[llvm] [GlobalISel] add KnownBits tracking for G_UMULH and G_SMULH (PR #158445)
Pragyansh Chaturvedi via llvm-commits
llvm-commits at lists.llvm.org
Sat Sep 20 13:22:19 PDT 2025
https://github.com/r41k0u updated https://github.com/llvm/llvm-project/pull/158445
>From 3574456e98da8b83b2da95e5cbac49d3c3e915ee Mon Sep 17 00:00:00 2001
From: Pragyansh Chaturvedi <pragyanshchaturvedi18 at gmail.com>
Date: Sat, 13 Sep 2025 22:04:51 +0000
Subject: [PATCH 1/7] [GlobalISel] add KnownBits tracking for G_UMULH and
G_SMULH
---
.../CodeGen/GlobalISel/GISelValueTracking.cpp | 16 ++++++++++++++++
1 file changed, 16 insertions(+)
diff --git a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
index 0cf44e02254de..1b8d192e53fcd 100644
--- a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
@@ -366,6 +366,22 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known,
Known = KnownBits::mul(Known, Known2);
break;
}
+ case TargetOpcode::G_UMULH: {
+ computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
+ Depth + 1);
+ computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
+ Depth + 1);
+ Known = KnownBits::mulhu(Known, Known2);
+ break;
+ }
+ case TargetOpcode::G_SMULH: {
+ computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
+ Depth + 1);
+ computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
+ Depth + 1);
+ Known = KnownBits::mulhs(Known, Known2);
+ break;
+ }
case TargetOpcode::G_SELECT: {
computeKnownBitsMin(MI.getOperand(2).getReg(), MI.getOperand(3).getReg(),
Known, DemandedElts, Depth + 1);
>From 468f4e48dc203f836cba188889ee7b214d3d4023 Mon Sep 17 00:00:00 2001
From: Pragyansh Chaturvedi <pragyanshchaturvedi18 at gmail.com>
Date: Sun, 14 Sep 2025 20:59:09 +0000
Subject: [PATCH 2/7] [GlobalISel] add GISelValueTracking tests for G_UMULH and
G_SMULH
---
.../AArch64/GlobalISel/knownbits-smulh.mir | 137 ++++++++++++++++++
.../AArch64/GlobalISel/knownbits-umulh.mir | 137 ++++++++++++++++++
2 files changed, 274 insertions(+)
create mode 100644 llvm/test/CodeGen/AArch64/GlobalISel/knownbits-smulh.mir
create mode 100644 llvm/test/CodeGen/AArch64/GlobalISel/knownbits-umulh.mir
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-smulh.mir b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-smulh.mir
new file mode 100644
index 0000000000000..b9cde9587c78c
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-smulh.mir
@@ -0,0 +1,137 @@
+# NOTE: Assertions have been autogenerated by utils/update_givaluetracking_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=aarch64 -passes='print<gisel-value-tracking>' -filetype=null %s 2>&1 | FileCheck %s
+
+---
+name: Cst
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @Cst
+ ; CHECK-NEXT: %0:_ KnownBits:00010011 SignBits:3
+ ; CHECK-NEXT: %1:_ KnownBits:11100000 SignBits:3
+ ; CHECK-NEXT: %2:_ KnownBits:11111101 SignBits:6
+ %0:_(s8) = G_CONSTANT i8 19
+ %1:_(s8) = G_CONSTANT i8 224
+ %2:_(s8) = G_SMULH %0, %1
+...
+---
+name: CstZero
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @CstZero
+ ; CHECK-NEXT: %0:_ KnownBits:11111111 SignBits:8
+ ; CHECK-NEXT: %1:_ KnownBits:00000000 SignBits:8
+ ; CHECK-NEXT: %2:_ KnownBits:00000000 SignBits:8
+ %0:_(s8) = G_CONSTANT i8 255
+ %1:_(s8) = G_CONSTANT i8 0
+ %2:_(s8) = G_SMULH %0, %1
+...
+---
+name: ScalarVar
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @ScalarVar
+ ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %1:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1
+ %0:_(s8) = COPY $b0
+ %1:_(s8) = COPY $b1
+ %2:_(s8) = G_SMULH %0, %1
+...
+---
+name: ScalarZero
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @ScalarZero
+ ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %1:_ KnownBits:00000000 SignBits:8
+ ; CHECK-NEXT: %2:_ KnownBits:00000000 SignBits:8
+ %0:_(s8) = COPY $b0
+ %1:_(s8) = G_CONSTANT i8 0
+ %2:_(s8) = G_SMULH %0, %1
+...
+---
+name: ScalarVarAbs
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @ScalarVarAbs
+ ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %1:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %2:_ KnownBits:???????????????? SignBits:9
+ ; CHECK-NEXT: %3:_ KnownBits:0000000000000001 SignBits:15
+ ; CHECK-NEXT: %4:_ KnownBits:???????????????? SignBits:1
+ %0:_(s8) = COPY $b0
+ %1:_(s8) = G_ABS %0
+ %2:_(s16) = G_SEXT %1
+ %3:_(s16) = G_CONSTANT i16 1
+ %4:_(s16) = G_SMULH %2, %3
+...
+---
+name: SplatVecCst
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @SplatVecCst
+ ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %2:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %3:_ KnownBits:00000000 SignBits:8
+ %0:_(s8) = G_CONSTANT i8 250
+ %1:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8)
+ %2:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8)
+ %3:_(<vscale x 16 x s8>) = G_SMULH %1, %2
+...
+---
+name: SplatVecPartScalar
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @SplatVecPartScalar
+ ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %3:_ KnownBits:00001111 SignBits:4
+ ; CHECK-NEXT: %4:_ KnownBits:0000???? SignBits:4
+ ; CHECK-NEXT: %5:_ KnownBits:0000???? SignBits:4
+ ; CHECK-NEXT: %6:_ KnownBits:???????? SignBits:1
+ %0:_(s8) = G_CONSTANT i8 250
+ %1:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8)
+ %2:_(s8) = G_IMPLICIT_DEF
+ %3:_(s8) = G_CONSTANT i8 15
+ %4:_(s8) = G_AND %2, %3
+ %5:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %4(s8)
+ %6:_(<vscale x 16 x s8>) = G_SMULH %1, %5
+...
+---
+name: VecCst
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @VecCst
+ ; CHECK-NEXT: %0:_ KnownBits:00011001 SignBits:3
+ ; CHECK-NEXT: %1:_ KnownBits:11100001 SignBits:3
+ ; CHECK-NEXT: %2:_ KnownBits:?????001 SignBits:3
+ ; CHECK-NEXT: %3:_ KnownBits:?????001 SignBits:3
+ ; CHECK-NEXT: %4:_ KnownBits:???????? SignBits:1
+ %0:_(s8) = G_CONSTANT i8 25
+ %1:_(s8) = G_CONSTANT i8 225
+ %2:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %1:_(s8)
+ %3:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %1:_(s8)
+ %4:_(<2 x s8>) = G_SMULH %2, %3
+...
+---
+name: VecPartScalar
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @VecPartScalar
+ ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %3:_ KnownBits:00001111 SignBits:4
+ ; CHECK-NEXT: %4:_ KnownBits:0000???? SignBits:4
+ ; CHECK-NEXT: %5:_ KnownBits:0000???? SignBits:4
+ ; CHECK-NEXT: %6:_ KnownBits:???????? SignBits:1
+ %0:_(s8) = G_CONSTANT i8 250
+ %1:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %0:_(s8)
+ %2:_(s8) = G_IMPLICIT_DEF
+ %3:_(s8) = G_CONSTANT i8 15
+ %4:_(s8) = G_AND %2, %3
+ %5:_(<2 x s8>) = G_BUILD_VECTOR %4:_(s8), %4:_(s8)
+ %6:_(<2 x s8>) = G_SMULH %1, %5
+...
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-umulh.mir b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-umulh.mir
new file mode 100644
index 0000000000000..debdbaaeecf5c
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-umulh.mir
@@ -0,0 +1,137 @@
+# NOTE: Assertions have been autogenerated by utils/update_givaluetracking_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=aarch64 -passes='print<gisel-value-tracking>' -filetype=null %s 2>&1 | FileCheck %s
+
+---
+name: Cst
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @Cst
+ ; CHECK-NEXT: %0:_ KnownBits:00010011 SignBits:3
+ ; CHECK-NEXT: %1:_ KnownBits:11100000 SignBits:3
+ ; CHECK-NEXT: %2:_ KnownBits:00010000 SignBits:3
+ %0:_(s8) = G_CONSTANT i8 19
+ %1:_(s8) = G_CONSTANT i8 224
+ %2:_(s8) = G_UMULH %0, %1
+...
+---
+name: CstZero
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @CstZero
+ ; CHECK-NEXT: %0:_ KnownBits:11111111 SignBits:8
+ ; CHECK-NEXT: %1:_ KnownBits:00000000 SignBits:8
+ ; CHECK-NEXT: %2:_ KnownBits:00000000 SignBits:8
+ %0:_(s8) = G_CONSTANT i8 255
+ %1:_(s8) = G_CONSTANT i8 0
+ %2:_(s8) = G_UMULH %0, %1
+...
+---
+name: ScalarVar
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @ScalarVar
+ ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %1:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1
+ %0:_(s8) = COPY $b0
+ %1:_(s8) = COPY $b1
+ %2:_(s8) = G_UMULH %0, %1
+...
+---
+name: ScalarZero
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @ScalarZero
+ ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %1:_ KnownBits:00000000 SignBits:8
+ ; CHECK-NEXT: %2:_ KnownBits:00000000 SignBits:8
+ %0:_(s8) = COPY $b0
+ %1:_(s8) = G_CONSTANT i8 0
+ %2:_(s8) = G_UMULH %0, %1
+...
+---
+name: ScalarVarAbs
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @ScalarVarAbs
+ ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %1:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %2:_ KnownBits:???????????????? SignBits:9
+ ; CHECK-NEXT: %3:_ KnownBits:0000000000000001 SignBits:15
+ ; CHECK-NEXT: %4:_ KnownBits:0000000000000000 SignBits:16
+ %0:_(s8) = COPY $b0
+ %1:_(s8) = G_ABS %0
+ %2:_(s16) = G_SEXT %1
+ %3:_(s16) = G_CONSTANT i16 1
+ %4:_(s16) = G_UMULH %2, %3
+...
+---
+name: SplatVecCst
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @SplatVecCst
+ ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %2:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %3:_ KnownBits:11110100 SignBits:4
+ %0:_(s8) = G_CONSTANT i8 250
+ %1:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8)
+ %2:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8)
+ %3:_(<vscale x 16 x s8>) = G_UMULH %1, %2
+...
+---
+name: SplatVecPartScalar
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @SplatVecPartScalar
+ ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %3:_ KnownBits:00001111 SignBits:4
+ ; CHECK-NEXT: %4:_ KnownBits:0000???? SignBits:4
+ ; CHECK-NEXT: %5:_ KnownBits:0000???? SignBits:4
+ ; CHECK-NEXT: %6:_ KnownBits:0000???? SignBits:4
+ %0:_(s8) = G_CONSTANT i8 250
+ %1:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8)
+ %2:_(s8) = G_IMPLICIT_DEF
+ %3:_(s8) = G_CONSTANT i8 15
+ %4:_(s8) = G_AND %2, %3
+ %5:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %4(s8)
+ %6:_(<vscale x 16 x s8>) = G_UMULH %1, %5
+...
+---
+name: VecCst
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @VecCst
+ ; CHECK-NEXT: %0:_ KnownBits:00011001 SignBits:3
+ ; CHECK-NEXT: %1:_ KnownBits:11100001 SignBits:3
+ ; CHECK-NEXT: %2:_ KnownBits:?????001 SignBits:3
+ ; CHECK-NEXT: %3:_ KnownBits:?????001 SignBits:3
+ ; CHECK-NEXT: %4:_ KnownBits:???????? SignBits:1
+ %0:_(s8) = G_CONSTANT i8 25
+ %1:_(s8) = G_CONSTANT i8 225
+ %2:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %1:_(s8)
+ %3:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %1:_(s8)
+ %4:_(<2 x s8>) = G_UMULH %2, %3
+...
+---
+name: VecPartScalar
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: @VecPartScalar
+ ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5
+ ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1
+ ; CHECK-NEXT: %3:_ KnownBits:00001111 SignBits:4
+ ; CHECK-NEXT: %4:_ KnownBits:0000???? SignBits:4
+ ; CHECK-NEXT: %5:_ KnownBits:0000???? SignBits:4
+ ; CHECK-NEXT: %6:_ KnownBits:0000???? SignBits:4
+ %0:_(s8) = G_CONSTANT i8 250
+ %1:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %0:_(s8)
+ %2:_(s8) = G_IMPLICIT_DEF
+ %3:_(s8) = G_CONSTANT i8 15
+ %4:_(s8) = G_AND %2, %3
+ %5:_(<2 x s8>) = G_BUILD_VECTOR %4:_(s8), %4:_(s8)
+ %6:_(<2 x s8>) = G_UMULH %1, %5
+...
>From 27526494f9ab2b4e7a7b4b44ad85c2b6cd2e0143 Mon Sep 17 00:00:00 2001
From: Pragyansh Chaturvedi <pragyanshchaturvedi18 at gmail.com>
Date: Sun, 14 Sep 2025 21:22:37 +0000
Subject: [PATCH 3/7] [GlobalISel] update CodeGen/AArch64/pr58431.ll to agree
with KnownBits changes
---
llvm/test/CodeGen/AArch64/pr58431.ll | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/test/CodeGen/AArch64/pr58431.ll b/llvm/test/CodeGen/AArch64/pr58431.ll
index 467ceb062f249..a37300432bca0 100644
--- a/llvm/test/CodeGen/AArch64/pr58431.ll
+++ b/llvm/test/CodeGen/AArch64/pr58431.ll
@@ -9,7 +9,7 @@ define i32 @f(i64 %0) {
; CHECK-NEXT: mov w10, #10 // =0xa
; CHECK-NEXT: eor x8, x8, #0x8000000000000003
; CHECK-NEXT: umulh x8, x9, x8
-; CHECK-NEXT: msub x0, x8, x10, x9
+; CHECK-NEXT: umsubl x0, w8, w10, x9
; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%2 = trunc i64 %0 to i32
>From 25c37de60e54370ee948c59cf86abf8666e55ef4 Mon Sep 17 00:00:00 2001
From: Pragyansh Chaturvedi <pragyanshchaturvedi18 at gmail.com>
Date: Fri, 19 Sep 2025 12:03:14 +0000
Subject: [PATCH 4/7] [GlobalISel] Update AMDGPU tests after G_UMULH/G_SMULH
knownbits change
---
.../CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll | 325 ++++++++---------
.../CodeGen/AMDGPU/GlobalISel/srem.i64.ll | 344 +++++++++---------
.../CodeGen/AMDGPU/GlobalISel/udiv.i64.ll | 108 +++---
.../CodeGen/AMDGPU/GlobalISel/urem.i64.ll | 32 +-
4 files changed, 388 insertions(+), 421 deletions(-)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
index f57fc005b994b..c688db7c7bff3 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
@@ -2534,202 +2534,195 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v4
; GISEL-NEXT: v_cvt_f32_u32_e32 v3, v1
-; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v5, 0
-; GISEL-NEXT: v_sub_i32_e32 v10, vcc, 0, v1
-; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v5
+; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v9, 0
+; GISEL-NEXT: v_sub_i32_e32 v8, vcc, 0, v1
+; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v9
; GISEL-NEXT: v_rcp_iflag_f32_e32 v3, v3
-; GISEL-NEXT: v_subb_u32_e64 v11, s[4:5], 0, 0, vcc
+; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, 0, vcc
; GISEL-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3
; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v3
-; GISEL-NEXT: v_trunc_f32_e32 v7, v4
-; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v7
-; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v3
-; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v7
-; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v10, v9, 0
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v10, v12, v[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v4, v12, v3
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v9, v[7:8]
-; GISEL-NEXT: v_mul_hi_u32 v8, v9, v3
-; GISEL-NEXT: v_mul_hi_u32 v3, v12, v3
-; GISEL-NEXT: v_mul_lo_u32 v13, v9, v7
-; GISEL-NEXT: v_mul_lo_u32 v14, v12, v7
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13
+; GISEL-NEXT: v_trunc_f32_e32 v5, v4
+; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v5
+; GISEL-NEXT: v_cvt_u32_f32_e32 v7, v3
+; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v5
+; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v8, v7, 0
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v8, v11, v[4:5]
+; GISEL-NEXT: v_mul_hi_u32 v12, v7, v3
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v7, v[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3
+; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3
+; GISEL-NEXT: v_mul_lo_u32 v13, v7, v4
+; GISEL-NEXT: v_mul_lo_u32 v14, v11, v4
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
-; GISEL-NEXT: v_mul_hi_u32 v8, v9, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v13, v4
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12
+; GISEL-NEXT: v_mul_hi_u32 v12, v7, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v14, v3
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v8
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8
-; GISEL-NEXT: v_mul_hi_u32 v7, v12, v7
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v7, v4
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v3
-; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v12, v4, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v10, v9, 0
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v10, v12, v[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v4, v12, v3
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v12
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12
+; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v12, v5
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v3
+; GISEL-NEXT: v_addc_u32_e32 v11, vcc, v11, v4, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v8, v7, 0
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v8, v11, v[4:5]
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v7, v[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3
; GISEL-NEXT: v_and_b32_e32 v10, 0xffffff, v0
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v9, v[7:8]
-; GISEL-NEXT: v_mul_hi_u32 v0, v9, v3
-; GISEL-NEXT: v_mul_hi_u32 v3, v12, v3
-; GISEL-NEXT: v_mul_lo_u32 v8, v9, v7
-; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v2
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
+; GISEL-NEXT: v_mul_lo_u32 v8, v7, v4
+; GISEL-NEXT: v_mul_hi_u32 v0, v7, v3
+; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v4, v12, v7
+; GISEL-NEXT: v_mul_lo_u32 v5, v11, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
-; GISEL-NEXT: v_mul_hi_u32 v8, v9, v7
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
+; GISEL-NEXT: v_mul_hi_u32 v8, v7, v4
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v8
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
-; GISEL-NEXT: v_mul_hi_u32 v7, v12, v7
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
+; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
-; GISEL-NEXT: v_addc_u32_e32 v4, vcc, v12, v3, vcc
-; GISEL-NEXT: v_mul_lo_u32 v7, 0, v0
-; GISEL-NEXT: v_mul_lo_u32 v8, v10, v4
-; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6
-; GISEL-NEXT: v_mul_hi_u32 v6, v10, v0
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v7, v0
+; GISEL-NEXT: v_addc_u32_e32 v3, vcc, v11, v3, vcc
+; GISEL-NEXT: v_mul_lo_u32 v4, 0, v0
+; GISEL-NEXT: v_mul_lo_u32 v5, v10, v3
+; GISEL-NEXT: v_mul_hi_u32 v7, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v8
-; GISEL-NEXT: v_mul_lo_u32 v8, 0, v4
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v7, v6
-; GISEL-NEXT: v_mul_hi_u32 v7, v10, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
+; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v2
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
+; GISEL-NEXT: v_mul_lo_u32 v5, 0, v3
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v7
+; GISEL-NEXT: v_mul_hi_u32 v7, v10, v3
+; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
-; GISEL-NEXT: v_mul_hi_u32 v4, 0, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v1, v0, 0
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v9
-; GISEL-NEXT: v_cvt_f32_u32_e32 v9, v3
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v1, v4, v[7:8]
-; GISEL-NEXT: v_mac_f32_e32 v9, 0x4f800000, v5
-; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v9
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], 0, v0, v[7:8]
-; GISEL-NEXT: v_sub_i32_e32 v8, vcc, v10, v6
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v1, v0, 0
+; GISEL-NEXT: v_mul_hi_u32 v4, 0, v3
+; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6
+; GISEL-NEXT: v_mov_b32_e32 v5, v8
+; GISEL-NEXT: v_cvt_f32_u32_e32 v8, v3
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v4, v[5:6]
+; GISEL-NEXT: v_mac_f32_e32 v8, 0x4f800000, v9
+; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v8
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v0, v[5:6]
+; GISEL-NEXT: v_sub_i32_e32 v9, vcc, v10, v7
; GISEL-NEXT: v_mul_f32_e32 v2, 0x5f7ffffc, v2
-; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v2
-; GISEL-NEXT: v_subb_u32_e64 v9, s[4:5], 0, v7, vcc
-; GISEL-NEXT: v_sub_i32_e64 v10, s[4:5], 0, v7
-; GISEL-NEXT: v_trunc_f32_e32 v7, v5
-; GISEL-NEXT: v_mac_f32_e32 v2, 0xcf800000, v7
+; GISEL-NEXT: v_mul_f32_e32 v6, 0x2f800000, v2
+; GISEL-NEXT: v_trunc_f32_e32 v8, v6
+; GISEL-NEXT: v_mac_f32_e32 v2, 0xcf800000, v8
; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v2
+; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, v5, vcc
; GISEL-NEXT: v_sub_i32_e64 v13, s[4:5], 0, v3
; GISEL-NEXT: v_subb_u32_e64 v14, s[4:5], 0, 0, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v12, 0
-; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v7
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v8, v1
-; GISEL-NEXT: v_mov_b32_e32 v2, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, -1, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v15, v[2:3]
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v9, -1, v16, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v14, v12, v[6:7]
-; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v10, vcc
-; GISEL-NEXT: v_mul_lo_u32 v7, v15, v5
-; GISEL-NEXT: v_mul_lo_u32 v10, v12, v6
-; GISEL-NEXT: v_sub_i32_e32 v8, vcc, v8, v1
-; GISEL-NEXT: v_subbrev_u32_e32 v16, vcc, 0, v2, vcc
-; GISEL-NEXT: v_mul_hi_u32 v2, v12, v5
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v12, 0
+; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v8
+; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v5
+; GISEL-NEXT: v_mov_b32_e32 v2, v7
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v13, v15, v[2:3]
+; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v9, v1
+; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, s[4:5]
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v14, v12, v[7:8]
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v8, -1, v2, s[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v2, v15, v6
+; GISEL-NEXT: v_mul_lo_u32 v10, v12, v7
+; GISEL-NEXT: v_subbrev_u32_e32 v16, vcc, 0, v5, vcc
+; GISEL-NEXT: v_mul_hi_u32 v5, v12, v6
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v7, v2
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v5
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v7, v15, v6
-; GISEL-NEXT: v_mul_hi_u32 v5, v15, v5
+; GISEL-NEXT: v_mul_lo_u32 v5, v15, v7
+; GISEL-NEXT: v_mul_hi_u32 v6, v15, v6
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v10, v2
-; GISEL-NEXT: v_mul_hi_u32 v10, v12, v6
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; GISEL-NEXT: v_mul_hi_u32 v10, v12, v7
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10
-; GISEL-NEXT: v_mul_hi_u32 v6, v15, v6
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10
+; GISEL-NEXT: v_mul_hi_u32 v7, v15, v7
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v12, v2
-; GISEL-NEXT: v_addc_u32_e32 v10, vcc, v15, v5, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v7, 0
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, 1, v0
-; GISEL-NEXT: v_addc_u32_e32 v15, vcc, 0, v4, vcc
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v8, v1
-; GISEL-NEXT: v_mov_b32_e32 v1, v6
-; GISEL-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v13, v10, v[1:2]
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v16
-; GISEL-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v14, v7, v[1:2]
-; GISEL-NEXT: v_cndmask_b32_e32 v6, -1, v8, vcc
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, 1, v12
-; GISEL-NEXT: v_addc_u32_e32 v8, vcc, 0, v15, vcc
-; GISEL-NEXT: v_mul_lo_u32 v13, v10, v5
-; GISEL-NEXT: v_mul_lo_u32 v14, v7, v1
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
-; GISEL-NEXT: v_mul_hi_u32 v6, v7, v5
-; GISEL-NEXT: v_cndmask_b32_e32 v2, v12, v2, vcc
-; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v13, v14
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v12, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, s[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v12, v10, v1
-; GISEL-NEXT: v_mul_hi_u32 v5, v10, v5
-; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v13, v6
-; GISEL-NEXT: v_mul_hi_u32 v13, v7, v1
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v12, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v5, v13
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v12, v13
-; GISEL-NEXT: v_mul_hi_u32 v1, v10, v1
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v5, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v12, v6
-; GISEL-NEXT: v_add_i32_e64 v1, s[4:5], v1, v6
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5
-; GISEL-NEXT: v_addc_u32_e64 v1, s[4:5], v10, v1, s[4:5]
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v12, v2
+; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v15, v5, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v10, 0
+; GISEL-NEXT: v_sub_i32_e32 v9, vcc, v9, v1
+; GISEL-NEXT: v_mov_b32_e32 v2, v6
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v12, v[2:3]
+; GISEL-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v16, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v14, v10, v[6:7]
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, 1, v0
+; GISEL-NEXT: v_addc_u32_e32 v13, vcc, 0, v4, vcc
+; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v9, v1
+; GISEL-NEXT: v_mul_lo_u32 v7, v12, v5
+; GISEL-NEXT: v_mul_lo_u32 v9, v10, v6
+; GISEL-NEXT: v_mul_hi_u32 v14, v10, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, -1, vcc
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v15
+; GISEL-NEXT: v_cndmask_b32_e32 v1, -1, v1, vcc
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v14
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; GISEL-NEXT: v_mul_lo_u32 v14, v12, v6
+; GISEL-NEXT: v_mul_hi_u32 v5, v12, v5
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7
+; GISEL-NEXT: v_mul_hi_u32 v9, v10, v6
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v14, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v14, 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 v9, vcc, v14, v9
+; GISEL-NEXT: v_mul_hi_u32 v6, v12, v6
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v7
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v7
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v10, v5
+; GISEL-NEXT: v_addc_u32_e32 v7, vcc, v12, v6, vcc
; GISEL-NEXT: v_mul_lo_u32 v6, 0, v5
-; GISEL-NEXT: v_mul_lo_u32 v7, v11, v1
-; GISEL-NEXT: v_mul_hi_u32 v10, v11, v5
-; GISEL-NEXT: v_cndmask_b32_e32 v8, v15, v8, vcc
+; GISEL-NEXT: v_mul_lo_u32 v9, v11, v7
+; GISEL-NEXT: v_mul_hi_u32 v14, v11, v5
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, 1, v2
+; GISEL-NEXT: v_addc_u32_e32 v12, vcc, 0, v13, vcc
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v9
+; GISEL-NEXT: v_mul_lo_u32 v9, 0, v7
; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v7
-; GISEL-NEXT: v_mul_lo_u32 v7, 0, v1
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10
-; GISEL-NEXT: v_mul_hi_u32 v10, v11, v1
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v14
+; GISEL-NEXT: v_mul_hi_u32 v14, v11, v7
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v5, v6
-; GISEL-NEXT: v_mul_hi_u32 v1, 0, v1
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v10, 0
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v12
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v1, v7
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v9, v5
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v14
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v5, v6
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v9, 0
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
+; GISEL-NEXT: v_cndmask_b32_e32 v2, v2, v10, vcc
+; GISEL-NEXT: v_mul_hi_u32 v10, 0, v7
; GISEL-NEXT: v_mov_b32_e32 v1, v6
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v3, v12, v[1:2]
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9
+; GISEL-NEXT: v_cndmask_b32_e32 v12, v13, v12, vcc
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v8
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v3, v10, v[1:2]
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v10, v[6:7]
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v8, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v12, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v9, v[6:7]
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v11, v5
; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v6
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], 0, v6, vcc
@@ -2740,8 +2733,8 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
; GISEL-NEXT: v_subbrev_u32_e32 v5, vcc, 0, v5, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v4, -1, v6, s[4:5]
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v10
-; GISEL-NEXT: v_addc_u32_e32 v7, vcc, 0, v12, vcc
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v9
+; GISEL-NEXT: v_addc_u32_e32 v7, vcc, 0, v10, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v5
@@ -2752,8 +2745,8 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_cndmask_b32_e32 v2, v6, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
-; GISEL-NEXT: v_cndmask_b32_e32 v2, v10, v2, vcc
-; GISEL-NEXT: v_cndmask_b32_e32 v3, v12, v3, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v2, v9, v2, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v3, v10, v3, vcc
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_v2i64_24bit:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
index 19dc20c510041..682c5bccbe844 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
@@ -3029,203 +3029,193 @@ define <2 x i64> @v_srem_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v4
; GISEL-NEXT: v_cvt_f32_u32_e32 v3, v1
-; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v4, 0
-; GISEL-NEXT: v_sub_i32_e32 v11, vcc, 0, v1
-; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v4
+; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v7, 0
+; GISEL-NEXT: v_sub_i32_e32 v9, vcc, 0, v1
+; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v7
; GISEL-NEXT: v_rcp_iflag_f32_e32 v3, v3
-; GISEL-NEXT: v_subb_u32_e64 v12, s[4:5], 0, 0, vcc
+; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, 0, vcc
; GISEL-NEXT: v_and_b32_e32 v2, 0xffffff, v2
; GISEL-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3
-; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v3
-; GISEL-NEXT: v_trunc_f32_e32 v5, v5
+; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v3
+; GISEL-NEXT: v_trunc_f32_e32 v5, v4
; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v5
-; GISEL-NEXT: v_cvt_u32_f32_e32 v10, v3
-; GISEL-NEXT: v_cvt_u32_f32_e32 v5, v5
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v10, 0
-; GISEL-NEXT: v_mov_b32_e32 v3, v8
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v11, v5, v[3:4]
-; GISEL-NEXT: v_mul_lo_u32 v3, v5, v7
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v12, v10, v[8:9]
-; GISEL-NEXT: v_mul_hi_u32 v9, v10, v7
-; GISEL-NEXT: v_mul_hi_u32 v7, v5, v7
-; GISEL-NEXT: v_mul_lo_u32 v13, v10, v8
-; GISEL-NEXT: v_mul_lo_u32 v14, v5, v8
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v13
+; GISEL-NEXT: v_cvt_u32_f32_e32 v8, v3
+; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v5
+; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v9, v8, 0
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v9, v11, v[4:5]
+; GISEL-NEXT: v_mul_hi_u32 v12, v8, v3
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v8, v[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3
+; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3
+; GISEL-NEXT: v_mul_lo_u32 v13, v8, v4
+; GISEL-NEXT: v_mul_lo_u32 v14, v11, v4
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9
-; GISEL-NEXT: v_mul_hi_u32 v9, v10, v8
-; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v13, v3
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v14, v7
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12
+; GISEL-NEXT: v_mul_hi_u32 v12, v8, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v14, v3
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v13, v9
-; GISEL-NEXT: v_mul_hi_u32 v8, v5, v8
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v8, v7
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v3
-; GISEL-NEXT: v_addc_u32_e32 v5, vcc, v5, v7, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v10, 0
-; GISEL-NEXT: v_mov_b32_e32 v3, v8
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v11, v5, v[3:4]
-; GISEL-NEXT: v_mul_lo_u32 v3, v5, v7
-; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v0
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v12, v10, v[8:9]
-; GISEL-NEXT: v_mul_hi_u32 v0, v10, v7
-; GISEL-NEXT: v_mul_hi_u32 v7, v5, v7
-; GISEL-NEXT: v_mul_lo_u32 v9, v10, v8
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v12
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12
+; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v12, v5
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v3
+; GISEL-NEXT: v_addc_u32_e32 v11, vcc, v11, v4, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v9, v8, 0
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v9, v11, v[4:5]
+; GISEL-NEXT: v_mul_hi_u32 v9, v8, v3
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v8, v[4:5]
+; GISEL-NEXT: v_and_b32_e32 v10, 0xffffff, v0
+; GISEL-NEXT: v_mul_lo_u32 v0, v11, v3
+; GISEL-NEXT: v_mul_lo_u32 v5, v8, v4
+; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v9
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v3, v5, v8
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
-; GISEL-NEXT: v_mul_hi_u32 v9, v10, v8
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9
+; GISEL-NEXT: v_mul_lo_u32 v9, v11, v4
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
+; GISEL-NEXT: v_mul_hi_u32 v5, v8, v4
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v9, v3
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
-; GISEL-NEXT: v_mul_hi_u32 v8, v5, v8
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v9, v5
+; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v8, v3
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0
-; GISEL-NEXT: v_addc_u32_e32 v5, vcc, v5, v3, vcc
-; GISEL-NEXT: v_mul_lo_u32 v7, 0, v0
-; GISEL-NEXT: v_mul_lo_u32 v8, v11, v5
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
+; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v11, v3, vcc
+; GISEL-NEXT: v_mul_lo_u32 v4, 0, v0
+; GISEL-NEXT: v_mul_lo_u32 v5, v10, v8
; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6
-; GISEL-NEXT: v_mul_hi_u32 v6, v11, v0
+; GISEL-NEXT: v_mul_hi_u32 v6, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v8
-; GISEL-NEXT: v_mul_lo_u32 v8, 0, v5
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v7, v6
-; GISEL-NEXT: v_mul_hi_u32 v7, v11, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v0, v6
-; GISEL-NEXT: v_mul_hi_u32 v9, 0, v5
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v8, 0
-; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v7, v0
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v0
-; GISEL-NEXT: v_mov_b32_e32 v0, v6
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v1, v7, v[0:1]
-; GISEL-NEXT: v_cvt_f32_u32_e32 v0, v3
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v8, v[6:7]
-; GISEL-NEXT: v_mac_f32_e32 v0, 0x4f800000, v4
-; GISEL-NEXT: v_rcp_iflag_f32_e32 v0, v0
-; GISEL-NEXT: v_sub_i32_e32 v7, vcc, v11, v5
-; GISEL-NEXT: v_subb_u32_e64 v8, s[4:5], 0, v6, vcc
-; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v0
-; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v0
-; GISEL-NEXT: v_trunc_f32_e32 v9, v4
-; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v9
-; GISEL-NEXT: v_cvt_u32_f32_e32 v10, v0
-; GISEL-NEXT: v_sub_i32_e64 v11, s[4:5], 0, v3
-; GISEL-NEXT: v_subb_u32_e64 v12, s[4:5], 0, 0, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v10, 0
-; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v9
-; GISEL-NEXT: v_sub_i32_e64 v13, s[4:5], 0, v6
-; GISEL-NEXT: v_mov_b32_e32 v0, v5
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v9, v[0:1]
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v7, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v10, v[5:6]
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v8
-; GISEL-NEXT: v_cndmask_b32_e64 v14, -1, v0, s[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v0, v9, v4
-; GISEL-NEXT: v_mul_lo_u32 v6, v10, v5
-; GISEL-NEXT: v_mul_hi_u32 v15, v10, v4
-; GISEL-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v13, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v15
-; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v15, v9, v5
-; GISEL-NEXT: v_mul_hi_u32 v4, v9, v4
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0
-; GISEL-NEXT: v_mul_hi_u32 v6, v10, v5
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v15, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
+; GISEL-NEXT: v_mul_lo_u32 v5, 0, v8
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v15, v6
-; GISEL-NEXT: v_mul_hi_u32 v5, v9, v5
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
+; GISEL-NEXT: v_mul_hi_u32 v6, v10, v8
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v6, v4
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v4
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v0
-; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v9, v4, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v10, 0
-; GISEL-NEXT: v_sub_i32_e32 v15, vcc, v7, v1
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v0, v4
+; GISEL-NEXT: v_cvt_f32_u32_e32 v0, v3
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v1, v9, 0
+; GISEL-NEXT: v_mul_hi_u32 v6, 0, v8
+; GISEL-NEXT: v_mac_f32_e32 v0, 0x4f800000, v7
+; GISEL-NEXT: v_rcp_iflag_f32_e32 v7, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v5
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v9, v[0:1]
-; GISEL-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v13, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v10, v[5:6]
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v15, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v13
-; GISEL-NEXT: v_cndmask_b32_e32 v0, -1, v16, vcc
-; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v15, v1
-; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v13, vcc
-; GISEL-NEXT: v_mul_lo_u32 v11, v9, v4
-; GISEL-NEXT: v_mul_lo_u32 v12, v10, v5
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0
-; GISEL-NEXT: v_mul_hi_u32 v0, v10, v4
-; GISEL-NEXT: v_mul_hi_u32 v4, v9, v4
-; 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 v0, s[4:5], v11, v0
-; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v11, v9, v5
-; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v12, v0
-; GISEL-NEXT: v_mul_hi_u32 v12, v10, v5
-; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v11, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v11, s[4:5], v11, v12
-; GISEL-NEXT: v_mul_hi_u32 v5, v9, v5
-; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v4, v0
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v6, v[0:1]
+; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v7
+; GISEL-NEXT: v_mul_f32_e32 v7, 0x2f800000, v0
+; GISEL-NEXT: v_trunc_f32_e32 v11, v7
+; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v11
+; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v0
+; GISEL-NEXT: v_sub_i32_e32 v13, vcc, 0, v3
+; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v11
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v13, v12, 0
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v9, v[5:6]
+; GISEL-NEXT: v_mov_b32_e32 v0, v8
+; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v13, v11, v[0:1]
+; GISEL-NEXT: v_subb_u32_e64 v14, s[4:5], 0, 0, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v14, v12, v[8:9]
+; GISEL-NEXT: v_sub_i32_e32 v10, vcc, v10, v4
+; GISEL-NEXT: v_subb_u32_e64 v15, s[4:5], 0, v5, vcc
+; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], 0, v5
+; GISEL-NEXT: v_mul_lo_u32 v4, v11, v7
+; GISEL-NEXT: v_mul_lo_u32 v5, v12, v8
+; GISEL-NEXT: v_mul_hi_u32 v9, v12, v7
+; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v10, v1
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v9
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v11, v4
+; GISEL-NEXT: v_mul_lo_u32 v9, v11, v8
+; GISEL-NEXT: v_mul_hi_u32 v7, v11, v7
; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v5, v4
-; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v10, v0
-; GISEL-NEXT: v_addc_u32_e64 v4, s[4:5], v9, v4, s[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v5, 0, v0
-; GISEL-NEXT: v_mul_lo_u32 v9, v2, v4
-; GISEL-NEXT: v_cndmask_b32_e32 v10, v13, v6, vcc
-; GISEL-NEXT: v_mul_hi_u32 v6, v2, v0
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v9
-; GISEL-NEXT: v_mul_lo_u32 v9, 0, v4
-; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6
-; GISEL-NEXT: v_mul_hi_u32 v6, v2, v4
+; GISEL-NEXT: v_mul_hi_u32 v5, v12, v8
+; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v9, v7
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v9, v7
+; GISEL-NEXT: v_mul_hi_u32 v8, v11, v8
+; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v5, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5
+; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v8, v5
+; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v12, v4
+; GISEL-NEXT: v_addc_u32_e64 v8, s[4:5], v11, v5, s[4:5]
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v13, v7, 0
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v15
+; GISEL-NEXT: v_subbrev_u32_e32 v11, vcc, 0, v0, vcc
+; GISEL-NEXT: v_mov_b32_e32 v0, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v9, -1, v6, s[4:5]
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v8, v[0:1]
+; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v10, v1
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v14, v7, v[5:6]
+; GISEL-NEXT: v_mul_lo_u32 v12, v8, v4
+; GISEL-NEXT: v_subbrev_u32_e32 v11, vcc, 0, v11, vcc
+; GISEL-NEXT: v_mul_lo_u32 v13, v7, v5
+; GISEL-NEXT: v_mul_hi_u32 v14, v7, v4
+; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, vcc
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v11
+; GISEL-NEXT: v_cndmask_b32_e32 v6, -1, v6, vcc
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v13
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v14
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; GISEL-NEXT: v_mul_lo_u32 v14, v8, v5
+; GISEL-NEXT: v_mul_hi_u32 v4, v8, v4
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12
+; GISEL-NEXT: v_mul_hi_u32 v13, v7, v5
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v14, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v13, vcc, v14, v13
+; GISEL-NEXT: v_mul_hi_u32 v5, v8, v5
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v7, v4
+; GISEL-NEXT: v_addc_u32_e32 v7, vcc, v8, v5, vcc
+; GISEL-NEXT: v_mul_lo_u32 v5, 0, v4
+; GISEL-NEXT: v_mul_lo_u32 v8, v2, v7
+; GISEL-NEXT: v_mul_hi_u32 v13, v2, v4
+; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v0, v1
+; GISEL-NEXT: v_subbrev_u32_e32 v12, vcc, 0, v11, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
+; GISEL-NEXT: v_mul_lo_u32 v8, 0, v7
+; GISEL-NEXT: v_mul_hi_u32 v4, 0, v4
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13
+; GISEL-NEXT: v_mul_hi_u32 v13, v2, v7
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v0, v5
-; GISEL-NEXT: v_mul_hi_u32 v11, 0, v4
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v9, 0
-; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v11, v0
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v4, v5
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v8, 0
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
+; GISEL-NEXT: v_mul_hi_u32 v6, 0, v7
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v0, v1, vcc
; GISEL-NEXT: v_mov_b32_e32 v0, v5
+; GISEL-NEXT: v_cndmask_b32_e32 v7, v11, v12, vcc
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v6, v[0:1]
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14
-; GISEL-NEXT: v_cndmask_b32_e32 v0, v7, v1, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v9, v[5:6]
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v8, v10, vcc
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9
+; GISEL-NEXT: v_cndmask_b32_e32 v0, v10, v1, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v8, v[5:6]
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v7, vcc
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v4
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], 0, v5, vcc
; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v5
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
index 4de10788a6bd7..ded985ec3a1ec 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
@@ -2053,90 +2053,82 @@ define <2 x i64> @v_udiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_mul_lo_u32 v12, v6, v2
; GISEL-NEXT: v_mul_lo_u32 v13, 0, v2
; GISEL-NEXT: v_mul_hi_u32 v14, v6, v2
-; GISEL-NEXT: v_mul_hi_u32 v2, 0, v2
-; GISEL-NEXT: v_mul_lo_u32 v15, v0, v5
+; GISEL-NEXT: v_mul_hi_u32 v15, 0, v2
+; GISEL-NEXT: v_mul_lo_u32 v2, v0, v5
; GISEL-NEXT: v_mul_lo_u32 v16, 0, v5
; GISEL-NEXT: v_mul_hi_u32 v17, v0, v5
; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v15
+; GISEL-NEXT: v_mul_lo_u32 v12, v3, v15
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v10, v2
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v16, v7
+; GISEL-NEXT: v_mul_lo_u32 v10, v1, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v14
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v11
+; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v17
-; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8
-; GISEL-NEXT: v_mul_lo_u32 v9, v3, v4
-; GISEL-NEXT: v_mul_lo_u32 v12, 0, v4
-; GISEL-NEXT: v_mul_hi_u32 v13, v3, v4
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10
-; GISEL-NEXT: v_mul_lo_u32 v11, v1, v7
-; GISEL-NEXT: v_mul_lo_u32 v14, 0, v7
-; GISEL-NEXT: v_mul_hi_u32 v15, v1, v7
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v2, v8
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
-; GISEL-NEXT: v_mul_lo_u32 v2, v3, v8
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, 1, v4
-; GISEL-NEXT: v_addc_u32_e32 v16, vcc, 0, v8, vcc
-; GISEL-NEXT: v_mul_lo_u32 v17, v1, v5
-; GISEL-NEXT: v_add_i32_e32 v18, vcc, 1, v7
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v7, v2
+; GISEL-NEXT: v_mul_lo_u32 v7, v3, v4
+; GISEL-NEXT: v_mul_lo_u32 v8, 0, v4
+; GISEL-NEXT: v_mul_hi_u32 v9, v3, v4
+; GISEL-NEXT: v_add_i32_e32 v11, vcc, 1, v4
+; GISEL-NEXT: v_addc_u32_e32 v13, vcc, 0, v15, vcc
+; GISEL-NEXT: v_mul_lo_u32 v14, v1, v2
+; GISEL-NEXT: v_mul_lo_u32 v16, 0, v2
+; GISEL-NEXT: v_mul_hi_u32 v17, v1, v2
+; GISEL-NEXT: v_add_i32_e32 v18, vcc, 1, v2
; GISEL-NEXT: v_addc_u32_e32 v19, vcc, 0, v5, vcc
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v12, v2
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v14, v17
-; GISEL-NEXT: v_add_i32_e32 v14, vcc, 1, v10
-; GISEL-NEXT: v_addc_u32_e32 v17, vcc, 0, v16, vcc
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v13
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v15
-; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v9
-; GISEL-NEXT: v_subb_u32_e64 v9, s[4:5], 0, v2, vcc
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v12
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v16, v10
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, 1, v11
+; GISEL-NEXT: v_addc_u32_e32 v16, vcc, 0, v13, vcc
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v9
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v17
+; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v7
+; GISEL-NEXT: v_subb_u32_e64 v7, s[4:5], 0, v8, vcc
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v6, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, -1, s[4:5]
-; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], v0, v11
-; GISEL-NEXT: v_subb_u32_e64 v11, s[6:7], 0, v12, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, -1, s[4:5]
+; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], v0, v14
+; GISEL-NEXT: v_subb_u32_e64 v14, s[6:7], 0, v9, s[4:5]
; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v0, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, -1, s[6:7]
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v9
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[8:9], 0, v11
-; GISEL-NEXT: v_add_i32_e64 v9, s[10:11], 1, v18
-; GISEL-NEXT: v_addc_u32_e64 v11, s[10:11], 0, v19, s[10:11]
-; GISEL-NEXT: v_sub_i32_e64 v2, s[10:11], 0, v2
-; GISEL-NEXT: v_sub_i32_e64 v12, s[10:11], 0, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v13, -1, v13, s[6:7]
-; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v15, -1, v15, s[8:9]
-; GISEL-NEXT: v_subbrev_u32_e64 v12, vcc, 0, v12, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, -1, s[6:7]
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v7
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[8:9], 0, v14
+; GISEL-NEXT: v_add_i32_e64 v7, s[10:11], 1, v18
+; GISEL-NEXT: v_addc_u32_e64 v14, s[10:11], 0, v19, s[10:11]
+; GISEL-NEXT: v_sub_i32_e64 v8, s[10:11], 0, v8
+; GISEL-NEXT: v_sub_i32_e64 v9, s[10:11], 0, v9
+; GISEL-NEXT: v_cndmask_b32_e64 v10, -1, v10, s[6:7]
+; GISEL-NEXT: v_subbrev_u32_e32 v8, vcc, 0, v8, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v17, -1, v17, s[8:9]
+; GISEL-NEXT: v_subbrev_u32_e64 v9, vcc, 0, v9, s[4:5]
; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v3
-; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
+; GISEL-NEXT: v_subbrev_u32_e32 v8, vcc, 0, v8, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v6, v3
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, -1, vcc
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
-; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v12, vcc
+; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v9, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v2
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v8
; GISEL-NEXT: v_cndmask_b32_e32 v1, -1, v3, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v6
; GISEL-NEXT: v_cndmask_b32_e32 v0, -1, v0, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v10, v14, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v11, v12, vcc
; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v0
-; GISEL-NEXT: v_cndmask_b32_e64 v2, v18, v9, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e32 v3, v16, v17, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v13
+; GISEL-NEXT: v_cndmask_b32_e64 v3, v18, v7, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e32 v6, v13, v16, vcc
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
; GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v4, v19, v11, s[4:5]
-; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v15
-; GISEL-NEXT: v_cndmask_b32_e64 v2, v7, v2, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v8, v3, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v4, v19, v14, s[4:5]
+; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v17
+; GISEL-NEXT: v_cndmask_b32_e64 v2, v2, v3, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v6, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v3, v5, v4, s[4:5]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
index a41ec8e7ce3ea..be5543b9b5b7f 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
@@ -2058,42 +2058,34 @@ define <2 x i64> @v_urem_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8
+; GISEL-NEXT: v_mul_lo_u32 v2, v3, v2
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v15
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v16, v7
+; GISEL-NEXT: v_mul_lo_u32 v5, v1, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v14
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v11
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v17
-; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8
-; GISEL-NEXT: v_mul_lo_u32 v9, v3, v4
-; GISEL-NEXT: v_mul_lo_u32 v12, 0, v4
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
+; GISEL-NEXT: v_mul_lo_u32 v8, v3, v4
+; GISEL-NEXT: v_mul_lo_u32 v9, 0, v4
; GISEL-NEXT: v_mul_hi_u32 v4, v3, v4
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10
-; GISEL-NEXT: v_mul_lo_u32 v11, v1, v7
-; GISEL-NEXT: v_mul_lo_u32 v13, 0, v7
+; GISEL-NEXT: v_mul_lo_u32 v10, v1, v7
+; GISEL-NEXT: v_mul_lo_u32 v11, 0, v7
; GISEL-NEXT: v_mul_hi_u32 v7, v1, v7
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v8
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
-; GISEL-NEXT: v_mul_lo_u32 v2, v3, v2
-; GISEL-NEXT: v_mul_lo_u32 v5, v1, v5
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v12, v2
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v9, v2
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v11, v5
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v4
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v7
-; GISEL-NEXT: v_sub_i32_e32 v5, vcc, v6, v9
+; GISEL-NEXT: v_sub_i32_e32 v5, vcc, v6, v8
; GISEL-NEXT: v_subb_u32_e64 v6, s[4:5], 0, v2, vcc
; GISEL-NEXT: v_sub_i32_e64 v2, s[4:5], 0, v2
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v5, v3
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
-; GISEL-NEXT: v_sub_i32_e64 v8, s[4:5], v0, v11
+; GISEL-NEXT: v_sub_i32_e64 v8, s[4:5], v0, v10
; GISEL-NEXT: v_subb_u32_e64 v9, s[6:7], 0, v4, s[4:5]
; GISEL-NEXT: v_sub_i32_e64 v0, s[6:7], 0, v4
; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v8, v1
>From 7020b7892fe6e880e4315e5d1d82a935cbb057ae Mon Sep 17 00:00:00 2001
From: Pragyansh Chaturvedi <pragyanshchaturvedi18 at gmail.com>
Date: Fri, 19 Sep 2025 12:04:09 +0000
Subject: [PATCH 5/7] [GlobalISel] Early exit if the first arg is fully unknown
in G_UMULH/G_SMULH
---
llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp | 2 ++
1 file changed, 2 insertions(+)
diff --git a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
index 1b8d192e53fcd..ca6c7c56d6a01 100644
--- a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
@@ -369,6 +369,7 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known,
case TargetOpcode::G_UMULH: {
computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
Depth + 1);
+ if (Known.isUnknown()) break;
computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
Depth + 1);
Known = KnownBits::mulhu(Known, Known2);
@@ -377,6 +378,7 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known,
case TargetOpcode::G_SMULH: {
computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
Depth + 1);
+ if (Known.isUnknown()) break;
computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
Depth + 1);
Known = KnownBits::mulhs(Known, Known2);
>From 003acadca22a89841abf202a5f2c4bb20454fa2c Mon Sep 17 00:00:00 2001
From: Pragyansh Chaturvedi <pragyanshchaturvedi18 at gmail.com>
Date: Fri, 19 Sep 2025 12:22:24 +0000
Subject: [PATCH 6/7] [GlobalISel] clang-format
---
llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
index ca6c7c56d6a01..dd02df817df03 100644
--- a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
@@ -369,7 +369,8 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known,
case TargetOpcode::G_UMULH: {
computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
Depth + 1);
- if (Known.isUnknown()) break;
+ if (Known.isUnknown())
+ break;
computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
Depth + 1);
Known = KnownBits::mulhu(Known, Known2);
@@ -378,7 +379,8 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known,
case TargetOpcode::G_SMULH: {
computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
Depth + 1);
- if (Known.isUnknown()) break;
+ if (Known.isUnknown())
+ break;
computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
Depth + 1);
Known = KnownBits::mulhs(Known, Known2);
>From 5263f16d26e90b5b492926b058c763892cab4c2d Mon Sep 17 00:00:00 2001
From: Pragyansh Chaturvedi <pragyanshchaturvedi18 at gmail.com>
Date: Sat, 20 Sep 2025 20:22:00 +0000
Subject: [PATCH 7/7] [GlobalISel] Update AMDGPU tests
---
.../CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll | 325 +++++++++--------
.../CodeGen/AMDGPU/GlobalISel/srem.i64.ll | 344 +++++++++---------
.../CodeGen/AMDGPU/GlobalISel/udiv.i64.ll | 108 +++---
.../CodeGen/AMDGPU/GlobalISel/urem.i64.ll | 32 +-
4 files changed, 421 insertions(+), 388 deletions(-)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
index c688db7c7bff3..f57fc005b994b 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
@@ -2534,195 +2534,202 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v4
; GISEL-NEXT: v_cvt_f32_u32_e32 v3, v1
-; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v9, 0
-; GISEL-NEXT: v_sub_i32_e32 v8, vcc, 0, v1
-; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v9
+; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v5, 0
+; GISEL-NEXT: v_sub_i32_e32 v10, vcc, 0, v1
+; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v5
; GISEL-NEXT: v_rcp_iflag_f32_e32 v3, v3
-; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, 0, vcc
+; GISEL-NEXT: v_subb_u32_e64 v11, s[4:5], 0, 0, vcc
; GISEL-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3
; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v3
-; GISEL-NEXT: v_trunc_f32_e32 v5, v4
-; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v5
-; GISEL-NEXT: v_cvt_u32_f32_e32 v7, v3
-; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v5
-; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v8, v7, 0
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v8, v11, v[4:5]
-; GISEL-NEXT: v_mul_hi_u32 v12, v7, v3
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v7, v[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3
-; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3
-; GISEL-NEXT: v_mul_lo_u32 v13, v7, v4
-; GISEL-NEXT: v_mul_lo_u32 v14, v11, v4
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13
+; GISEL-NEXT: v_trunc_f32_e32 v7, v4
+; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v7
+; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v3
+; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v7
+; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v10, v9, 0
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v10, v12, v[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v4, v12, v3
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v9, v[7:8]
+; GISEL-NEXT: v_mul_hi_u32 v8, v9, v3
+; GISEL-NEXT: v_mul_hi_u32 v3, v12, v3
+; GISEL-NEXT: v_mul_lo_u32 v13, v9, v7
+; GISEL-NEXT: v_mul_lo_u32 v14, v12, v7
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12
-; GISEL-NEXT: v_mul_hi_u32 v12, v7, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
+; GISEL-NEXT: v_mul_hi_u32 v8, v9, v7
+; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v13, v4
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v14, v3
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12
-; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v12, v5
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v3
-; GISEL-NEXT: v_addc_u32_e32 v11, vcc, v11, v4, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v8, v7, 0
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v8, v11, v[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v7, v[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v8
+; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8
+; GISEL-NEXT: v_mul_hi_u32 v7, v12, v7
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v7, v4
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v3
+; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v12, v4, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v10, v9, 0
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v10, v12, v[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v4, v12, v3
; GISEL-NEXT: v_and_b32_e32 v10, 0xffffff, v0
-; GISEL-NEXT: v_mul_lo_u32 v8, v7, v4
-; GISEL-NEXT: v_mul_hi_u32 v0, v7, v3
-; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v9, v[7:8]
+; GISEL-NEXT: v_mul_hi_u32 v0, v9, v3
+; GISEL-NEXT: v_mul_hi_u32 v3, v12, v3
+; GISEL-NEXT: v_mul_lo_u32 v8, v9, v7
+; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v2
+; 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 v0, vcc, v5, v0
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v5, v11, v4
+; GISEL-NEXT: v_mul_lo_u32 v4, v12, v7
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
-; GISEL-NEXT: v_mul_hi_u32 v8, v7, v4
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
+; GISEL-NEXT: v_mul_hi_u32 v8, v9, v7
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3
+; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v8
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
-; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
+; GISEL-NEXT: v_mul_hi_u32 v7, v12, v7
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v7, v0
-; GISEL-NEXT: v_addc_u32_e32 v3, vcc, v11, v3, vcc
-; GISEL-NEXT: v_mul_lo_u32 v4, 0, v0
-; GISEL-NEXT: v_mul_lo_u32 v5, v10, v3
-; GISEL-NEXT: v_mul_hi_u32 v7, v10, v0
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
+; GISEL-NEXT: v_addc_u32_e32 v4, vcc, v12, v3, vcc
+; GISEL-NEXT: v_mul_lo_u32 v7, 0, v0
+; GISEL-NEXT: v_mul_lo_u32 v8, v10, v4
+; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6
+; GISEL-NEXT: v_mul_hi_u32 v6, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0
-; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v2
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
-; GISEL-NEXT: v_mul_lo_u32 v5, 0, v3
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v7
-; GISEL-NEXT: v_mul_hi_u32 v7, v10, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v8
+; GISEL-NEXT: v_mul_lo_u32 v8, 0, v4
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v7, v6
+; GISEL-NEXT: v_mul_hi_u32 v7, v10, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v7
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v1, v0, 0
-; GISEL-NEXT: v_mul_hi_u32 v4, 0, v3
-; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6
-; GISEL-NEXT: v_mov_b32_e32 v5, v8
-; GISEL-NEXT: v_cvt_f32_u32_e32 v8, v3
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v4, v[5:6]
-; GISEL-NEXT: v_mac_f32_e32 v8, 0x4f800000, v9
-; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v8
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v0, v[5:6]
-; GISEL-NEXT: v_sub_i32_e32 v9, vcc, v10, v7
+; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
+; GISEL-NEXT: v_mul_hi_u32 v4, 0, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v1, v0, 0
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v9
+; GISEL-NEXT: v_cvt_f32_u32_e32 v9, v3
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v1, v4, v[7:8]
+; GISEL-NEXT: v_mac_f32_e32 v9, 0x4f800000, v5
+; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v9
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], 0, v0, v[7:8]
+; GISEL-NEXT: v_sub_i32_e32 v8, vcc, v10, v6
; GISEL-NEXT: v_mul_f32_e32 v2, 0x5f7ffffc, v2
-; GISEL-NEXT: v_mul_f32_e32 v6, 0x2f800000, v2
-; GISEL-NEXT: v_trunc_f32_e32 v8, v6
-; GISEL-NEXT: v_mac_f32_e32 v2, 0xcf800000, v8
+; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v2
+; GISEL-NEXT: v_subb_u32_e64 v9, s[4:5], 0, v7, vcc
+; GISEL-NEXT: v_sub_i32_e64 v10, s[4:5], 0, v7
+; GISEL-NEXT: v_trunc_f32_e32 v7, v5
+; GISEL-NEXT: v_mac_f32_e32 v2, 0xcf800000, v7
; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v2
-; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, v5, vcc
; GISEL-NEXT: v_sub_i32_e64 v13, s[4:5], 0, v3
; GISEL-NEXT: v_subb_u32_e64 v14, s[4:5], 0, 0, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v12, 0
-; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v8
-; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v5
-; GISEL-NEXT: v_mov_b32_e32 v2, v7
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v13, v15, v[2:3]
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v9, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v14, v12, v[7:8]
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v10
-; GISEL-NEXT: v_cndmask_b32_e64 v8, -1, v2, s[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v2, v15, v6
-; GISEL-NEXT: v_mul_lo_u32 v10, v12, v7
-; GISEL-NEXT: v_subbrev_u32_e32 v16, vcc, 0, v5, vcc
-; GISEL-NEXT: v_mul_hi_u32 v5, v12, v6
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v10
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v12, 0
+; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v7
+; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v8, v1
+; GISEL-NEXT: v_mov_b32_e32 v2, v6
+; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, -1, s[4:5]
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v15, v[2:3]
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v9
+; GISEL-NEXT: v_cndmask_b32_e64 v9, -1, v16, s[4:5]
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v14, v12, v[6:7]
+; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v10, vcc
+; GISEL-NEXT: v_mul_lo_u32 v7, v15, v5
+; GISEL-NEXT: v_mul_lo_u32 v10, v12, v6
+; GISEL-NEXT: v_sub_i32_e32 v8, vcc, v8, v1
+; GISEL-NEXT: v_subbrev_u32_e32 v16, vcc, 0, v2, vcc
+; GISEL-NEXT: v_mul_hi_u32 v2, v12, v5
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v5
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v7, v2
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v5, v15, v7
-; GISEL-NEXT: v_mul_hi_u32 v6, v15, v6
+; GISEL-NEXT: v_mul_lo_u32 v7, v15, v6
+; GISEL-NEXT: v_mul_hi_u32 v5, v15, v5
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v10, v2
-; GISEL-NEXT: v_mul_hi_u32 v10, v12, v7
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; GISEL-NEXT: v_mul_hi_u32 v10, v12, v6
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10
-; GISEL-NEXT: v_mul_hi_u32 v7, v15, v7
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10
+; GISEL-NEXT: v_mul_hi_u32 v6, v15, v6
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v12, v2
-; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v15, v5, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v10, 0
-; GISEL-NEXT: v_sub_i32_e32 v9, vcc, v9, v1
-; GISEL-NEXT: v_mov_b32_e32 v2, v6
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v12, v[2:3]
-; GISEL-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v16, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v14, v10, v[6:7]
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, 1, v0
-; GISEL-NEXT: v_addc_u32_e32 v13, vcc, 0, v4, vcc
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v9, v1
-; GISEL-NEXT: v_mul_lo_u32 v7, v12, v5
-; GISEL-NEXT: v_mul_lo_u32 v9, v10, v6
-; GISEL-NEXT: v_mul_hi_u32 v14, v10, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v15
-; GISEL-NEXT: v_cndmask_b32_e32 v1, -1, v1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v14
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v14, v12, v6
-; GISEL-NEXT: v_mul_hi_u32 v5, v12, v5
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7
-; GISEL-NEXT: v_mul_hi_u32 v9, v10, v6
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v14, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v14, 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 v9, vcc, v14, v9
-; GISEL-NEXT: v_mul_hi_u32 v6, v12, v6
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v7
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v10, v5
-; GISEL-NEXT: v_addc_u32_e32 v7, vcc, v12, v6, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v12, v2
+; GISEL-NEXT: v_addc_u32_e32 v10, vcc, v15, v5, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v7, 0
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, 1, v0
+; GISEL-NEXT: v_addc_u32_e32 v15, vcc, 0, v4, vcc
+; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v8, v1
+; GISEL-NEXT: v_mov_b32_e32 v1, v6
+; GISEL-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v13, v10, v[1:2]
+; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v16
+; GISEL-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v14, v7, v[1:2]
+; GISEL-NEXT: v_cndmask_b32_e32 v6, -1, v8, vcc
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, 1, v12
+; GISEL-NEXT: v_addc_u32_e32 v8, vcc, 0, v15, vcc
+; GISEL-NEXT: v_mul_lo_u32 v13, v10, v5
+; GISEL-NEXT: v_mul_lo_u32 v14, v7, v1
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
+; GISEL-NEXT: v_mul_hi_u32 v6, v7, v5
+; GISEL-NEXT: v_cndmask_b32_e32 v2, v12, v2, vcc
+; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v13, v14
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v12, v6
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, s[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v12, v10, v1
+; GISEL-NEXT: v_mul_hi_u32 v5, v10, v5
+; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v13, v6
+; GISEL-NEXT: v_mul_hi_u32 v13, v7, v1
+; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v12, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v5, v13
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v12, v13
+; GISEL-NEXT: v_mul_hi_u32 v1, v10, v1
+; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v5, v6
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v12, v6
+; GISEL-NEXT: v_add_i32_e64 v1, s[4:5], v1, v6
+; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5
+; GISEL-NEXT: v_addc_u32_e64 v1, s[4:5], v10, v1, s[4:5]
; GISEL-NEXT: v_mul_lo_u32 v6, 0, v5
-; GISEL-NEXT: v_mul_lo_u32 v9, v11, v7
-; GISEL-NEXT: v_mul_hi_u32 v14, v11, v5
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, 1, v2
-; GISEL-NEXT: v_addc_u32_e32 v12, vcc, 0, v13, vcc
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v9
-; GISEL-NEXT: v_mul_lo_u32 v9, 0, v7
+; GISEL-NEXT: v_mul_lo_u32 v7, v11, v1
+; GISEL-NEXT: v_mul_hi_u32 v10, v11, v5
+; GISEL-NEXT: v_cndmask_b32_e32 v8, v15, v8, vcc
; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v14
-; GISEL-NEXT: v_mul_hi_u32 v14, v11, v7
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v7
+; GISEL-NEXT: v_mul_lo_u32 v7, 0, v1
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10
+; GISEL-NEXT: v_mul_hi_u32 v10, v11, v1
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v9, v5
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v14
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v5, v6
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v9, 0
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
-; GISEL-NEXT: v_cndmask_b32_e32 v2, v2, v10, vcc
-; GISEL-NEXT: v_mul_hi_u32 v10, 0, v7
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v5, v6
+; GISEL-NEXT: v_mul_hi_u32 v1, 0, v1
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v10, 0
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v12
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v1, v7
; GISEL-NEXT: v_mov_b32_e32 v1, v6
-; GISEL-NEXT: v_cndmask_b32_e32 v12, v13, v12, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v8
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v3, v10, v[1:2]
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v3, v12, v[1:2]
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v12, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v9, v[6:7]
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v10, v[6:7]
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v8, vcc
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v11, v5
; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v6
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], 0, v6, vcc
@@ -2733,8 +2740,8 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
; GISEL-NEXT: v_subbrev_u32_e32 v5, vcc, 0, v5, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v4, -1, v6, s[4:5]
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v9
-; GISEL-NEXT: v_addc_u32_e32 v7, vcc, 0, v10, vcc
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v10
+; GISEL-NEXT: v_addc_u32_e32 v7, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v5
@@ -2745,8 +2752,8 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_cndmask_b32_e32 v2, v6, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
-; GISEL-NEXT: v_cndmask_b32_e32 v2, v9, v2, vcc
-; GISEL-NEXT: v_cndmask_b32_e32 v3, v10, v3, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v2, v10, v2, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v3, v12, v3, vcc
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_v2i64_24bit:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
index 682c5bccbe844..19dc20c510041 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
@@ -3029,193 +3029,203 @@ define <2 x i64> @v_srem_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v4
; GISEL-NEXT: v_cvt_f32_u32_e32 v3, v1
-; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v7, 0
-; GISEL-NEXT: v_sub_i32_e32 v9, vcc, 0, v1
-; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v7
+; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v4, 0
+; GISEL-NEXT: v_sub_i32_e32 v11, vcc, 0, v1
+; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v4
; GISEL-NEXT: v_rcp_iflag_f32_e32 v3, v3
-; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, 0, vcc
+; GISEL-NEXT: v_subb_u32_e64 v12, s[4:5], 0, 0, vcc
; GISEL-NEXT: v_and_b32_e32 v2, 0xffffff, v2
; GISEL-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3
-; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v3
-; GISEL-NEXT: v_trunc_f32_e32 v5, v4
+; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v3
+; GISEL-NEXT: v_trunc_f32_e32 v5, v5
; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v5
-; GISEL-NEXT: v_cvt_u32_f32_e32 v8, v3
-; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v5
-; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v9, v8, 0
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v9, v11, v[4:5]
-; GISEL-NEXT: v_mul_hi_u32 v12, v8, v3
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v8, v[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3
-; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3
-; GISEL-NEXT: v_mul_lo_u32 v13, v8, v4
-; GISEL-NEXT: v_mul_lo_u32 v14, v11, v4
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13
+; GISEL-NEXT: v_cvt_u32_f32_e32 v10, v3
+; GISEL-NEXT: v_cvt_u32_f32_e32 v5, v5
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v10, 0
+; GISEL-NEXT: v_mov_b32_e32 v3, v8
+; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v11, v5, v[3:4]
+; GISEL-NEXT: v_mul_lo_u32 v3, v5, v7
+; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v12, v10, v[8:9]
+; GISEL-NEXT: v_mul_hi_u32 v9, v10, v7
+; GISEL-NEXT: v_mul_hi_u32 v7, v5, v7
+; GISEL-NEXT: v_mul_lo_u32 v13, v10, v8
+; GISEL-NEXT: v_mul_lo_u32 v14, v5, v8
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v13
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12
-; GISEL-NEXT: v_mul_hi_u32 v12, v8, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v14, v3
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9
+; GISEL-NEXT: v_mul_hi_u32 v9, v10, v8
+; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v13, v3
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v14, v7
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12
-; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v12, v5
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v3
-; GISEL-NEXT: v_addc_u32_e32 v11, vcc, v11, v4, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v9, v8, 0
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v9, v11, v[4:5]
-; GISEL-NEXT: v_mul_hi_u32 v9, v8, v3
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v8, v[4:5]
-; GISEL-NEXT: v_and_b32_e32 v10, 0xffffff, v0
-; GISEL-NEXT: v_mul_lo_u32 v0, v11, v3
-; GISEL-NEXT: v_mul_lo_u32 v5, v8, v4
-; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v9
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v13, v9
+; GISEL-NEXT: v_mul_hi_u32 v8, v5, v8
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v8, v7
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v3
+; GISEL-NEXT: v_addc_u32_e32 v5, vcc, v5, v7, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v10, 0
+; GISEL-NEXT: v_mov_b32_e32 v3, v8
+; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v11, v5, v[3:4]
+; GISEL-NEXT: v_mul_lo_u32 v3, v5, v7
+; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v0
+; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v12, v10, v[8:9]
+; GISEL-NEXT: v_mul_hi_u32 v0, v10, v7
+; GISEL-NEXT: v_mul_hi_u32 v7, v5, v7
+; GISEL-NEXT: v_mul_lo_u32 v9, v10, v8
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v9, v11, v4
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
-; GISEL-NEXT: v_mul_hi_u32 v5, v8, v4
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v9, v3
+; GISEL-NEXT: v_mul_lo_u32 v3, v5, v8
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
+; GISEL-NEXT: v_mul_hi_u32 v9, v10, v8
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v7
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v9, v5
-; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
+; GISEL-NEXT: v_mul_hi_u32 v8, v5, v8
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
-; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v11, v3, vcc
-; GISEL-NEXT: v_mul_lo_u32 v4, 0, v0
-; GISEL-NEXT: v_mul_lo_u32 v5, v10, v8
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3
+; GISEL-NEXT: v_add_i32_e32 v3, vcc, v8, v3
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0
+; GISEL-NEXT: v_addc_u32_e32 v5, vcc, v5, v3, vcc
+; GISEL-NEXT: v_mul_lo_u32 v7, 0, v0
+; GISEL-NEXT: v_mul_lo_u32 v8, v11, v5
; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6
-; GISEL-NEXT: v_mul_hi_u32 v6, v10, v0
+; GISEL-NEXT: v_mul_hi_u32 v6, v11, v0
; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
-; GISEL-NEXT: v_mul_lo_u32 v5, 0, v8
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v8
+; GISEL-NEXT: v_mul_lo_u32 v8, 0, v5
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v7, v6
+; GISEL-NEXT: v_mul_hi_u32 v7, v11, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v7
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v0, v6
+; GISEL-NEXT: v_mul_hi_u32 v9, 0, v5
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v8, 0
+; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v7, v0
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v0
+; GISEL-NEXT: v_mov_b32_e32 v0, v6
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v1, v7, v[0:1]
+; GISEL-NEXT: v_cvt_f32_u32_e32 v0, v3
+; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v8, v[6:7]
+; GISEL-NEXT: v_mac_f32_e32 v0, 0x4f800000, v4
+; GISEL-NEXT: v_rcp_iflag_f32_e32 v0, v0
+; GISEL-NEXT: v_sub_i32_e32 v7, vcc, v11, v5
+; GISEL-NEXT: v_subb_u32_e64 v8, s[4:5], 0, v6, vcc
+; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v0
+; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v0
+; GISEL-NEXT: v_trunc_f32_e32 v9, v4
+; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v9
+; GISEL-NEXT: v_cvt_u32_f32_e32 v10, v0
+; GISEL-NEXT: v_sub_i32_e64 v11, s[4:5], 0, v3
+; GISEL-NEXT: v_subb_u32_e64 v12, s[4:5], 0, 0, s[4:5]
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v10, 0
+; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v9
+; GISEL-NEXT: v_sub_i32_e64 v13, s[4:5], 0, v6
+; GISEL-NEXT: v_mov_b32_e32 v0, v5
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v9, v[0:1]
+; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v7, v1
+; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5]
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v10, v[5:6]
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v8
+; GISEL-NEXT: v_cndmask_b32_e64 v14, -1, v0, s[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v0, v9, v4
+; GISEL-NEXT: v_mul_lo_u32 v6, v10, v5
+; GISEL-NEXT: v_mul_hi_u32 v15, v10, v4
+; GISEL-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v13, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v15
+; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
+; GISEL-NEXT: v_mul_lo_u32 v15, v9, v5
+; GISEL-NEXT: v_mul_hi_u32 v4, v9, v4
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0
+; GISEL-NEXT: v_mul_hi_u32 v6, v10, v5
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v15, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v6
-; GISEL-NEXT: v_mul_hi_u32 v6, v10, v8
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v15, v6
+; GISEL-NEXT: v_mul_hi_u32 v5, v9, v5
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v0, v4
-; GISEL-NEXT: v_cvt_f32_u32_e32 v0, v3
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v1, v9, 0
-; GISEL-NEXT: v_mul_hi_u32 v6, 0, v8
-; GISEL-NEXT: v_mac_f32_e32 v0, 0x4f800000, v7
-; GISEL-NEXT: v_rcp_iflag_f32_e32 v7, v0
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v6, v4
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v4
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v0
+; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v9, v4, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v10, 0
+; GISEL-NEXT: v_sub_i32_e32 v15, vcc, v7, v1
; GISEL-NEXT: v_mov_b32_e32 v0, v5
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v6, v[0:1]
-; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v7
-; GISEL-NEXT: v_mul_f32_e32 v7, 0x2f800000, v0
-; GISEL-NEXT: v_trunc_f32_e32 v11, v7
-; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v11
-; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v0
-; GISEL-NEXT: v_sub_i32_e32 v13, vcc, 0, v3
-; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v11
-; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v13, v12, 0
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v9, v[5:6]
-; GISEL-NEXT: v_mov_b32_e32 v0, v8
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v13, v11, v[0:1]
-; GISEL-NEXT: v_subb_u32_e64 v14, s[4:5], 0, 0, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v14, v12, v[8:9]
-; GISEL-NEXT: v_sub_i32_e32 v10, vcc, v10, v4
-; GISEL-NEXT: v_subb_u32_e64 v15, s[4:5], 0, v5, vcc
-; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], 0, v5
-; GISEL-NEXT: v_mul_lo_u32 v4, v11, v7
-; GISEL-NEXT: v_mul_lo_u32 v5, v12, v8
-; GISEL-NEXT: v_mul_hi_u32 v9, v12, v7
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v10, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v9
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v9, v[0:1]
+; GISEL-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v13, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v10, v[5:6]
+; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v15, v1
+; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, -1, vcc
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v13
+; GISEL-NEXT: v_cndmask_b32_e32 v0, -1, v16, vcc
+; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v15, v1
+; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v13, vcc
+; GISEL-NEXT: v_mul_lo_u32 v11, v9, v4
+; GISEL-NEXT: v_mul_lo_u32 v12, v10, v5
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0
+; GISEL-NEXT: v_mul_hi_u32 v0, v10, v4
+; GISEL-NEXT: v_mul_hi_u32 v4, v9, v4
+; 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 v0, s[4:5], v11, v0
+; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v11, v9, v5
+; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v12, v0
+; GISEL-NEXT: v_mul_hi_u32 v12, v10, v5
+; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v11, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v12
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v11, s[4:5], v11, v12
+; GISEL-NEXT: v_mul_hi_u32 v5, v9, v5
+; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v4, v0
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, s[4:5]
-; GISEL-NEXT: v_mul_lo_u32 v9, v11, v8
-; GISEL-NEXT: v_mul_hi_u32 v7, v11, v7
+; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v11, v4
; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v5, v4
-; GISEL-NEXT: v_mul_hi_u32 v5, v12, v8
-; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v9, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v9, v7
-; GISEL-NEXT: v_mul_hi_u32 v8, v11, v8
-; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v5, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v8, v5
-; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v12, v4
-; GISEL-NEXT: v_addc_u32_e64 v8, s[4:5], v11, v5, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v13, v7, 0
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v15
-; GISEL-NEXT: v_subbrev_u32_e32 v11, vcc, 0, v0, vcc
-; GISEL-NEXT: v_mov_b32_e32 v0, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v9, -1, v6, s[4:5]
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v8, v[0:1]
-; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v10, v1
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v14, v7, v[5:6]
-; GISEL-NEXT: v_mul_lo_u32 v12, v8, v4
-; GISEL-NEXT: v_subbrev_u32_e32 v11, vcc, 0, v11, vcc
-; GISEL-NEXT: v_mul_lo_u32 v13, v7, v5
-; GISEL-NEXT: v_mul_hi_u32 v14, v7, v4
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v11
-; GISEL-NEXT: v_cndmask_b32_e32 v6, -1, v6, vcc
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v13
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v14
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v14, v8, v5
-; GISEL-NEXT: v_mul_hi_u32 v4, v8, v4
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12
-; GISEL-NEXT: v_mul_hi_u32 v13, v7, v5
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v14, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v13, vcc, v14, v13
-; GISEL-NEXT: v_mul_hi_u32 v5, v8, v5
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v7, v4
-; GISEL-NEXT: v_addc_u32_e32 v7, vcc, v8, v5, vcc
-; GISEL-NEXT: v_mul_lo_u32 v5, 0, v4
-; GISEL-NEXT: v_mul_lo_u32 v8, v2, v7
-; GISEL-NEXT: v_mul_hi_u32 v13, v2, v4
-; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v0, v1
-; GISEL-NEXT: v_subbrev_u32_e32 v12, vcc, 0, v11, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
-; GISEL-NEXT: v_mul_lo_u32 v8, 0, v7
-; GISEL-NEXT: v_mul_hi_u32 v4, 0, v4
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13
-; GISEL-NEXT: v_mul_hi_u32 v13, v2, v7
+; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v10, v0
+; GISEL-NEXT: v_addc_u32_e64 v4, s[4:5], v9, v4, s[4:5]
+; GISEL-NEXT: v_mul_lo_u32 v5, 0, v0
+; GISEL-NEXT: v_mul_lo_u32 v9, v2, v4
+; GISEL-NEXT: v_cndmask_b32_e32 v10, v13, v6, vcc
+; GISEL-NEXT: v_mul_hi_u32 v6, v2, v0
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v9
+; GISEL-NEXT: v_mul_lo_u32 v9, 0, v4
+; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6
+; GISEL-NEXT: v_mul_hi_u32 v6, v2, v4
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v4, v5
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v8, 0
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
-; GISEL-NEXT: v_mul_hi_u32 v6, 0, v7
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v0, v1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v0, v5
+; GISEL-NEXT: v_mul_hi_u32 v11, 0, v4
+; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v9, 0
+; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v11, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v5
-; GISEL-NEXT: v_cndmask_b32_e32 v7, v11, v12, vcc
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v6, v[0:1]
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9
-; GISEL-NEXT: v_cndmask_b32_e32 v0, v10, v1, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v8, v[5:6]
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v7, vcc
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14
+; GISEL-NEXT: v_cndmask_b32_e32 v0, v7, v1, vcc
+; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v9, v[5:6]
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v8, v10, vcc
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v4
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], 0, v5, vcc
; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v5
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
index ded985ec3a1ec..4de10788a6bd7 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
@@ -2053,82 +2053,90 @@ define <2 x i64> @v_udiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_mul_lo_u32 v12, v6, v2
; GISEL-NEXT: v_mul_lo_u32 v13, 0, v2
; GISEL-NEXT: v_mul_hi_u32 v14, v6, v2
-; GISEL-NEXT: v_mul_hi_u32 v15, 0, v2
-; GISEL-NEXT: v_mul_lo_u32 v2, v0, v5
+; GISEL-NEXT: v_mul_hi_u32 v2, 0, v2
+; GISEL-NEXT: v_mul_lo_u32 v15, v0, v5
; GISEL-NEXT: v_mul_lo_u32 v16, 0, v5
; GISEL-NEXT: v_mul_hi_u32 v17, v0, v5
; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8
-; GISEL-NEXT: v_mul_lo_u32 v12, v3, v15
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v10, v2
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v15
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v16, v7
-; GISEL-NEXT: v_mul_lo_u32 v10, v1, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v14
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v11
-; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
+; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v17
+; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v7, v2
-; GISEL-NEXT: v_mul_lo_u32 v7, v3, v4
-; GISEL-NEXT: v_mul_lo_u32 v8, 0, v4
-; GISEL-NEXT: v_mul_hi_u32 v9, v3, v4
-; GISEL-NEXT: v_add_i32_e32 v11, vcc, 1, v4
-; GISEL-NEXT: v_addc_u32_e32 v13, vcc, 0, v15, vcc
-; GISEL-NEXT: v_mul_lo_u32 v14, v1, v2
-; GISEL-NEXT: v_mul_lo_u32 v16, 0, v2
-; GISEL-NEXT: v_mul_hi_u32 v17, v1, v2
-; GISEL-NEXT: v_add_i32_e32 v18, vcc, 1, v2
+; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8
+; GISEL-NEXT: v_mul_lo_u32 v9, v3, v4
+; GISEL-NEXT: v_mul_lo_u32 v12, 0, v4
+; GISEL-NEXT: v_mul_hi_u32 v13, v3, v4
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10
+; GISEL-NEXT: v_mul_lo_u32 v11, v1, v7
+; GISEL-NEXT: v_mul_lo_u32 v14, 0, v7
+; GISEL-NEXT: v_mul_hi_u32 v15, v1, v7
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v2, v8
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
+; GISEL-NEXT: v_mul_lo_u32 v2, v3, v8
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, 1, v4
+; GISEL-NEXT: v_addc_u32_e32 v16, vcc, 0, v8, vcc
+; GISEL-NEXT: v_mul_lo_u32 v17, v1, v5
+; GISEL-NEXT: v_add_i32_e32 v18, vcc, 1, v7
; GISEL-NEXT: v_addc_u32_e32 v19, vcc, 0, v5, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v12
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v16, v10
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, 1, v11
-; GISEL-NEXT: v_addc_u32_e32 v16, vcc, 0, v13, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v9
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v17
-; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v7
-; GISEL-NEXT: v_subb_u32_e64 v7, s[4:5], 0, v8, vcc
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v12, v2
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v14, v17
+; GISEL-NEXT: v_add_i32_e32 v14, vcc, 1, v10
+; GISEL-NEXT: v_addc_u32_e32 v17, vcc, 0, v16, vcc
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v13
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v15
+; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v9
+; GISEL-NEXT: v_subb_u32_e64 v9, s[4:5], 0, v2, vcc
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v6, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, -1, s[4:5]
-; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], v0, v14
-; GISEL-NEXT: v_subb_u32_e64 v14, s[6:7], 0, v9, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, -1, s[4:5]
+; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], v0, v11
+; GISEL-NEXT: v_subb_u32_e64 v11, s[6:7], 0, v12, s[4:5]
; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v0, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, -1, s[6:7]
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v7
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[8:9], 0, v14
-; GISEL-NEXT: v_add_i32_e64 v7, s[10:11], 1, v18
-; GISEL-NEXT: v_addc_u32_e64 v14, s[10:11], 0, v19, s[10:11]
-; GISEL-NEXT: v_sub_i32_e64 v8, s[10:11], 0, v8
-; GISEL-NEXT: v_sub_i32_e64 v9, s[10:11], 0, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v10, -1, v10, s[6:7]
-; GISEL-NEXT: v_subbrev_u32_e32 v8, vcc, 0, v8, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v17, -1, v17, s[8:9]
-; GISEL-NEXT: v_subbrev_u32_e64 v9, vcc, 0, v9, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, -1, s[6:7]
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v9
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[8:9], 0, v11
+; GISEL-NEXT: v_add_i32_e64 v9, s[10:11], 1, v18
+; GISEL-NEXT: v_addc_u32_e64 v11, s[10:11], 0, v19, s[10:11]
+; GISEL-NEXT: v_sub_i32_e64 v2, s[10:11], 0, v2
+; GISEL-NEXT: v_sub_i32_e64 v12, s[10:11], 0, v12
+; GISEL-NEXT: v_cndmask_b32_e64 v13, -1, v13, s[6:7]
+; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v15, -1, v15, s[8:9]
+; GISEL-NEXT: v_subbrev_u32_e64 v12, vcc, 0, v12, s[4:5]
; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v3
-; GISEL-NEXT: v_subbrev_u32_e32 v8, vcc, 0, v8, vcc
+; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v6, v3
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, -1, vcc
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
-; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v9, vcc
+; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v8
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v2
; GISEL-NEXT: v_cndmask_b32_e32 v1, -1, v3, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v6
; GISEL-NEXT: v_cndmask_b32_e32 v0, -1, v0, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v11, v12, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v10, v14, vcc
; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v0
-; GISEL-NEXT: v_cndmask_b32_e64 v3, v18, v7, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e32 v6, v13, v16, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v2, v18, v9, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e32 v3, v16, v17, vcc
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v13
; GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v4, v19, v14, s[4:5]
-; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v17
-; GISEL-NEXT: v_cndmask_b32_e64 v2, v2, v3, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v6, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v4, v19, v11, s[4:5]
+; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v15
+; GISEL-NEXT: v_cndmask_b32_e64 v2, v7, v2, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v8, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v3, v5, v4, s[4:5]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
index be5543b9b5b7f..a41ec8e7ce3ea 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
@@ -2058,34 +2058,42 @@ define <2 x i64> @v_urem_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8
-; GISEL-NEXT: v_mul_lo_u32 v2, v3, v2
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v15
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v16, v7
-; GISEL-NEXT: v_mul_lo_u32 v5, v1, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v14
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v11
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
+; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v17
+; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
-; GISEL-NEXT: v_mul_lo_u32 v8, v3, v4
-; GISEL-NEXT: v_mul_lo_u32 v9, 0, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8
+; GISEL-NEXT: v_mul_lo_u32 v9, v3, v4
+; GISEL-NEXT: v_mul_lo_u32 v12, 0, v4
; GISEL-NEXT: v_mul_hi_u32 v4, v3, v4
-; GISEL-NEXT: v_mul_lo_u32 v10, v1, v7
-; GISEL-NEXT: v_mul_lo_u32 v11, 0, v7
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10
+; GISEL-NEXT: v_mul_lo_u32 v11, v1, v7
+; GISEL-NEXT: v_mul_lo_u32 v13, 0, v7
; GISEL-NEXT: v_mul_hi_u32 v7, v1, v7
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v9, v2
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v11, v5
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v8
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
+; GISEL-NEXT: v_mul_lo_u32 v2, v3, v2
+; GISEL-NEXT: v_mul_lo_u32 v5, v1, v5
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v12, v2
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v4
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v7
-; GISEL-NEXT: v_sub_i32_e32 v5, vcc, v6, v8
+; GISEL-NEXT: v_sub_i32_e32 v5, vcc, v6, v9
; GISEL-NEXT: v_subb_u32_e64 v6, s[4:5], 0, v2, vcc
; GISEL-NEXT: v_sub_i32_e64 v2, s[4:5], 0, v2
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v5, v3
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
-; GISEL-NEXT: v_sub_i32_e64 v8, s[4:5], v0, v10
+; GISEL-NEXT: v_sub_i32_e64 v8, s[4:5], v0, v11
; GISEL-NEXT: v_subb_u32_e64 v9, s[6:7], 0, v4, s[4:5]
; GISEL-NEXT: v_sub_i32_e64 v0, s[6:7], 0, v4
; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v8, v1
More information about the llvm-commits
mailing list