], 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[4:5], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
@@ -1178,12 +1176,11 @@
 ; GCN-IR-NEXT:    s_sub_u32 s10, s12, s16
 ; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[18:19], s[10:11], 63
-; GCN-IR-NEXT:    s_mov_b32 s13, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[10:11], 63
 ; GCN-IR-NEXT:    s_or_b64 s[14:15], s[14:15], s[18:19]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[18:19], s[10:11], 63
-; GCN-IR-NEXT:    s_xor_b64 s[20:21], s[14:15], -1
-; GCN-IR-NEXT:    s_and_b64 s[18:19], s[20:21], s[18:19]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[18:19]
+; GCN-IR-NEXT:    s_or_b64 s[18:19], s[14:15], s[20:21]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[18:19]
+; GCN-IR-NEXT:    s_mov_b32 s13, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB9_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s14, s10, 1
@@ -1380,20 +1377,19 @@
 ; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[2:3], s[6:7]
 ; GCN-IR-NEXT:    s_sub_u32 s4, s2, s6
 ; GCN-IR-NEXT:    s_subb_u32 s5, s3, s6
-; GCN-IR-NEXT:    s_flbit_i32_b32 s2, s4
-; GCN-IR-NEXT:    s_add_i32 s2, s2, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s3, s5
-; GCN-IR-NEXT:    s_min_u32 s8, s2, s3
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s4
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s5
+; GCN-IR-NEXT:    s_min_u32 s8, s6, s7
 ; GCN-IR-NEXT:    s_add_u32 s6, s8, 0xffffffc5
 ; GCN-IR-NEXT:    s_addc_u32 s7, 0, -1
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[4:5], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[4:5], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[10:11], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[2:3], s[10:11]
+; GCN-IR-NEXT:    s_or_b64 s[2:3], s[10:11], s[12:13]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[2:3]
 ; GCN-IR-NEXT:    s_mov_b64 s[2:3], 0
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[10:11], s[12:13]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_xor_b64 s[14:15], s[10:11], -1
-; GCN-IR-NEXT:    s_and_b64 s[12:13], s[14:15], s[12:13]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[12:13]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB10_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s10, s6, 1
@@ -1997,7 +1993,7 @@
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v1, v1
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v0|
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, 24, v0
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
@@ -2059,7 +2055,7 @@
 ; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, s4
 ; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
 ; GCN-IR-NEXT:    s_movk_i32 s3, 0x5b7f
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s3
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    s_mov_b32 s5, s1
diff --git a/llvm/test/CodeGen/AMDGPU/udiv64.ll b/llvm/test/CodeGen/AMDGPU/udiv64.ll
--- a/llvm/test/CodeGen/AMDGPU/udiv64.ll
+++ b/llvm/test/CodeGen/AMDGPU/udiv64.ll
@@ -141,12 +141,11 @@
 ; GCN-IR-NEXT:    s_sub_u32 s8, s10, s14
 ; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    s_mov_b32 s11, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[8:9], 63
 ; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[16:17]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    s_xor_b64 s[18:19], s[12:13], -1
-; GCN-IR-NEXT:    s_and_b64 s[16:17], s[18:19], s[16:17]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[16:17]
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[18:19]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
+; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
@@ -816,12 +815,11 @@
 ; GCN-IR-NEXT:    s_sub_u32 s6, s10, s14
 ; GCN-IR-NEXT:    s_subb_u32 s7, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[6:7], 63
-; GCN-IR-NEXT:    s_mov_b32 s11, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[6:7], 63
 ; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[16:17]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[16:17], s[6:7], 63
-; GCN-IR-NEXT:    s_xor_b64 s[18:19], s[12:13], -1
-; GCN-IR-NEXT:    s_and_b64 s[16:17], s[18:19], s[16:17]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[16:17]
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[18:19]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
+; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB7_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s12, s6, 1
@@ -995,20 +993,19 @@
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_flbit_i32_b32 s4, s2
-; GCN-IR-NEXT:    s_flbit_i32_b32 s5, s3
-; GCN-IR-NEXT:    s_add_i32 s4, s4, 32
-; GCN-IR-NEXT:    s_min_u32 s8, s4, s5
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_min_u32 s8, s6, s7
 ; GCN-IR-NEXT:    s_add_u32 s6, s8, 0xffffffc5
 ; GCN-IR-NEXT:    s_addc_u32 s7, 0, -1
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[10:11], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[4:5], s[10:11]
+; GCN-IR-NEXT:    s_or_b64 s[4:5], s[10:11], s[12:13]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[4:5]
 ; GCN-IR-NEXT:    s_mov_b64 s[4:5], 0
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[10:11], s[12:13]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_xor_b64 s[14:15], s[10:11], -1
-; GCN-IR-NEXT:    s_and_b64 s[12:13], s[14:15], s[12:13]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[12:13]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB8_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s10, s6, 1
@@ -1444,20 +1441,19 @@
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_flbit_i32_b32 s4, s2
-; GCN-IR-NEXT:    s_flbit_i32_b32 s5, s3
-; GCN-IR-NEXT:    s_add_i32 s4, s4, 32
-; GCN-IR-NEXT:    s_min_u32 s10, s4, s5
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_min_u32 s10, s6, s7
 ; GCN-IR-NEXT:    s_sub_u32 s6, 59, s10
 ; GCN-IR-NEXT:    s_subb_u32 s7, 0, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[8:9], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    s_or_b64 s[8:9], s[4:5], s[8:9]
+; GCN-IR-NEXT:    s_or_b64 s[4:5], s[8:9], s[12:13]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[4:5]
 ; GCN-IR-NEXT:    s_mov_b64 s[4:5], 0
-; GCN-IR-NEXT:    s_or_b64 s[8:9], s[8:9], s[12:13]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_xor_b64 s[14:15], s[8:9], -1
-; GCN-IR-NEXT:    s_and_b64 s[12:13], s[14:15], s[12:13]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[12:13]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB11_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s8, s6, 1
diff --git a/llvm/test/CodeGen/AMDGPU/urem64.ll b/llvm/test/CodeGen/AMDGPU/urem64.ll
--- a/llvm/test/CodeGen/AMDGPU/urem64.ll
+++ b/llvm/test/CodeGen/AMDGPU/urem64.ll
@@ -140,12 +140,11 @@
 ; GCN-IR-NEXT:    s_sub_u32 s8, s10, s14
 ; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    s_mov_b32 s11, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[8:9], 63
 ; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[16:17]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    s_xor_b64 s[18:19], s[12:13], -1
-; GCN-IR-NEXT:    s_and_b64 s[16:17], s[18:19], s[16:17]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[16:17]
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[18:19]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
+; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
@@ -202,8 +201,8 @@
 ; GCN-IR-NEXT:    v_mul_lo_u32 v3, s5, v0
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, s4, v0
 ; GCN-IR-NEXT:    s_mov_b32 s11, 0xf000
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
 ; GCN-IR-NEXT:    s_mov_b32 s10, -1
@@ -817,20 +816,19 @@
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_flbit_i32_b32 s4, s2
-; GCN-IR-NEXT:    s_flbit_i32_b32 s5, s3
-; GCN-IR-NEXT:    s_add_i32 s4, s4, 32
-; GCN-IR-NEXT:    s_min_u32 s8, s4, s5
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_min_u32 s8, s6, s7
 ; GCN-IR-NEXT:    s_add_u32 s6, s8, 0xffffffc5
 ; GCN-IR-NEXT:    s_addc_u32 s7, 0, -1
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[10:11], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[4:5], s[10:11]
+; GCN-IR-NEXT:    s_or_b64 s[4:5], s[10:11], s[12:13]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[4:5]
 ; GCN-IR-NEXT:    s_mov_b64 s[4:5], 0
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[10:11], s[12:13]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_xor_b64 s[14:15], s[10:11], -1
-; GCN-IR-NEXT:    s_and_b64 s[12:13], s[14:15], s[12:13]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[12:13]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB6_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s10, s6, 1
@@ -1005,20 +1003,19 @@
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_flbit_i32_b32 s4, s2
-; GCN-IR-NEXT:    s_flbit_i32_b32 s5, s3
-; GCN-IR-NEXT:    s_add_i32 s4, s4, 32
-; GCN-IR-NEXT:    s_min_u32 s8, s4, s5
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_min_u32 s8, s6, s7
 ; GCN-IR-NEXT:    s_sub_u32 s6, 59, s8
 ; GCN-IR-NEXT:    s_subb_u32 s7, 0, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[10:11], s[6:7], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[4:5], s[10:11]
+; GCN-IR-NEXT:    s_or_b64 s[4:5], s[10:11], s[12:13]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[4:5]
 ; GCN-IR-NEXT:    s_mov_b64 s[4:5], 0
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[10:11], s[12:13]
-; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_xor_b64 s[14:15], s[10:11], -1
-; GCN-IR-NEXT:    s_and_b64 s[12:13], s[14:15], s[12:13]
-; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[12:13]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB7_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s10, s6, 1
@@ -1071,7 +1068,7 @@
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, 24
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0