[llvm] r354783 - [SLPVectorizer][X86] Add fixed smul/umul tests

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Mon Feb 25 05:26:30 PST 2019


Author: rksimon
Date: Mon Feb 25 05:26:30 2019
New Revision: 354783

URL: http://llvm.org/viewvc/llvm-project?rev=354783&view=rev
Log:
[SLPVectorizer][X86] Add fixed smul/umul tests

Baseline tests - fixed mul intrinsics aren't flagged as vectorizable yet

Added:
    llvm/trunk/test/Transforms/SLPVectorizer/X86/arith-fix.ll

Added: llvm/trunk/test/Transforms/SLPVectorizer/X86/arith-fix.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SLPVectorizer/X86/arith-fix.ll?rev=354783&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/SLPVectorizer/X86/arith-fix.ll (added)
+++ llvm/trunk/test/Transforms/SLPVectorizer/X86/arith-fix.ll Mon Feb 25 05:26:30 2019
@@ -0,0 +1,2007 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -mtriple=x86_64-unknown -basicaa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=slm -basicaa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,SLM
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basicaa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basicaa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=knl -basicaa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -mattr=-prefer-256-bit -basicaa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -mattr=+prefer-256-bit -basicaa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256BW
+
+ at a64 = common global [8 x i64] zeroinitializer, align 64
+ at b64 = common global [8 x i64] zeroinitializer, align 64
+ at c64 = common global [8 x i64] zeroinitializer, align 64
+ at a32 = common global [16 x i32] zeroinitializer, align 64
+ at b32 = common global [16 x i32] zeroinitializer, align 64
+ at c32 = common global [16 x i32] zeroinitializer, align 64
+ at a16 = common global [32 x i16] zeroinitializer, align 64
+ at b16 = common global [32 x i16] zeroinitializer, align 64
+ at c16 = common global [32 x i16] zeroinitializer, align 64
+ at a8  = common global [64 x i8] zeroinitializer, align 64
+ at b8  = common global [64 x i8] zeroinitializer, align 64
+ at c8  = common global [64 x i8] zeroinitializer, align 64
+
+declare i64 @llvm.smul.fix.i64(i64, i64, i32)
+declare i32 @llvm.smul.fix.i32(i32, i32, i32)
+declare i16 @llvm.smul.fix.i16(i16, i16, i32)
+declare i8  @llvm.smul.fix.i8 (i8 , i8 , i32)
+
+define void @smul_v8i64() {
+; CHECK-LABEL: @smul_v8i64(
+; CHECK-NEXT:    [[A0:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 0), align 8
+; CHECK-NEXT:    [[A1:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 1), align 8
+; CHECK-NEXT:    [[A2:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 2), align 8
+; CHECK-NEXT:    [[A3:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 3), align 8
+; CHECK-NEXT:    [[A4:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 4), align 8
+; CHECK-NEXT:    [[A5:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 5), align 8
+; CHECK-NEXT:    [[A6:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 6), align 8
+; CHECK-NEXT:    [[A7:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 7), align 8
+; CHECK-NEXT:    [[B0:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 0), align 8
+; CHECK-NEXT:    [[B1:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 1), align 8
+; CHECK-NEXT:    [[B2:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 2), align 8
+; CHECK-NEXT:    [[B3:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 3), align 8
+; CHECK-NEXT:    [[B4:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 4), align 8
+; CHECK-NEXT:    [[B5:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 5), align 8
+; CHECK-NEXT:    [[B6:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 6), align 8
+; CHECK-NEXT:    [[B7:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 7), align 8
+; CHECK-NEXT:    [[R0:%.*]] = call i64 @llvm.smul.fix.i64(i64 [[A0]], i64 [[B0]], i32 3)
+; CHECK-NEXT:    [[R1:%.*]] = call i64 @llvm.smul.fix.i64(i64 [[A1]], i64 [[B1]], i32 3)
+; CHECK-NEXT:    [[R2:%.*]] = call i64 @llvm.smul.fix.i64(i64 [[A2]], i64 [[B2]], i32 3)
+; CHECK-NEXT:    [[R3:%.*]] = call i64 @llvm.smul.fix.i64(i64 [[A3]], i64 [[B3]], i32 3)
+; CHECK-NEXT:    [[R4:%.*]] = call i64 @llvm.smul.fix.i64(i64 [[A4]], i64 [[B4]], i32 3)
+; CHECK-NEXT:    [[R5:%.*]] = call i64 @llvm.smul.fix.i64(i64 [[A5]], i64 [[B5]], i32 3)
+; CHECK-NEXT:    [[R6:%.*]] = call i64 @llvm.smul.fix.i64(i64 [[A6]], i64 [[B6]], i32 3)
+; CHECK-NEXT:    [[R7:%.*]] = call i64 @llvm.smul.fix.i64(i64 [[A7]], i64 [[B7]], i32 3)
+; CHECK-NEXT:    store i64 [[R0]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 0), align 8
+; CHECK-NEXT:    store i64 [[R1]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 1), align 8
+; CHECK-NEXT:    store i64 [[R2]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 2), align 8
+; CHECK-NEXT:    store i64 [[R3]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 3), align 8
+; CHECK-NEXT:    store i64 [[R4]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 4), align 8
+; CHECK-NEXT:    store i64 [[R5]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 5), align 8
+; CHECK-NEXT:    store i64 [[R6]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 6), align 8
+; CHECK-NEXT:    store i64 [[R7]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 7), align 8
+; CHECK-NEXT:    ret void
+;
+  %a0 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 0), align 8
+  %a1 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 1), align 8
+  %a2 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 2), align 8
+  %a3 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 3), align 8
+  %a4 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 4), align 8
+  %a5 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 5), align 8
+  %a6 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 6), align 8
+  %a7 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 7), align 8
+  %b0 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 0), align 8
+  %b1 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 1), align 8
+  %b2 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 2), align 8
+  %b3 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 3), align 8
+  %b4 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 4), align 8
+  %b5 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 5), align 8
+  %b6 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 6), align 8
+  %b7 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 7), align 8
+  %r0 = call i64 @llvm.smul.fix.i64(i64 %a0, i64 %b0, i32 3)
+  %r1 = call i64 @llvm.smul.fix.i64(i64 %a1, i64 %b1, i32 3)
+  %r2 = call i64 @llvm.smul.fix.i64(i64 %a2, i64 %b2, i32 3)
+  %r3 = call i64 @llvm.smul.fix.i64(i64 %a3, i64 %b3, i32 3)
+  %r4 = call i64 @llvm.smul.fix.i64(i64 %a4, i64 %b4, i32 3)
+  %r5 = call i64 @llvm.smul.fix.i64(i64 %a5, i64 %b5, i32 3)
+  %r6 = call i64 @llvm.smul.fix.i64(i64 %a6, i64 %b6, i32 3)
+  %r7 = call i64 @llvm.smul.fix.i64(i64 %a7, i64 %b7, i32 3)
+  store i64 %r0, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 0), align 8
+  store i64 %r1, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 1), align 8
+  store i64 %r2, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 2), align 8
+  store i64 %r3, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 3), align 8
+  store i64 %r4, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 4), align 8
+  store i64 %r5, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 5), align 8
+  store i64 %r6, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 6), align 8
+  store i64 %r7, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 7), align 8
+  ret void
+}
+
+define void @smul_v16i32() {
+; CHECK-LABEL: @smul_v16i32(
+; CHECK-NEXT:    [[A0:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 0), align 4
+; CHECK-NEXT:    [[A1:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 1), align 4
+; CHECK-NEXT:    [[A2:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 2), align 4
+; CHECK-NEXT:    [[A3:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 3), align 4
+; CHECK-NEXT:    [[A4:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 4), align 4
+; CHECK-NEXT:    [[A5:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 5), align 4
+; CHECK-NEXT:    [[A6:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 6), align 4
+; CHECK-NEXT:    [[A7:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 7), align 4
+; CHECK-NEXT:    [[A8:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 8), align 4
+; CHECK-NEXT:    [[A9:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 9), align 4
+; CHECK-NEXT:    [[A10:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 10), align 4
+; CHECK-NEXT:    [[A11:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 11), align 4
+; CHECK-NEXT:    [[A12:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 12), align 4
+; CHECK-NEXT:    [[A13:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 13), align 4
+; CHECK-NEXT:    [[A14:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 14), align 4
+; CHECK-NEXT:    [[A15:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 15), align 4
+; CHECK-NEXT:    [[B0:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 0), align 4
+; CHECK-NEXT:    [[B1:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 1), align 4
+; CHECK-NEXT:    [[B2:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 2), align 4
+; CHECK-NEXT:    [[B3:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 3), align 4
+; CHECK-NEXT:    [[B4:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 4), align 4
+; CHECK-NEXT:    [[B5:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 5), align 4
+; CHECK-NEXT:    [[B6:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 6), align 4
+; CHECK-NEXT:    [[B7:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 7), align 4
+; CHECK-NEXT:    [[B8:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 8), align 4
+; CHECK-NEXT:    [[B9:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 9), align 4
+; CHECK-NEXT:    [[B10:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 10), align 4
+; CHECK-NEXT:    [[B11:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 11), align 4
+; CHECK-NEXT:    [[B12:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 12), align 4
+; CHECK-NEXT:    [[B13:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 13), align 4
+; CHECK-NEXT:    [[B14:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 14), align 4
+; CHECK-NEXT:    [[B15:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 15), align 4
+; CHECK-NEXT:    [[R0:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A0]], i32 [[B0]], i32 3)
+; CHECK-NEXT:    [[R1:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A1]], i32 [[B1]], i32 3)
+; CHECK-NEXT:    [[R2:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A2]], i32 [[B2]], i32 3)
+; CHECK-NEXT:    [[R3:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A3]], i32 [[B3]], i32 3)
+; CHECK-NEXT:    [[R4:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A4]], i32 [[B4]], i32 3)
+; CHECK-NEXT:    [[R5:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A5]], i32 [[B5]], i32 3)
+; CHECK-NEXT:    [[R6:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A6]], i32 [[B6]], i32 3)
+; CHECK-NEXT:    [[R7:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A7]], i32 [[B7]], i32 3)
+; CHECK-NEXT:    [[R8:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A8]], i32 [[B8]], i32 3)
+; CHECK-NEXT:    [[R9:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A9]], i32 [[B9]], i32 3)
+; CHECK-NEXT:    [[R10:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A10]], i32 [[B10]], i32 3)
+; CHECK-NEXT:    [[R11:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A11]], i32 [[B11]], i32 3)
+; CHECK-NEXT:    [[R12:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A12]], i32 [[B12]], i32 3)
+; CHECK-NEXT:    [[R13:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A13]], i32 [[B13]], i32 3)
+; CHECK-NEXT:    [[R14:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A14]], i32 [[B14]], i32 3)
+; CHECK-NEXT:    [[R15:%.*]] = call i32 @llvm.smul.fix.i32(i32 [[A15]], i32 [[B15]], i32 3)
+; CHECK-NEXT:    store i32 [[R0]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 0), align 4
+; CHECK-NEXT:    store i32 [[R1]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 1), align 4
+; CHECK-NEXT:    store i32 [[R2]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 2), align 4
+; CHECK-NEXT:    store i32 [[R3]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 3), align 4
+; CHECK-NEXT:    store i32 [[R4]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 4), align 4
+; CHECK-NEXT:    store i32 [[R5]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 5), align 4
+; CHECK-NEXT:    store i32 [[R6]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 6), align 4
+; CHECK-NEXT:    store i32 [[R7]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 7), align 4
+; CHECK-NEXT:    store i32 [[R8]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 8), align 4
+; CHECK-NEXT:    store i32 [[R9]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 9), align 4
+; CHECK-NEXT:    store i32 [[R10]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 10), align 4
+; CHECK-NEXT:    store i32 [[R11]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 11), align 4
+; CHECK-NEXT:    store i32 [[R12]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 12), align 4
+; CHECK-NEXT:    store i32 [[R13]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 13), align 4
+; CHECK-NEXT:    store i32 [[R14]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 14), align 4
+; CHECK-NEXT:    store i32 [[R15]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 15), align 4
+; CHECK-NEXT:    ret void
+;
+  %a0  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 0 ), align 4
+  %a1  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 1 ), align 4
+  %a2  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 2 ), align 4
+  %a3  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 3 ), align 4
+  %a4  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 4 ), align 4
+  %a5  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 5 ), align 4
+  %a6  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 6 ), align 4
+  %a7  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 7 ), align 4
+  %a8  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 8 ), align 4
+  %a9  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 9 ), align 4
+  %a10 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 10), align 4
+  %a11 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 11), align 4
+  %a12 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 12), align 4
+  %a13 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 13), align 4
+  %a14 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 14), align 4
+  %a15 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 15), align 4
+  %b0  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 0 ), align 4
+  %b1  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 1 ), align 4
+  %b2  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 2 ), align 4
+  %b3  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 3 ), align 4
+  %b4  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 4 ), align 4
+  %b5  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 5 ), align 4
+  %b6  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 6 ), align 4
+  %b7  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 7 ), align 4
+  %b8  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 8 ), align 4
+  %b9  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 9 ), align 4
+  %b10 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 10), align 4
+  %b11 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 11), align 4
+  %b12 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 12), align 4
+  %b13 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 13), align 4
+  %b14 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 14), align 4
+  %b15 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 15), align 4
+  %r0  = call i32 @llvm.smul.fix.i32(i32 %a0 , i32 %b0 , i32 3)
+  %r1  = call i32 @llvm.smul.fix.i32(i32 %a1 , i32 %b1 , i32 3)
+  %r2  = call i32 @llvm.smul.fix.i32(i32 %a2 , i32 %b2 , i32 3)
+  %r3  = call i32 @llvm.smul.fix.i32(i32 %a3 , i32 %b3 , i32 3)
+  %r4  = call i32 @llvm.smul.fix.i32(i32 %a4 , i32 %b4 , i32 3)
+  %r5  = call i32 @llvm.smul.fix.i32(i32 %a5 , i32 %b5 , i32 3)
+  %r6  = call i32 @llvm.smul.fix.i32(i32 %a6 , i32 %b6 , i32 3)
+  %r7  = call i32 @llvm.smul.fix.i32(i32 %a7 , i32 %b7 , i32 3)
+  %r8  = call i32 @llvm.smul.fix.i32(i32 %a8 , i32 %b8 , i32 3)
+  %r9  = call i32 @llvm.smul.fix.i32(i32 %a9 , i32 %b9 , i32 3)
+  %r10 = call i32 @llvm.smul.fix.i32(i32 %a10, i32 %b10, i32 3)
+  %r11 = call i32 @llvm.smul.fix.i32(i32 %a11, i32 %b11, i32 3)
+  %r12 = call i32 @llvm.smul.fix.i32(i32 %a12, i32 %b12, i32 3)
+  %r13 = call i32 @llvm.smul.fix.i32(i32 %a13, i32 %b13, i32 3)
+  %r14 = call i32 @llvm.smul.fix.i32(i32 %a14, i32 %b14, i32 3)
+  %r15 = call i32 @llvm.smul.fix.i32(i32 %a15, i32 %b15, i32 3)
+  store i32 %r0 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 0 ), align 4
+  store i32 %r1 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 1 ), align 4
+  store i32 %r2 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 2 ), align 4
+  store i32 %r3 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 3 ), align 4
+  store i32 %r4 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 4 ), align 4
+  store i32 %r5 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 5 ), align 4
+  store i32 %r6 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 6 ), align 4
+  store i32 %r7 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 7 ), align 4
+  store i32 %r8 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 8 ), align 4
+  store i32 %r9 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 9 ), align 4
+  store i32 %r10, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 10), align 4
+  store i32 %r11, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 11), align 4
+  store i32 %r12, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 12), align 4
+  store i32 %r13, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 13), align 4
+  store i32 %r14, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 14), align 4
+  store i32 %r15, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 15), align 4
+  ret void
+}
+
+define void @smul_v32i16() {
+; CHECK-LABEL: @smul_v32i16(
+; CHECK-NEXT:    [[A0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 0), align 2
+; CHECK-NEXT:    [[A1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 1), align 2
+; CHECK-NEXT:    [[A2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 2), align 2
+; CHECK-NEXT:    [[A3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 3), align 2
+; CHECK-NEXT:    [[A4:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 4), align 2
+; CHECK-NEXT:    [[A5:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 5), align 2
+; CHECK-NEXT:    [[A6:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 6), align 2
+; CHECK-NEXT:    [[A7:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 7), align 2
+; CHECK-NEXT:    [[A8:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 8), align 2
+; CHECK-NEXT:    [[A9:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 9), align 2
+; CHECK-NEXT:    [[A10:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 10), align 2
+; CHECK-NEXT:    [[A11:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 11), align 2
+; CHECK-NEXT:    [[A12:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 12), align 2
+; CHECK-NEXT:    [[A13:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 13), align 2
+; CHECK-NEXT:    [[A14:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 14), align 2
+; CHECK-NEXT:    [[A15:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 15), align 2
+; CHECK-NEXT:    [[A16:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 16), align 2
+; CHECK-NEXT:    [[A17:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 17), align 2
+; CHECK-NEXT:    [[A18:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 18), align 2
+; CHECK-NEXT:    [[A19:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 19), align 2
+; CHECK-NEXT:    [[A20:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 20), align 2
+; CHECK-NEXT:    [[A21:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 21), align 2
+; CHECK-NEXT:    [[A22:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 22), align 2
+; CHECK-NEXT:    [[A23:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 23), align 2
+; CHECK-NEXT:    [[A24:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 24), align 2
+; CHECK-NEXT:    [[A25:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 25), align 2
+; CHECK-NEXT:    [[A26:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 26), align 2
+; CHECK-NEXT:    [[A27:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 27), align 2
+; CHECK-NEXT:    [[A28:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 28), align 2
+; CHECK-NEXT:    [[A29:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 29), align 2
+; CHECK-NEXT:    [[A30:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 30), align 2
+; CHECK-NEXT:    [[A31:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 31), align 2
+; CHECK-NEXT:    [[B0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 0), align 2
+; CHECK-NEXT:    [[B1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 1), align 2
+; CHECK-NEXT:    [[B2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 2), align 2
+; CHECK-NEXT:    [[B3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 3), align 2
+; CHECK-NEXT:    [[B4:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 4), align 2
+; CHECK-NEXT:    [[B5:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 5), align 2
+; CHECK-NEXT:    [[B6:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 6), align 2
+; CHECK-NEXT:    [[B7:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 7), align 2
+; CHECK-NEXT:    [[B8:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 8), align 2
+; CHECK-NEXT:    [[B9:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 9), align 2
+; CHECK-NEXT:    [[B10:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 10), align 2
+; CHECK-NEXT:    [[B11:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 11), align 2
+; CHECK-NEXT:    [[B12:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 12), align 2
+; CHECK-NEXT:    [[B13:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 13), align 2
+; CHECK-NEXT:    [[B14:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 14), align 2
+; CHECK-NEXT:    [[B15:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 15), align 2
+; CHECK-NEXT:    [[B16:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 16), align 2
+; CHECK-NEXT:    [[B17:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 17), align 2
+; CHECK-NEXT:    [[B18:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 18), align 2
+; CHECK-NEXT:    [[B19:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 19), align 2
+; CHECK-NEXT:    [[B20:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 20), align 2
+; CHECK-NEXT:    [[B21:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 21), align 2
+; CHECK-NEXT:    [[B22:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 22), align 2
+; CHECK-NEXT:    [[B23:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 23), align 2
+; CHECK-NEXT:    [[B24:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 24), align 2
+; CHECK-NEXT:    [[B25:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 25), align 2
+; CHECK-NEXT:    [[B26:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 26), align 2
+; CHECK-NEXT:    [[B27:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 27), align 2
+; CHECK-NEXT:    [[B28:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 28), align 2
+; CHECK-NEXT:    [[B29:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 29), align 2
+; CHECK-NEXT:    [[B30:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 30), align 2
+; CHECK-NEXT:    [[B31:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 31), align 2
+; CHECK-NEXT:    [[R0:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A0]], i16 [[B0]], i32 3)
+; CHECK-NEXT:    [[R1:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A1]], i16 [[B1]], i32 3)
+; CHECK-NEXT:    [[R2:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A2]], i16 [[B2]], i32 3)
+; CHECK-NEXT:    [[R3:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A3]], i16 [[B3]], i32 3)
+; CHECK-NEXT:    [[R4:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A4]], i16 [[B4]], i32 3)
+; CHECK-NEXT:    [[R5:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A5]], i16 [[B5]], i32 3)
+; CHECK-NEXT:    [[R6:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A6]], i16 [[B6]], i32 3)
+; CHECK-NEXT:    [[R7:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A7]], i16 [[B7]], i32 3)
+; CHECK-NEXT:    [[R8:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A8]], i16 [[B8]], i32 3)
+; CHECK-NEXT:    [[R9:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A9]], i16 [[B9]], i32 3)
+; CHECK-NEXT:    [[R10:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A10]], i16 [[B10]], i32 3)
+; CHECK-NEXT:    [[R11:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A11]], i16 [[B11]], i32 3)
+; CHECK-NEXT:    [[R12:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A12]], i16 [[B12]], i32 3)
+; CHECK-NEXT:    [[R13:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A13]], i16 [[B13]], i32 3)
+; CHECK-NEXT:    [[R14:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A14]], i16 [[B14]], i32 3)
+; CHECK-NEXT:    [[R15:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A15]], i16 [[B15]], i32 3)
+; CHECK-NEXT:    [[R16:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A16]], i16 [[B16]], i32 3)
+; CHECK-NEXT:    [[R17:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A17]], i16 [[B17]], i32 3)
+; CHECK-NEXT:    [[R18:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A18]], i16 [[B18]], i32 3)
+; CHECK-NEXT:    [[R19:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A19]], i16 [[B19]], i32 3)
+; CHECK-NEXT:    [[R20:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A20]], i16 [[B20]], i32 3)
+; CHECK-NEXT:    [[R21:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A21]], i16 [[B21]], i32 3)
+; CHECK-NEXT:    [[R22:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A22]], i16 [[B22]], i32 3)
+; CHECK-NEXT:    [[R23:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A23]], i16 [[B23]], i32 3)
+; CHECK-NEXT:    [[R24:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A24]], i16 [[B24]], i32 3)
+; CHECK-NEXT:    [[R25:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A25]], i16 [[B25]], i32 3)
+; CHECK-NEXT:    [[R26:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A26]], i16 [[B26]], i32 3)
+; CHECK-NEXT:    [[R27:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A27]], i16 [[B27]], i32 3)
+; CHECK-NEXT:    [[R28:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A28]], i16 [[B28]], i32 3)
+; CHECK-NEXT:    [[R29:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A29]], i16 [[B29]], i32 3)
+; CHECK-NEXT:    [[R30:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A30]], i16 [[B30]], i32 3)
+; CHECK-NEXT:    [[R31:%.*]] = call i16 @llvm.smul.fix.i16(i16 [[A31]], i16 [[B31]], i32 3)
+; CHECK-NEXT:    store i16 [[R0]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 0), align 2
+; CHECK-NEXT:    store i16 [[R1]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 1), align 2
+; CHECK-NEXT:    store i16 [[R2]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 2), align 2
+; CHECK-NEXT:    store i16 [[R3]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 3), align 2
+; CHECK-NEXT:    store i16 [[R4]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 4), align 2
+; CHECK-NEXT:    store i16 [[R5]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 5), align 2
+; CHECK-NEXT:    store i16 [[R6]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 6), align 2
+; CHECK-NEXT:    store i16 [[R7]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 7), align 2
+; CHECK-NEXT:    store i16 [[R8]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 8), align 2
+; CHECK-NEXT:    store i16 [[R9]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 9), align 2
+; CHECK-NEXT:    store i16 [[R10]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 10), align 2
+; CHECK-NEXT:    store i16 [[R11]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 11), align 2
+; CHECK-NEXT:    store i16 [[R12]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 12), align 2
+; CHECK-NEXT:    store i16 [[R13]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 13), align 2
+; CHECK-NEXT:    store i16 [[R14]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 14), align 2
+; CHECK-NEXT:    store i16 [[R15]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 15), align 2
+; CHECK-NEXT:    store i16 [[R16]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 16), align 2
+; CHECK-NEXT:    store i16 [[R17]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 17), align 2
+; CHECK-NEXT:    store i16 [[R18]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 18), align 2
+; CHECK-NEXT:    store i16 [[R19]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 19), align 2
+; CHECK-NEXT:    store i16 [[R20]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 20), align 2
+; CHECK-NEXT:    store i16 [[R21]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 21), align 2
+; CHECK-NEXT:    store i16 [[R22]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 22), align 2
+; CHECK-NEXT:    store i16 [[R23]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 23), align 2
+; CHECK-NEXT:    store i16 [[R24]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 24), align 2
+; CHECK-NEXT:    store i16 [[R25]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 25), align 2
+; CHECK-NEXT:    store i16 [[R26]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 26), align 2
+; CHECK-NEXT:    store i16 [[R27]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 27), align 2
+; CHECK-NEXT:    store i16 [[R28]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 28), align 2
+; CHECK-NEXT:    store i16 [[R29]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 29), align 2
+; CHECK-NEXT:    store i16 [[R30]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 30), align 2
+; CHECK-NEXT:    store i16 [[R31]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 31), align 2
+; CHECK-NEXT:    ret void
+;
+  %a0  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 0 ), align 2
+  %a1  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 1 ), align 2
+  %a2  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 2 ), align 2
+  %a3  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 3 ), align 2
+  %a4  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 4 ), align 2
+  %a5  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 5 ), align 2
+  %a6  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 6 ), align 2
+  %a7  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 7 ), align 2
+  %a8  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 8 ), align 2
+  %a9  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 9 ), align 2
+  %a10 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 10), align 2
+  %a11 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 11), align 2
+  %a12 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 12), align 2
+  %a13 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 13), align 2
+  %a14 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 14), align 2
+  %a15 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 15), align 2
+  %a16 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 16), align 2
+  %a17 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 17), align 2
+  %a18 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 18), align 2
+  %a19 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 19), align 2
+  %a20 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 20), align 2
+  %a21 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 21), align 2
+  %a22 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 22), align 2
+  %a23 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 23), align 2
+  %a24 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 24), align 2
+  %a25 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 25), align 2
+  %a26 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 26), align 2
+  %a27 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 27), align 2
+  %a28 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 28), align 2
+  %a29 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 29), align 2
+  %a30 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 30), align 2
+  %a31 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 31), align 2
+  %b0  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 0 ), align 2
+  %b1  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 1 ), align 2
+  %b2  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 2 ), align 2
+  %b3  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 3 ), align 2
+  %b4  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 4 ), align 2
+  %b5  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 5 ), align 2
+  %b6  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 6 ), align 2
+  %b7  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 7 ), align 2
+  %b8  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 8 ), align 2
+  %b9  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 9 ), align 2
+  %b10 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 10), align 2
+  %b11 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 11), align 2
+  %b12 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 12), align 2
+  %b13 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 13), align 2
+  %b14 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 14), align 2
+  %b15 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 15), align 2
+  %b16 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 16), align 2
+  %b17 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 17), align 2
+  %b18 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 18), align 2
+  %b19 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 19), align 2
+  %b20 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 20), align 2
+  %b21 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 21), align 2
+  %b22 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 22), align 2
+  %b23 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 23), align 2
+  %b24 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 24), align 2
+  %b25 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 25), align 2
+  %b26 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 26), align 2
+  %b27 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 27), align 2
+  %b28 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 28), align 2
+  %b29 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 29), align 2
+  %b30 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 30), align 2
+  %b31 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 31), align 2
+  %r0  = call i16 @llvm.smul.fix.i16(i16 %a0 , i16 %b0 , i32 3)
+  %r1  = call i16 @llvm.smul.fix.i16(i16 %a1 , i16 %b1 , i32 3)
+  %r2  = call i16 @llvm.smul.fix.i16(i16 %a2 , i16 %b2 , i32 3)
+  %r3  = call i16 @llvm.smul.fix.i16(i16 %a3 , i16 %b3 , i32 3)
+  %r4  = call i16 @llvm.smul.fix.i16(i16 %a4 , i16 %b4 , i32 3)
+  %r5  = call i16 @llvm.smul.fix.i16(i16 %a5 , i16 %b5 , i32 3)
+  %r6  = call i16 @llvm.smul.fix.i16(i16 %a6 , i16 %b6 , i32 3)
+  %r7  = call i16 @llvm.smul.fix.i16(i16 %a7 , i16 %b7 , i32 3)
+  %r8  = call i16 @llvm.smul.fix.i16(i16 %a8 , i16 %b8 , i32 3)
+  %r9  = call i16 @llvm.smul.fix.i16(i16 %a9 , i16 %b9 , i32 3)
+  %r10 = call i16 @llvm.smul.fix.i16(i16 %a10, i16 %b10, i32 3)
+  %r11 = call i16 @llvm.smul.fix.i16(i16 %a11, i16 %b11, i32 3)
+  %r12 = call i16 @llvm.smul.fix.i16(i16 %a12, i16 %b12, i32 3)
+  %r13 = call i16 @llvm.smul.fix.i16(i16 %a13, i16 %b13, i32 3)
+  %r14 = call i16 @llvm.smul.fix.i16(i16 %a14, i16 %b14, i32 3)
+  %r15 = call i16 @llvm.smul.fix.i16(i16 %a15, i16 %b15, i32 3)
+  %r16 = call i16 @llvm.smul.fix.i16(i16 %a16, i16 %b16, i32 3)
+  %r17 = call i16 @llvm.smul.fix.i16(i16 %a17, i16 %b17, i32 3)
+  %r18 = call i16 @llvm.smul.fix.i16(i16 %a18, i16 %b18, i32 3)
+  %r19 = call i16 @llvm.smul.fix.i16(i16 %a19, i16 %b19, i32 3)
+  %r20 = call i16 @llvm.smul.fix.i16(i16 %a20, i16 %b20, i32 3)
+  %r21 = call i16 @llvm.smul.fix.i16(i16 %a21, i16 %b21, i32 3)
+  %r22 = call i16 @llvm.smul.fix.i16(i16 %a22, i16 %b22, i32 3)
+  %r23 = call i16 @llvm.smul.fix.i16(i16 %a23, i16 %b23, i32 3)
+  %r24 = call i16 @llvm.smul.fix.i16(i16 %a24, i16 %b24, i32 3)
+  %r25 = call i16 @llvm.smul.fix.i16(i16 %a25, i16 %b25, i32 3)
+  %r26 = call i16 @llvm.smul.fix.i16(i16 %a26, i16 %b26, i32 3)
+  %r27 = call i16 @llvm.smul.fix.i16(i16 %a27, i16 %b27, i32 3)
+  %r28 = call i16 @llvm.smul.fix.i16(i16 %a28, i16 %b28, i32 3)
+  %r29 = call i16 @llvm.smul.fix.i16(i16 %a29, i16 %b29, i32 3)
+  %r30 = call i16 @llvm.smul.fix.i16(i16 %a30, i16 %b30, i32 3)
+  %r31 = call i16 @llvm.smul.fix.i16(i16 %a31, i16 %b31, i32 3)
+  store i16 %r0 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 0 ), align 2
+  store i16 %r1 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 1 ), align 2
+  store i16 %r2 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 2 ), align 2
+  store i16 %r3 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 3 ), align 2
+  store i16 %r4 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 4 ), align 2
+  store i16 %r5 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 5 ), align 2
+  store i16 %r6 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 6 ), align 2
+  store i16 %r7 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 7 ), align 2
+  store i16 %r8 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 8 ), align 2
+  store i16 %r9 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 9 ), align 2
+  store i16 %r10, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 10), align 2
+  store i16 %r11, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 11), align 2
+  store i16 %r12, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 12), align 2
+  store i16 %r13, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 13), align 2
+  store i16 %r14, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 14), align 2
+  store i16 %r15, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 15), align 2
+  store i16 %r16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 16), align 2
+  store i16 %r17, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 17), align 2
+  store i16 %r18, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 18), align 2
+  store i16 %r19, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 19), align 2
+  store i16 %r20, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 20), align 2
+  store i16 %r21, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 21), align 2
+  store i16 %r22, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 22), align 2
+  store i16 %r23, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 23), align 2
+  store i16 %r24, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 24), align 2
+  store i16 %r25, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 25), align 2
+  store i16 %r26, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 26), align 2
+  store i16 %r27, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 27), align 2
+  store i16 %r28, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 28), align 2
+  store i16 %r29, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 29), align 2
+  store i16 %r30, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 30), align 2
+  store i16 %r31, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 31), align 2
+  ret void
+}
+
+define void @smul_v64i8() {
+; CHECK-LABEL: @smul_v64i8(
+; CHECK-NEXT:    [[A0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 0), align 1
+; CHECK-NEXT:    [[A1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 1), align 1
+; CHECK-NEXT:    [[A2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 2), align 1
+; CHECK-NEXT:    [[A3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 3), align 1
+; CHECK-NEXT:    [[A4:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 4), align 1
+; CHECK-NEXT:    [[A5:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 5), align 1
+; CHECK-NEXT:    [[A6:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 6), align 1
+; CHECK-NEXT:    [[A7:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 7), align 1
+; CHECK-NEXT:    [[A8:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 8), align 1
+; CHECK-NEXT:    [[A9:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 9), align 1
+; CHECK-NEXT:    [[A10:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 10), align 1
+; CHECK-NEXT:    [[A11:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 11), align 1
+; CHECK-NEXT:    [[A12:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 12), align 1
+; CHECK-NEXT:    [[A13:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 13), align 1
+; CHECK-NEXT:    [[A14:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 14), align 1
+; CHECK-NEXT:    [[A15:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 15), align 1
+; CHECK-NEXT:    [[A16:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 16), align 1
+; CHECK-NEXT:    [[A17:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 17), align 1
+; CHECK-NEXT:    [[A18:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 18), align 1
+; CHECK-NEXT:    [[A19:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 19), align 1
+; CHECK-NEXT:    [[A20:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 20), align 1
+; CHECK-NEXT:    [[A21:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 21), align 1
+; CHECK-NEXT:    [[A22:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 22), align 1
+; CHECK-NEXT:    [[A23:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 23), align 1
+; CHECK-NEXT:    [[A24:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 24), align 1
+; CHECK-NEXT:    [[A25:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 25), align 1
+; CHECK-NEXT:    [[A26:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 26), align 1
+; CHECK-NEXT:    [[A27:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 27), align 1
+; CHECK-NEXT:    [[A28:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 28), align 1
+; CHECK-NEXT:    [[A29:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 29), align 1
+; CHECK-NEXT:    [[A30:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 30), align 1
+; CHECK-NEXT:    [[A31:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 31), align 1
+; CHECK-NEXT:    [[A32:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 32), align 1
+; CHECK-NEXT:    [[A33:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 33), align 1
+; CHECK-NEXT:    [[A34:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 34), align 1
+; CHECK-NEXT:    [[A35:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 35), align 1
+; CHECK-NEXT:    [[A36:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 36), align 1
+; CHECK-NEXT:    [[A37:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 37), align 1
+; CHECK-NEXT:    [[A38:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 38), align 1
+; CHECK-NEXT:    [[A39:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 39), align 1
+; CHECK-NEXT:    [[A40:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 40), align 1
+; CHECK-NEXT:    [[A41:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 41), align 1
+; CHECK-NEXT:    [[A42:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 42), align 1
+; CHECK-NEXT:    [[A43:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 43), align 1
+; CHECK-NEXT:    [[A44:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 44), align 1
+; CHECK-NEXT:    [[A45:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 45), align 1
+; CHECK-NEXT:    [[A46:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 46), align 1
+; CHECK-NEXT:    [[A47:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 47), align 1
+; CHECK-NEXT:    [[A48:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 48), align 1
+; CHECK-NEXT:    [[A49:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 49), align 1
+; CHECK-NEXT:    [[A50:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 50), align 1
+; CHECK-NEXT:    [[A51:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 51), align 1
+; CHECK-NEXT:    [[A52:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 52), align 1
+; CHECK-NEXT:    [[A53:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 53), align 1
+; CHECK-NEXT:    [[A54:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 54), align 1
+; CHECK-NEXT:    [[A55:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 55), align 1
+; CHECK-NEXT:    [[A56:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 56), align 1
+; CHECK-NEXT:    [[A57:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 57), align 1
+; CHECK-NEXT:    [[A58:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 58), align 1
+; CHECK-NEXT:    [[A59:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 59), align 1
+; CHECK-NEXT:    [[A60:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 60), align 1
+; CHECK-NEXT:    [[A61:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 61), align 1
+; CHECK-NEXT:    [[A62:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 62), align 1
+; CHECK-NEXT:    [[A63:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 63), align 1
+; CHECK-NEXT:    [[B0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 0), align 1
+; CHECK-NEXT:    [[B1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 1), align 1
+; CHECK-NEXT:    [[B2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 2), align 1
+; CHECK-NEXT:    [[B3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 3), align 1
+; CHECK-NEXT:    [[B4:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 4), align 1
+; CHECK-NEXT:    [[B5:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 5), align 1
+; CHECK-NEXT:    [[B6:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 6), align 1
+; CHECK-NEXT:    [[B7:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 7), align 1
+; CHECK-NEXT:    [[B8:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 8), align 1
+; CHECK-NEXT:    [[B9:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 9), align 1
+; CHECK-NEXT:    [[B10:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 10), align 1
+; CHECK-NEXT:    [[B11:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 11), align 1
+; CHECK-NEXT:    [[B12:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 12), align 1
+; CHECK-NEXT:    [[B13:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 13), align 1
+; CHECK-NEXT:    [[B14:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 14), align 1
+; CHECK-NEXT:    [[B15:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 15), align 1
+; CHECK-NEXT:    [[B16:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 16), align 1
+; CHECK-NEXT:    [[B17:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 17), align 1
+; CHECK-NEXT:    [[B18:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 18), align 1
+; CHECK-NEXT:    [[B19:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 19), align 1
+; CHECK-NEXT:    [[B20:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 20), align 1
+; CHECK-NEXT:    [[B21:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 21), align 1
+; CHECK-NEXT:    [[B22:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 22), align 1
+; CHECK-NEXT:    [[B23:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 23), align 1
+; CHECK-NEXT:    [[B24:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 24), align 1
+; CHECK-NEXT:    [[B25:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 25), align 1
+; CHECK-NEXT:    [[B26:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 26), align 1
+; CHECK-NEXT:    [[B27:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 27), align 1
+; CHECK-NEXT:    [[B28:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 28), align 1
+; CHECK-NEXT:    [[B29:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 29), align 1
+; CHECK-NEXT:    [[B30:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 30), align 1
+; CHECK-NEXT:    [[B31:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 31), align 1
+; CHECK-NEXT:    [[B32:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 32), align 1
+; CHECK-NEXT:    [[B33:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 33), align 1
+; CHECK-NEXT:    [[B34:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 34), align 1
+; CHECK-NEXT:    [[B35:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 35), align 1
+; CHECK-NEXT:    [[B36:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 36), align 1
+; CHECK-NEXT:    [[B37:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 37), align 1
+; CHECK-NEXT:    [[B38:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 38), align 1
+; CHECK-NEXT:    [[B39:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 39), align 1
+; CHECK-NEXT:    [[B40:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 40), align 1
+; CHECK-NEXT:    [[B41:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 41), align 1
+; CHECK-NEXT:    [[B42:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 42), align 1
+; CHECK-NEXT:    [[B43:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 43), align 1
+; CHECK-NEXT:    [[B44:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 44), align 1
+; CHECK-NEXT:    [[B45:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 45), align 1
+; CHECK-NEXT:    [[B46:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 46), align 1
+; CHECK-NEXT:    [[B47:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 47), align 1
+; CHECK-NEXT:    [[B48:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 48), align 1
+; CHECK-NEXT:    [[B49:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 49), align 1
+; CHECK-NEXT:    [[B50:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 50), align 1
+; CHECK-NEXT:    [[B51:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 51), align 1
+; CHECK-NEXT:    [[B52:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 52), align 1
+; CHECK-NEXT:    [[B53:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 53), align 1
+; CHECK-NEXT:    [[B54:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 54), align 1
+; CHECK-NEXT:    [[B55:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 55), align 1
+; CHECK-NEXT:    [[B56:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 56), align 1
+; CHECK-NEXT:    [[B57:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 57), align 1
+; CHECK-NEXT:    [[B58:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 58), align 1
+; CHECK-NEXT:    [[B59:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 59), align 1
+; CHECK-NEXT:    [[B60:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 60), align 1
+; CHECK-NEXT:    [[B61:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 61), align 1
+; CHECK-NEXT:    [[B62:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 62), align 1
+; CHECK-NEXT:    [[B63:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 63), align 1
+; CHECK-NEXT:    [[R0:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A0]], i8 [[B0]], i32 3)
+; CHECK-NEXT:    [[R1:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A1]], i8 [[B1]], i32 3)
+; CHECK-NEXT:    [[R2:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A2]], i8 [[B2]], i32 3)
+; CHECK-NEXT:    [[R3:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A3]], i8 [[B3]], i32 3)
+; CHECK-NEXT:    [[R4:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A4]], i8 [[B4]], i32 3)
+; CHECK-NEXT:    [[R5:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A5]], i8 [[B5]], i32 3)
+; CHECK-NEXT:    [[R6:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A6]], i8 [[B6]], i32 3)
+; CHECK-NEXT:    [[R7:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A7]], i8 [[B7]], i32 3)
+; CHECK-NEXT:    [[R8:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A8]], i8 [[B8]], i32 3)
+; CHECK-NEXT:    [[R9:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A9]], i8 [[B9]], i32 3)
+; CHECK-NEXT:    [[R10:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A10]], i8 [[B10]], i32 3)
+; CHECK-NEXT:    [[R11:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A11]], i8 [[B11]], i32 3)
+; CHECK-NEXT:    [[R12:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A12]], i8 [[B12]], i32 3)
+; CHECK-NEXT:    [[R13:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A13]], i8 [[B13]], i32 3)
+; CHECK-NEXT:    [[R14:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A14]], i8 [[B14]], i32 3)
+; CHECK-NEXT:    [[R15:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A15]], i8 [[B15]], i32 3)
+; CHECK-NEXT:    [[R16:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A16]], i8 [[B16]], i32 3)
+; CHECK-NEXT:    [[R17:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A17]], i8 [[B17]], i32 3)
+; CHECK-NEXT:    [[R18:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A18]], i8 [[B18]], i32 3)
+; CHECK-NEXT:    [[R19:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A19]], i8 [[B19]], i32 3)
+; CHECK-NEXT:    [[R20:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A20]], i8 [[B20]], i32 3)
+; CHECK-NEXT:    [[R21:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A21]], i8 [[B21]], i32 3)
+; CHECK-NEXT:    [[R22:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A22]], i8 [[B22]], i32 3)
+; CHECK-NEXT:    [[R23:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A23]], i8 [[B23]], i32 3)
+; CHECK-NEXT:    [[R24:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A24]], i8 [[B24]], i32 3)
+; CHECK-NEXT:    [[R25:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A25]], i8 [[B25]], i32 3)
+; CHECK-NEXT:    [[R26:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A26]], i8 [[B26]], i32 3)
+; CHECK-NEXT:    [[R27:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A27]], i8 [[B27]], i32 3)
+; CHECK-NEXT:    [[R28:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A28]], i8 [[B28]], i32 3)
+; CHECK-NEXT:    [[R29:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A29]], i8 [[B29]], i32 3)
+; CHECK-NEXT:    [[R30:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A30]], i8 [[B30]], i32 3)
+; CHECK-NEXT:    [[R31:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A31]], i8 [[B31]], i32 3)
+; CHECK-NEXT:    [[R32:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A32]], i8 [[B32]], i32 3)
+; CHECK-NEXT:    [[R33:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A33]], i8 [[B33]], i32 3)
+; CHECK-NEXT:    [[R34:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A34]], i8 [[B34]], i32 3)
+; CHECK-NEXT:    [[R35:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A35]], i8 [[B35]], i32 3)
+; CHECK-NEXT:    [[R36:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A36]], i8 [[B36]], i32 3)
+; CHECK-NEXT:    [[R37:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A37]], i8 [[B37]], i32 3)
+; CHECK-NEXT:    [[R38:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A38]], i8 [[B38]], i32 3)
+; CHECK-NEXT:    [[R39:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A39]], i8 [[B39]], i32 3)
+; CHECK-NEXT:    [[R40:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A40]], i8 [[B40]], i32 3)
+; CHECK-NEXT:    [[R41:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A41]], i8 [[B41]], i32 3)
+; CHECK-NEXT:    [[R42:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A42]], i8 [[B42]], i32 3)
+; CHECK-NEXT:    [[R43:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A43]], i8 [[B43]], i32 3)
+; CHECK-NEXT:    [[R44:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A44]], i8 [[B44]], i32 3)
+; CHECK-NEXT:    [[R45:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A45]], i8 [[B45]], i32 3)
+; CHECK-NEXT:    [[R46:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A46]], i8 [[B46]], i32 3)
+; CHECK-NEXT:    [[R47:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A47]], i8 [[B47]], i32 3)
+; CHECK-NEXT:    [[R48:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A48]], i8 [[B48]], i32 3)
+; CHECK-NEXT:    [[R49:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A49]], i8 [[B49]], i32 3)
+; CHECK-NEXT:    [[R50:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A50]], i8 [[B50]], i32 3)
+; CHECK-NEXT:    [[R51:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A51]], i8 [[B51]], i32 3)
+; CHECK-NEXT:    [[R52:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A52]], i8 [[B52]], i32 3)
+; CHECK-NEXT:    [[R53:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A53]], i8 [[B53]], i32 3)
+; CHECK-NEXT:    [[R54:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A54]], i8 [[B54]], i32 3)
+; CHECK-NEXT:    [[R55:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A55]], i8 [[B55]], i32 3)
+; CHECK-NEXT:    [[R56:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A56]], i8 [[B56]], i32 3)
+; CHECK-NEXT:    [[R57:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A57]], i8 [[B57]], i32 3)
+; CHECK-NEXT:    [[R58:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A58]], i8 [[B58]], i32 3)
+; CHECK-NEXT:    [[R59:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A59]], i8 [[B59]], i32 3)
+; CHECK-NEXT:    [[R60:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A60]], i8 [[B60]], i32 3)
+; CHECK-NEXT:    [[R61:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A61]], i8 [[B61]], i32 3)
+; CHECK-NEXT:    [[R62:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A62]], i8 [[B62]], i32 3)
+; CHECK-NEXT:    [[R63:%.*]] = call i8 @llvm.smul.fix.i8(i8 [[A63]], i8 [[B63]], i32 3)
+; CHECK-NEXT:    store i8 [[R0]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 0), align 1
+; CHECK-NEXT:    store i8 [[R1]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 1), align 1
+; CHECK-NEXT:    store i8 [[R2]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 2), align 1
+; CHECK-NEXT:    store i8 [[R3]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 3), align 1
+; CHECK-NEXT:    store i8 [[R4]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 4), align 1
+; CHECK-NEXT:    store i8 [[R5]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 5), align 1
+; CHECK-NEXT:    store i8 [[R6]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 6), align 1
+; CHECK-NEXT:    store i8 [[R7]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 7), align 1
+; CHECK-NEXT:    store i8 [[R8]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 8), align 1
+; CHECK-NEXT:    store i8 [[R9]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 9), align 1
+; CHECK-NEXT:    store i8 [[R10]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 10), align 1
+; CHECK-NEXT:    store i8 [[R11]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 11), align 1
+; CHECK-NEXT:    store i8 [[R12]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 12), align 1
+; CHECK-NEXT:    store i8 [[R13]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 13), align 1
+; CHECK-NEXT:    store i8 [[R14]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 14), align 1
+; CHECK-NEXT:    store i8 [[R15]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 15), align 1
+; CHECK-NEXT:    store i8 [[R16]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 16), align 1
+; CHECK-NEXT:    store i8 [[R17]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 17), align 1
+; CHECK-NEXT:    store i8 [[R18]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 18), align 1
+; CHECK-NEXT:    store i8 [[R19]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 19), align 1
+; CHECK-NEXT:    store i8 [[R20]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 20), align 1
+; CHECK-NEXT:    store i8 [[R21]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 21), align 1
+; CHECK-NEXT:    store i8 [[R22]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 22), align 1
+; CHECK-NEXT:    store i8 [[R23]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 23), align 1
+; CHECK-NEXT:    store i8 [[R24]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 24), align 1
+; CHECK-NEXT:    store i8 [[R25]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 25), align 1
+; CHECK-NEXT:    store i8 [[R26]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 26), align 1
+; CHECK-NEXT:    store i8 [[R27]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 27), align 1
+; CHECK-NEXT:    store i8 [[R28]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 28), align 1
+; CHECK-NEXT:    store i8 [[R29]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 29), align 1
+; CHECK-NEXT:    store i8 [[R30]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 30), align 1
+; CHECK-NEXT:    store i8 [[R31]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 31), align 1
+; CHECK-NEXT:    store i8 [[R32]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 32), align 1
+; CHECK-NEXT:    store i8 [[R33]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 33), align 1
+; CHECK-NEXT:    store i8 [[R34]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 34), align 1
+; CHECK-NEXT:    store i8 [[R35]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 35), align 1
+; CHECK-NEXT:    store i8 [[R36]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 36), align 1
+; CHECK-NEXT:    store i8 [[R37]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 37), align 1
+; CHECK-NEXT:    store i8 [[R38]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 38), align 1
+; CHECK-NEXT:    store i8 [[R39]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 39), align 1
+; CHECK-NEXT:    store i8 [[R40]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 40), align 1
+; CHECK-NEXT:    store i8 [[R41]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 41), align 1
+; CHECK-NEXT:    store i8 [[R42]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 42), align 1
+; CHECK-NEXT:    store i8 [[R43]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 43), align 1
+; CHECK-NEXT:    store i8 [[R44]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 44), align 1
+; CHECK-NEXT:    store i8 [[R45]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 45), align 1
+; CHECK-NEXT:    store i8 [[R46]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 46), align 1
+; CHECK-NEXT:    store i8 [[R47]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 47), align 1
+; CHECK-NEXT:    store i8 [[R48]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 48), align 1
+; CHECK-NEXT:    store i8 [[R49]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 49), align 1
+; CHECK-NEXT:    store i8 [[R50]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 50), align 1
+; CHECK-NEXT:    store i8 [[R51]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 51), align 1
+; CHECK-NEXT:    store i8 [[R52]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 52), align 1
+; CHECK-NEXT:    store i8 [[R53]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 53), align 1
+; CHECK-NEXT:    store i8 [[R54]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 54), align 1
+; CHECK-NEXT:    store i8 [[R55]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 55), align 1
+; CHECK-NEXT:    store i8 [[R56]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 56), align 1
+; CHECK-NEXT:    store i8 [[R57]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 57), align 1
+; CHECK-NEXT:    store i8 [[R58]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 58), align 1
+; CHECK-NEXT:    store i8 [[R59]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 59), align 1
+; CHECK-NEXT:    store i8 [[R60]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 60), align 1
+; CHECK-NEXT:    store i8 [[R61]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 61), align 1
+; CHECK-NEXT:    store i8 [[R62]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 62), align 1
+; CHECK-NEXT:    store i8 [[R63]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 63), align 1
+; CHECK-NEXT:    ret void
+;
+  %a0  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 0 ), align 1
+  %a1  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 1 ), align 1
+  %a2  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 2 ), align 1
+  %a3  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 3 ), align 1
+  %a4  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 4 ), align 1
+  %a5  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 5 ), align 1
+  %a6  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 6 ), align 1
+  %a7  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 7 ), align 1
+  %a8  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 8 ), align 1
+  %a9  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 9 ), align 1
+  %a10 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 10), align 1
+  %a11 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 11), align 1
+  %a12 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 12), align 1
+  %a13 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 13), align 1
+  %a14 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 14), align 1
+  %a15 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 15), align 1
+  %a16 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 16), align 1
+  %a17 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 17), align 1
+  %a18 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 18), align 1
+  %a19 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 19), align 1
+  %a20 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 20), align 1
+  %a21 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 21), align 1
+  %a22 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 22), align 1
+  %a23 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 23), align 1
+  %a24 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 24), align 1
+  %a25 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 25), align 1
+  %a26 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 26), align 1
+  %a27 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 27), align 1
+  %a28 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 28), align 1
+  %a29 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 29), align 1
+  %a30 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 30), align 1
+  %a31 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 31), align 1
+  %a32 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 32), align 1
+  %a33 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 33), align 1
+  %a34 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 34), align 1
+  %a35 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 35), align 1
+  %a36 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 36), align 1
+  %a37 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 37), align 1
+  %a38 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 38), align 1
+  %a39 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 39), align 1
+  %a40 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 40), align 1
+  %a41 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 41), align 1
+  %a42 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 42), align 1
+  %a43 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 43), align 1
+  %a44 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 44), align 1
+  %a45 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 45), align 1
+  %a46 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 46), align 1
+  %a47 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 47), align 1
+  %a48 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 48), align 1
+  %a49 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 49), align 1
+  %a50 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 50), align 1
+  %a51 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 51), align 1
+  %a52 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 52), align 1
+  %a53 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 53), align 1
+  %a54 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 54), align 1
+  %a55 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 55), align 1
+  %a56 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 56), align 1
+  %a57 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 57), align 1
+  %a58 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 58), align 1
+  %a59 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 59), align 1
+  %a60 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 60), align 1
+  %a61 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 61), align 1
+  %a62 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 62), align 1
+  %a63 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 63), align 1
+  %b0  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 0 ), align 1
+  %b1  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 1 ), align 1
+  %b2  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 2 ), align 1
+  %b3  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 3 ), align 1
+  %b4  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 4 ), align 1
+  %b5  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 5 ), align 1
+  %b6  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 6 ), align 1
+  %b7  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 7 ), align 1
+  %b8  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 8 ), align 1
+  %b9  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 9 ), align 1
+  %b10 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 10), align 1
+  %b11 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 11), align 1
+  %b12 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 12), align 1
+  %b13 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 13), align 1
+  %b14 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 14), align 1
+  %b15 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 15), align 1
+  %b16 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 16), align 1
+  %b17 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 17), align 1
+  %b18 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 18), align 1
+  %b19 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 19), align 1
+  %b20 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 20), align 1
+  %b21 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 21), align 1
+  %b22 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 22), align 1
+  %b23 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 23), align 1
+  %b24 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 24), align 1
+  %b25 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 25), align 1
+  %b26 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 26), align 1
+  %b27 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 27), align 1
+  %b28 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 28), align 1
+  %b29 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 29), align 1
+  %b30 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 30), align 1
+  %b31 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 31), align 1
+  %b32 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 32), align 1
+  %b33 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 33), align 1
+  %b34 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 34), align 1
+  %b35 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 35), align 1
+  %b36 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 36), align 1
+  %b37 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 37), align 1
+  %b38 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 38), align 1
+  %b39 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 39), align 1
+  %b40 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 40), align 1
+  %b41 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 41), align 1
+  %b42 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 42), align 1
+  %b43 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 43), align 1
+  %b44 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 44), align 1
+  %b45 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 45), align 1
+  %b46 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 46), align 1
+  %b47 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 47), align 1
+  %b48 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 48), align 1
+  %b49 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 49), align 1
+  %b50 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 50), align 1
+  %b51 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 51), align 1
+  %b52 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 52), align 1
+  %b53 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 53), align 1
+  %b54 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 54), align 1
+  %b55 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 55), align 1
+  %b56 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 56), align 1
+  %b57 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 57), align 1
+  %b58 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 58), align 1
+  %b59 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 59), align 1
+  %b60 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 60), align 1
+  %b61 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 61), align 1
+  %b62 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 62), align 1
+  %b63 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 63), align 1
+  %r0  = call i8 @llvm.smul.fix.i8(i8 %a0 , i8 %b0 , i32 3)
+  %r1  = call i8 @llvm.smul.fix.i8(i8 %a1 , i8 %b1 , i32 3)
+  %r2  = call i8 @llvm.smul.fix.i8(i8 %a2 , i8 %b2 , i32 3)
+  %r3  = call i8 @llvm.smul.fix.i8(i8 %a3 , i8 %b3 , i32 3)
+  %r4  = call i8 @llvm.smul.fix.i8(i8 %a4 , i8 %b4 , i32 3)
+  %r5  = call i8 @llvm.smul.fix.i8(i8 %a5 , i8 %b5 , i32 3)
+  %r6  = call i8 @llvm.smul.fix.i8(i8 %a6 , i8 %b6 , i32 3)
+  %r7  = call i8 @llvm.smul.fix.i8(i8 %a7 , i8 %b7 , i32 3)
+  %r8  = call i8 @llvm.smul.fix.i8(i8 %a8 , i8 %b8 , i32 3)
+  %r9  = call i8 @llvm.smul.fix.i8(i8 %a9 , i8 %b9 , i32 3)
+  %r10 = call i8 @llvm.smul.fix.i8(i8 %a10, i8 %b10, i32 3)
+  %r11 = call i8 @llvm.smul.fix.i8(i8 %a11, i8 %b11, i32 3)
+  %r12 = call i8 @llvm.smul.fix.i8(i8 %a12, i8 %b12, i32 3)
+  %r13 = call i8 @llvm.smul.fix.i8(i8 %a13, i8 %b13, i32 3)
+  %r14 = call i8 @llvm.smul.fix.i8(i8 %a14, i8 %b14, i32 3)
+  %r15 = call i8 @llvm.smul.fix.i8(i8 %a15, i8 %b15, i32 3)
+  %r16 = call i8 @llvm.smul.fix.i8(i8 %a16, i8 %b16, i32 3)
+  %r17 = call i8 @llvm.smul.fix.i8(i8 %a17, i8 %b17, i32 3)
+  %r18 = call i8 @llvm.smul.fix.i8(i8 %a18, i8 %b18, i32 3)
+  %r19 = call i8 @llvm.smul.fix.i8(i8 %a19, i8 %b19, i32 3)
+  %r20 = call i8 @llvm.smul.fix.i8(i8 %a20, i8 %b20, i32 3)
+  %r21 = call i8 @llvm.smul.fix.i8(i8 %a21, i8 %b21, i32 3)
+  %r22 = call i8 @llvm.smul.fix.i8(i8 %a22, i8 %b22, i32 3)
+  %r23 = call i8 @llvm.smul.fix.i8(i8 %a23, i8 %b23, i32 3)
+  %r24 = call i8 @llvm.smul.fix.i8(i8 %a24, i8 %b24, i32 3)
+  %r25 = call i8 @llvm.smul.fix.i8(i8 %a25, i8 %b25, i32 3)
+  %r26 = call i8 @llvm.smul.fix.i8(i8 %a26, i8 %b26, i32 3)
+  %r27 = call i8 @llvm.smul.fix.i8(i8 %a27, i8 %b27, i32 3)
+  %r28 = call i8 @llvm.smul.fix.i8(i8 %a28, i8 %b28, i32 3)
+  %r29 = call i8 @llvm.smul.fix.i8(i8 %a29, i8 %b29, i32 3)
+  %r30 = call i8 @llvm.smul.fix.i8(i8 %a30, i8 %b30, i32 3)
+  %r31 = call i8 @llvm.smul.fix.i8(i8 %a31, i8 %b31, i32 3)
+  %r32 = call i8 @llvm.smul.fix.i8(i8 %a32, i8 %b32, i32 3)
+  %r33 = call i8 @llvm.smul.fix.i8(i8 %a33, i8 %b33, i32 3)
+  %r34 = call i8 @llvm.smul.fix.i8(i8 %a34, i8 %b34, i32 3)
+  %r35 = call i8 @llvm.smul.fix.i8(i8 %a35, i8 %b35, i32 3)
+  %r36 = call i8 @llvm.smul.fix.i8(i8 %a36, i8 %b36, i32 3)
+  %r37 = call i8 @llvm.smul.fix.i8(i8 %a37, i8 %b37, i32 3)
+  %r38 = call i8 @llvm.smul.fix.i8(i8 %a38, i8 %b38, i32 3)
+  %r39 = call i8 @llvm.smul.fix.i8(i8 %a39, i8 %b39, i32 3)
+  %r40 = call i8 @llvm.smul.fix.i8(i8 %a40, i8 %b40, i32 3)
+  %r41 = call i8 @llvm.smul.fix.i8(i8 %a41, i8 %b41, i32 3)
+  %r42 = call i8 @llvm.smul.fix.i8(i8 %a42, i8 %b42, i32 3)
+  %r43 = call i8 @llvm.smul.fix.i8(i8 %a43, i8 %b43, i32 3)
+  %r44 = call i8 @llvm.smul.fix.i8(i8 %a44, i8 %b44, i32 3)
+  %r45 = call i8 @llvm.smul.fix.i8(i8 %a45, i8 %b45, i32 3)
+  %r46 = call i8 @llvm.smul.fix.i8(i8 %a46, i8 %b46, i32 3)
+  %r47 = call i8 @llvm.smul.fix.i8(i8 %a47, i8 %b47, i32 3)
+  %r48 = call i8 @llvm.smul.fix.i8(i8 %a48, i8 %b48, i32 3)
+  %r49 = call i8 @llvm.smul.fix.i8(i8 %a49, i8 %b49, i32 3)
+  %r50 = call i8 @llvm.smul.fix.i8(i8 %a50, i8 %b50, i32 3)
+  %r51 = call i8 @llvm.smul.fix.i8(i8 %a51, i8 %b51, i32 3)
+  %r52 = call i8 @llvm.smul.fix.i8(i8 %a52, i8 %b52, i32 3)
+  %r53 = call i8 @llvm.smul.fix.i8(i8 %a53, i8 %b53, i32 3)
+  %r54 = call i8 @llvm.smul.fix.i8(i8 %a54, i8 %b54, i32 3)
+  %r55 = call i8 @llvm.smul.fix.i8(i8 %a55, i8 %b55, i32 3)
+  %r56 = call i8 @llvm.smul.fix.i8(i8 %a56, i8 %b56, i32 3)
+  %r57 = call i8 @llvm.smul.fix.i8(i8 %a57, i8 %b57, i32 3)
+  %r58 = call i8 @llvm.smul.fix.i8(i8 %a58, i8 %b58, i32 3)
+  %r59 = call i8 @llvm.smul.fix.i8(i8 %a59, i8 %b59, i32 3)
+  %r60 = call i8 @llvm.smul.fix.i8(i8 %a60, i8 %b60, i32 3)
+  %r61 = call i8 @llvm.smul.fix.i8(i8 %a61, i8 %b61, i32 3)
+  %r62 = call i8 @llvm.smul.fix.i8(i8 %a62, i8 %b62, i32 3)
+  %r63 = call i8 @llvm.smul.fix.i8(i8 %a63, i8 %b63, i32 3)
+  store i8 %r0 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 0 ), align 1
+  store i8 %r1 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 1 ), align 1
+  store i8 %r2 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 2 ), align 1
+  store i8 %r3 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 3 ), align 1
+  store i8 %r4 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 4 ), align 1
+  store i8 %r5 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 5 ), align 1
+  store i8 %r6 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 6 ), align 1
+  store i8 %r7 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 7 ), align 1
+  store i8 %r8 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 8 ), align 1
+  store i8 %r9 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 9 ), align 1
+  store i8 %r10, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 10), align 1
+  store i8 %r11, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 11), align 1
+  store i8 %r12, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 12), align 1
+  store i8 %r13, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 13), align 1
+  store i8 %r14, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 14), align 1
+  store i8 %r15, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 15), align 1
+  store i8 %r16, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 16), align 1
+  store i8 %r17, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 17), align 1
+  store i8 %r18, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 18), align 1
+  store i8 %r19, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 19), align 1
+  store i8 %r20, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 20), align 1
+  store i8 %r21, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 21), align 1
+  store i8 %r22, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 22), align 1
+  store i8 %r23, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 23), align 1
+  store i8 %r24, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 24), align 1
+  store i8 %r25, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 25), align 1
+  store i8 %r26, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 26), align 1
+  store i8 %r27, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 27), align 1
+  store i8 %r28, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 28), align 1
+  store i8 %r29, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 29), align 1
+  store i8 %r30, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 30), align 1
+  store i8 %r31, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 31), align 1
+  store i8 %r32, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 32), align 1
+  store i8 %r33, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 33), align 1
+  store i8 %r34, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 34), align 1
+  store i8 %r35, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 35), align 1
+  store i8 %r36, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 36), align 1
+  store i8 %r37, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 37), align 1
+  store i8 %r38, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 38), align 1
+  store i8 %r39, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 39), align 1
+  store i8 %r40, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 40), align 1
+  store i8 %r41, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 41), align 1
+  store i8 %r42, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 42), align 1
+  store i8 %r43, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 43), align 1
+  store i8 %r44, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 44), align 1
+  store i8 %r45, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 45), align 1
+  store i8 %r46, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 46), align 1
+  store i8 %r47, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 47), align 1
+  store i8 %r48, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 48), align 1
+  store i8 %r49, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 49), align 1
+  store i8 %r50, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 50), align 1
+  store i8 %r51, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 51), align 1
+  store i8 %r52, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 52), align 1
+  store i8 %r53, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 53), align 1
+  store i8 %r54, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 54), align 1
+  store i8 %r55, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 55), align 1
+  store i8 %r56, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 56), align 1
+  store i8 %r57, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 57), align 1
+  store i8 %r58, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 58), align 1
+  store i8 %r59, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 59), align 1
+  store i8 %r60, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 60), align 1
+  store i8 %r61, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 61), align 1
+  store i8 %r62, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 62), align 1
+  store i8 %r63, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 63), align 1
+  ret void
+}
+
+declare i64 @llvm.umul.fix.i64(i64, i64, i32)
+declare i32 @llvm.umul.fix.i32(i32, i32, i32)
+declare i16 @llvm.umul.fix.i16(i16, i16, i32)
+declare i8  @llvm.umul.fix.i8 (i8 , i8 , i32)
+
+define void @umul_v8i64() {
+; CHECK-LABEL: @umul_v8i64(
+; CHECK-NEXT:    [[A0:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 0), align 8
+; CHECK-NEXT:    [[A1:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 1), align 8
+; CHECK-NEXT:    [[A2:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 2), align 8
+; CHECK-NEXT:    [[A3:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 3), align 8
+; CHECK-NEXT:    [[A4:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 4), align 8
+; CHECK-NEXT:    [[A5:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 5), align 8
+; CHECK-NEXT:    [[A6:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 6), align 8
+; CHECK-NEXT:    [[A7:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 7), align 8
+; CHECK-NEXT:    [[B0:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 0), align 8
+; CHECK-NEXT:    [[B1:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 1), align 8
+; CHECK-NEXT:    [[B2:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 2), align 8
+; CHECK-NEXT:    [[B3:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 3), align 8
+; CHECK-NEXT:    [[B4:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 4), align 8
+; CHECK-NEXT:    [[B5:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 5), align 8
+; CHECK-NEXT:    [[B6:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 6), align 8
+; CHECK-NEXT:    [[B7:%.*]] = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 7), align 8
+; CHECK-NEXT:    [[R0:%.*]] = call i64 @llvm.umul.fix.i64(i64 [[A0]], i64 [[B0]], i32 3)
+; CHECK-NEXT:    [[R1:%.*]] = call i64 @llvm.umul.fix.i64(i64 [[A1]], i64 [[B1]], i32 3)
+; CHECK-NEXT:    [[R2:%.*]] = call i64 @llvm.umul.fix.i64(i64 [[A2]], i64 [[B2]], i32 3)
+; CHECK-NEXT:    [[R3:%.*]] = call i64 @llvm.umul.fix.i64(i64 [[A3]], i64 [[B3]], i32 3)
+; CHECK-NEXT:    [[R4:%.*]] = call i64 @llvm.umul.fix.i64(i64 [[A4]], i64 [[B4]], i32 3)
+; CHECK-NEXT:    [[R5:%.*]] = call i64 @llvm.umul.fix.i64(i64 [[A5]], i64 [[B5]], i32 3)
+; CHECK-NEXT:    [[R6:%.*]] = call i64 @llvm.umul.fix.i64(i64 [[A6]], i64 [[B6]], i32 3)
+; CHECK-NEXT:    [[R7:%.*]] = call i64 @llvm.umul.fix.i64(i64 [[A7]], i64 [[B7]], i32 3)
+; CHECK-NEXT:    store i64 [[R0]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 0), align 8
+; CHECK-NEXT:    store i64 [[R1]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 1), align 8
+; CHECK-NEXT:    store i64 [[R2]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 2), align 8
+; CHECK-NEXT:    store i64 [[R3]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 3), align 8
+; CHECK-NEXT:    store i64 [[R4]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 4), align 8
+; CHECK-NEXT:    store i64 [[R5]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 5), align 8
+; CHECK-NEXT:    store i64 [[R6]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 6), align 8
+; CHECK-NEXT:    store i64 [[R7]], i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 7), align 8
+; CHECK-NEXT:    ret void
+;
+  %a0 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 0), align 8
+  %a1 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 1), align 8
+  %a2 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 2), align 8
+  %a3 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 3), align 8
+  %a4 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 4), align 8
+  %a5 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 5), align 8
+  %a6 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 6), align 8
+  %a7 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @a64, i32 0, i64 7), align 8
+  %b0 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 0), align 8
+  %b1 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 1), align 8
+  %b2 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 2), align 8
+  %b3 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 3), align 8
+  %b4 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 4), align 8
+  %b5 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 5), align 8
+  %b6 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 6), align 8
+  %b7 = load i64, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @b64, i32 0, i64 7), align 8
+  %r0 = call i64 @llvm.umul.fix.i64(i64 %a0, i64 %b0, i32 3)
+  %r1 = call i64 @llvm.umul.fix.i64(i64 %a1, i64 %b1, i32 3)
+  %r2 = call i64 @llvm.umul.fix.i64(i64 %a2, i64 %b2, i32 3)
+  %r3 = call i64 @llvm.umul.fix.i64(i64 %a3, i64 %b3, i32 3)
+  %r4 = call i64 @llvm.umul.fix.i64(i64 %a4, i64 %b4, i32 3)
+  %r5 = call i64 @llvm.umul.fix.i64(i64 %a5, i64 %b5, i32 3)
+  %r6 = call i64 @llvm.umul.fix.i64(i64 %a6, i64 %b6, i32 3)
+  %r7 = call i64 @llvm.umul.fix.i64(i64 %a7, i64 %b7, i32 3)
+  store i64 %r0, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 0), align 8
+  store i64 %r1, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 1), align 8
+  store i64 %r2, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 2), align 8
+  store i64 %r3, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 3), align 8
+  store i64 %r4, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 4), align 8
+  store i64 %r5, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 5), align 8
+  store i64 %r6, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 6), align 8
+  store i64 %r7, i64* getelementptr inbounds ([8 x i64], [8 x i64]* @c64, i32 0, i64 7), align 8
+  ret void
+}
+
+define void @umul_v16i32() {
+; CHECK-LABEL: @umul_v16i32(
+; CHECK-NEXT:    [[A0:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 0), align 4
+; CHECK-NEXT:    [[A1:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 1), align 4
+; CHECK-NEXT:    [[A2:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 2), align 4
+; CHECK-NEXT:    [[A3:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 3), align 4
+; CHECK-NEXT:    [[A4:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 4), align 4
+; CHECK-NEXT:    [[A5:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 5), align 4
+; CHECK-NEXT:    [[A6:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 6), align 4
+; CHECK-NEXT:    [[A7:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 7), align 4
+; CHECK-NEXT:    [[A8:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 8), align 4
+; CHECK-NEXT:    [[A9:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 9), align 4
+; CHECK-NEXT:    [[A10:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 10), align 4
+; CHECK-NEXT:    [[A11:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 11), align 4
+; CHECK-NEXT:    [[A12:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 12), align 4
+; CHECK-NEXT:    [[A13:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 13), align 4
+; CHECK-NEXT:    [[A14:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 14), align 4
+; CHECK-NEXT:    [[A15:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 15), align 4
+; CHECK-NEXT:    [[B0:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 0), align 4
+; CHECK-NEXT:    [[B1:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 1), align 4
+; CHECK-NEXT:    [[B2:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 2), align 4
+; CHECK-NEXT:    [[B3:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 3), align 4
+; CHECK-NEXT:    [[B4:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 4), align 4
+; CHECK-NEXT:    [[B5:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 5), align 4
+; CHECK-NEXT:    [[B6:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 6), align 4
+; CHECK-NEXT:    [[B7:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 7), align 4
+; CHECK-NEXT:    [[B8:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 8), align 4
+; CHECK-NEXT:    [[B9:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 9), align 4
+; CHECK-NEXT:    [[B10:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 10), align 4
+; CHECK-NEXT:    [[B11:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 11), align 4
+; CHECK-NEXT:    [[B12:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 12), align 4
+; CHECK-NEXT:    [[B13:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 13), align 4
+; CHECK-NEXT:    [[B14:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 14), align 4
+; CHECK-NEXT:    [[B15:%.*]] = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 15), align 4
+; CHECK-NEXT:    [[R0:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A0]], i32 [[B0]], i32 3)
+; CHECK-NEXT:    [[R1:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A1]], i32 [[B1]], i32 3)
+; CHECK-NEXT:    [[R2:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A2]], i32 [[B2]], i32 3)
+; CHECK-NEXT:    [[R3:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A3]], i32 [[B3]], i32 3)
+; CHECK-NEXT:    [[R4:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A4]], i32 [[B4]], i32 3)
+; CHECK-NEXT:    [[R5:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A5]], i32 [[B5]], i32 3)
+; CHECK-NEXT:    [[R6:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A6]], i32 [[B6]], i32 3)
+; CHECK-NEXT:    [[R7:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A7]], i32 [[B7]], i32 3)
+; CHECK-NEXT:    [[R8:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A8]], i32 [[B8]], i32 3)
+; CHECK-NEXT:    [[R9:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A9]], i32 [[B9]], i32 3)
+; CHECK-NEXT:    [[R10:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A10]], i32 [[B10]], i32 3)
+; CHECK-NEXT:    [[R11:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A11]], i32 [[B11]], i32 3)
+; CHECK-NEXT:    [[R12:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A12]], i32 [[B12]], i32 3)
+; CHECK-NEXT:    [[R13:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A13]], i32 [[B13]], i32 3)
+; CHECK-NEXT:    [[R14:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A14]], i32 [[B14]], i32 3)
+; CHECK-NEXT:    [[R15:%.*]] = call i32 @llvm.umul.fix.i32(i32 [[A15]], i32 [[B15]], i32 3)
+; CHECK-NEXT:    store i32 [[R0]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 0), align 4
+; CHECK-NEXT:    store i32 [[R1]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 1), align 4
+; CHECK-NEXT:    store i32 [[R2]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 2), align 4
+; CHECK-NEXT:    store i32 [[R3]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 3), align 4
+; CHECK-NEXT:    store i32 [[R4]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 4), align 4
+; CHECK-NEXT:    store i32 [[R5]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 5), align 4
+; CHECK-NEXT:    store i32 [[R6]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 6), align 4
+; CHECK-NEXT:    store i32 [[R7]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 7), align 4
+; CHECK-NEXT:    store i32 [[R8]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 8), align 4
+; CHECK-NEXT:    store i32 [[R9]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 9), align 4
+; CHECK-NEXT:    store i32 [[R10]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 10), align 4
+; CHECK-NEXT:    store i32 [[R11]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 11), align 4
+; CHECK-NEXT:    store i32 [[R12]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 12), align 4
+; CHECK-NEXT:    store i32 [[R13]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 13), align 4
+; CHECK-NEXT:    store i32 [[R14]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 14), align 4
+; CHECK-NEXT:    store i32 [[R15]], i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 15), align 4
+; CHECK-NEXT:    ret void
+;
+  %a0  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 0 ), align 4
+  %a1  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 1 ), align 4
+  %a2  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 2 ), align 4
+  %a3  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 3 ), align 4
+  %a4  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 4 ), align 4
+  %a5  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 5 ), align 4
+  %a6  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 6 ), align 4
+  %a7  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 7 ), align 4
+  %a8  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 8 ), align 4
+  %a9  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 9 ), align 4
+  %a10 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 10), align 4
+  %a11 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 11), align 4
+  %a12 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 12), align 4
+  %a13 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 13), align 4
+  %a14 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 14), align 4
+  %a15 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @a32, i32 0, i64 15), align 4
+  %b0  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 0 ), align 4
+  %b1  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 1 ), align 4
+  %b2  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 2 ), align 4
+  %b3  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 3 ), align 4
+  %b4  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 4 ), align 4
+  %b5  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 5 ), align 4
+  %b6  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 6 ), align 4
+  %b7  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 7 ), align 4
+  %b8  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 8 ), align 4
+  %b9  = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 9 ), align 4
+  %b10 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 10), align 4
+  %b11 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 11), align 4
+  %b12 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 12), align 4
+  %b13 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 13), align 4
+  %b14 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 14), align 4
+  %b15 = load i32, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @b32, i32 0, i64 15), align 4
+  %r0  = call i32 @llvm.umul.fix.i32(i32 %a0 , i32 %b0 , i32 3)
+  %r1  = call i32 @llvm.umul.fix.i32(i32 %a1 , i32 %b1 , i32 3)
+  %r2  = call i32 @llvm.umul.fix.i32(i32 %a2 , i32 %b2 , i32 3)
+  %r3  = call i32 @llvm.umul.fix.i32(i32 %a3 , i32 %b3 , i32 3)
+  %r4  = call i32 @llvm.umul.fix.i32(i32 %a4 , i32 %b4 , i32 3)
+  %r5  = call i32 @llvm.umul.fix.i32(i32 %a5 , i32 %b5 , i32 3)
+  %r6  = call i32 @llvm.umul.fix.i32(i32 %a6 , i32 %b6 , i32 3)
+  %r7  = call i32 @llvm.umul.fix.i32(i32 %a7 , i32 %b7 , i32 3)
+  %r8  = call i32 @llvm.umul.fix.i32(i32 %a8 , i32 %b8 , i32 3)
+  %r9  = call i32 @llvm.umul.fix.i32(i32 %a9 , i32 %b9 , i32 3)
+  %r10 = call i32 @llvm.umul.fix.i32(i32 %a10, i32 %b10, i32 3)
+  %r11 = call i32 @llvm.umul.fix.i32(i32 %a11, i32 %b11, i32 3)
+  %r12 = call i32 @llvm.umul.fix.i32(i32 %a12, i32 %b12, i32 3)
+  %r13 = call i32 @llvm.umul.fix.i32(i32 %a13, i32 %b13, i32 3)
+  %r14 = call i32 @llvm.umul.fix.i32(i32 %a14, i32 %b14, i32 3)
+  %r15 = call i32 @llvm.umul.fix.i32(i32 %a15, i32 %b15, i32 3)
+  store i32 %r0 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 0 ), align 4
+  store i32 %r1 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 1 ), align 4
+  store i32 %r2 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 2 ), align 4
+  store i32 %r3 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 3 ), align 4
+  store i32 %r4 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 4 ), align 4
+  store i32 %r5 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 5 ), align 4
+  store i32 %r6 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 6 ), align 4
+  store i32 %r7 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 7 ), align 4
+  store i32 %r8 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 8 ), align 4
+  store i32 %r9 , i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 9 ), align 4
+  store i32 %r10, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 10), align 4
+  store i32 %r11, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 11), align 4
+  store i32 %r12, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 12), align 4
+  store i32 %r13, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 13), align 4
+  store i32 %r14, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 14), align 4
+  store i32 %r15, i32* getelementptr inbounds ([16 x i32], [16 x i32]* @c32, i32 0, i64 15), align 4
+  ret void
+}
+
+define void @umul_v32i16() {
+; CHECK-LABEL: @umul_v32i16(
+; CHECK-NEXT:    [[A0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 0), align 2
+; CHECK-NEXT:    [[A1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 1), align 2
+; CHECK-NEXT:    [[A2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 2), align 2
+; CHECK-NEXT:    [[A3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 3), align 2
+; CHECK-NEXT:    [[A4:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 4), align 2
+; CHECK-NEXT:    [[A5:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 5), align 2
+; CHECK-NEXT:    [[A6:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 6), align 2
+; CHECK-NEXT:    [[A7:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 7), align 2
+; CHECK-NEXT:    [[A8:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 8), align 2
+; CHECK-NEXT:    [[A9:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 9), align 2
+; CHECK-NEXT:    [[A10:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 10), align 2
+; CHECK-NEXT:    [[A11:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 11), align 2
+; CHECK-NEXT:    [[A12:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 12), align 2
+; CHECK-NEXT:    [[A13:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 13), align 2
+; CHECK-NEXT:    [[A14:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 14), align 2
+; CHECK-NEXT:    [[A15:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 15), align 2
+; CHECK-NEXT:    [[A16:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 16), align 2
+; CHECK-NEXT:    [[A17:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 17), align 2
+; CHECK-NEXT:    [[A18:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 18), align 2
+; CHECK-NEXT:    [[A19:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 19), align 2
+; CHECK-NEXT:    [[A20:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 20), align 2
+; CHECK-NEXT:    [[A21:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 21), align 2
+; CHECK-NEXT:    [[A22:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 22), align 2
+; CHECK-NEXT:    [[A23:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 23), align 2
+; CHECK-NEXT:    [[A24:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 24), align 2
+; CHECK-NEXT:    [[A25:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 25), align 2
+; CHECK-NEXT:    [[A26:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 26), align 2
+; CHECK-NEXT:    [[A27:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 27), align 2
+; CHECK-NEXT:    [[A28:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 28), align 2
+; CHECK-NEXT:    [[A29:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 29), align 2
+; CHECK-NEXT:    [[A30:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 30), align 2
+; CHECK-NEXT:    [[A31:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 31), align 2
+; CHECK-NEXT:    [[B0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 0), align 2
+; CHECK-NEXT:    [[B1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 1), align 2
+; CHECK-NEXT:    [[B2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 2), align 2
+; CHECK-NEXT:    [[B3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 3), align 2
+; CHECK-NEXT:    [[B4:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 4), align 2
+; CHECK-NEXT:    [[B5:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 5), align 2
+; CHECK-NEXT:    [[B6:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 6), align 2
+; CHECK-NEXT:    [[B7:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 7), align 2
+; CHECK-NEXT:    [[B8:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 8), align 2
+; CHECK-NEXT:    [[B9:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 9), align 2
+; CHECK-NEXT:    [[B10:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 10), align 2
+; CHECK-NEXT:    [[B11:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 11), align 2
+; CHECK-NEXT:    [[B12:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 12), align 2
+; CHECK-NEXT:    [[B13:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 13), align 2
+; CHECK-NEXT:    [[B14:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 14), align 2
+; CHECK-NEXT:    [[B15:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 15), align 2
+; CHECK-NEXT:    [[B16:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 16), align 2
+; CHECK-NEXT:    [[B17:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 17), align 2
+; CHECK-NEXT:    [[B18:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 18), align 2
+; CHECK-NEXT:    [[B19:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 19), align 2
+; CHECK-NEXT:    [[B20:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 20), align 2
+; CHECK-NEXT:    [[B21:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 21), align 2
+; CHECK-NEXT:    [[B22:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 22), align 2
+; CHECK-NEXT:    [[B23:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 23), align 2
+; CHECK-NEXT:    [[B24:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 24), align 2
+; CHECK-NEXT:    [[B25:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 25), align 2
+; CHECK-NEXT:    [[B26:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 26), align 2
+; CHECK-NEXT:    [[B27:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 27), align 2
+; CHECK-NEXT:    [[B28:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 28), align 2
+; CHECK-NEXT:    [[B29:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 29), align 2
+; CHECK-NEXT:    [[B30:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 30), align 2
+; CHECK-NEXT:    [[B31:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 31), align 2
+; CHECK-NEXT:    [[R0:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A0]], i16 [[B0]], i32 3)
+; CHECK-NEXT:    [[R1:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A1]], i16 [[B1]], i32 3)
+; CHECK-NEXT:    [[R2:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A2]], i16 [[B2]], i32 3)
+; CHECK-NEXT:    [[R3:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A3]], i16 [[B3]], i32 3)
+; CHECK-NEXT:    [[R4:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A4]], i16 [[B4]], i32 3)
+; CHECK-NEXT:    [[R5:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A5]], i16 [[B5]], i32 3)
+; CHECK-NEXT:    [[R6:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A6]], i16 [[B6]], i32 3)
+; CHECK-NEXT:    [[R7:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A7]], i16 [[B7]], i32 3)
+; CHECK-NEXT:    [[R8:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A8]], i16 [[B8]], i32 3)
+; CHECK-NEXT:    [[R9:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A9]], i16 [[B9]], i32 3)
+; CHECK-NEXT:    [[R10:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A10]], i16 [[B10]], i32 3)
+; CHECK-NEXT:    [[R11:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A11]], i16 [[B11]], i32 3)
+; CHECK-NEXT:    [[R12:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A12]], i16 [[B12]], i32 3)
+; CHECK-NEXT:    [[R13:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A13]], i16 [[B13]], i32 3)
+; CHECK-NEXT:    [[R14:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A14]], i16 [[B14]], i32 3)
+; CHECK-NEXT:    [[R15:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A15]], i16 [[B15]], i32 3)
+; CHECK-NEXT:    [[R16:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A16]], i16 [[B16]], i32 3)
+; CHECK-NEXT:    [[R17:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A17]], i16 [[B17]], i32 3)
+; CHECK-NEXT:    [[R18:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A18]], i16 [[B18]], i32 3)
+; CHECK-NEXT:    [[R19:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A19]], i16 [[B19]], i32 3)
+; CHECK-NEXT:    [[R20:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A20]], i16 [[B20]], i32 3)
+; CHECK-NEXT:    [[R21:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A21]], i16 [[B21]], i32 3)
+; CHECK-NEXT:    [[R22:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A22]], i16 [[B22]], i32 3)
+; CHECK-NEXT:    [[R23:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A23]], i16 [[B23]], i32 3)
+; CHECK-NEXT:    [[R24:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A24]], i16 [[B24]], i32 3)
+; CHECK-NEXT:    [[R25:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A25]], i16 [[B25]], i32 3)
+; CHECK-NEXT:    [[R26:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A26]], i16 [[B26]], i32 3)
+; CHECK-NEXT:    [[R27:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A27]], i16 [[B27]], i32 3)
+; CHECK-NEXT:    [[R28:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A28]], i16 [[B28]], i32 3)
+; CHECK-NEXT:    [[R29:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A29]], i16 [[B29]], i32 3)
+; CHECK-NEXT:    [[R30:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A30]], i16 [[B30]], i32 3)
+; CHECK-NEXT:    [[R31:%.*]] = call i16 @llvm.umul.fix.i16(i16 [[A31]], i16 [[B31]], i32 3)
+; CHECK-NEXT:    store i16 [[R0]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 0), align 2
+; CHECK-NEXT:    store i16 [[R1]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 1), align 2
+; CHECK-NEXT:    store i16 [[R2]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 2), align 2
+; CHECK-NEXT:    store i16 [[R3]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 3), align 2
+; CHECK-NEXT:    store i16 [[R4]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 4), align 2
+; CHECK-NEXT:    store i16 [[R5]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 5), align 2
+; CHECK-NEXT:    store i16 [[R6]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 6), align 2
+; CHECK-NEXT:    store i16 [[R7]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 7), align 2
+; CHECK-NEXT:    store i16 [[R8]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 8), align 2
+; CHECK-NEXT:    store i16 [[R9]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 9), align 2
+; CHECK-NEXT:    store i16 [[R10]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 10), align 2
+; CHECK-NEXT:    store i16 [[R11]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 11), align 2
+; CHECK-NEXT:    store i16 [[R12]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 12), align 2
+; CHECK-NEXT:    store i16 [[R13]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 13), align 2
+; CHECK-NEXT:    store i16 [[R14]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 14), align 2
+; CHECK-NEXT:    store i16 [[R15]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 15), align 2
+; CHECK-NEXT:    store i16 [[R16]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 16), align 2
+; CHECK-NEXT:    store i16 [[R17]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 17), align 2
+; CHECK-NEXT:    store i16 [[R18]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 18), align 2
+; CHECK-NEXT:    store i16 [[R19]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 19), align 2
+; CHECK-NEXT:    store i16 [[R20]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 20), align 2
+; CHECK-NEXT:    store i16 [[R21]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 21), align 2
+; CHECK-NEXT:    store i16 [[R22]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 22), align 2
+; CHECK-NEXT:    store i16 [[R23]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 23), align 2
+; CHECK-NEXT:    store i16 [[R24]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 24), align 2
+; CHECK-NEXT:    store i16 [[R25]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 25), align 2
+; CHECK-NEXT:    store i16 [[R26]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 26), align 2
+; CHECK-NEXT:    store i16 [[R27]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 27), align 2
+; CHECK-NEXT:    store i16 [[R28]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 28), align 2
+; CHECK-NEXT:    store i16 [[R29]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 29), align 2
+; CHECK-NEXT:    store i16 [[R30]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 30), align 2
+; CHECK-NEXT:    store i16 [[R31]], i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 31), align 2
+; CHECK-NEXT:    ret void
+;
+  %a0  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 0 ), align 2
+  %a1  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 1 ), align 2
+  %a2  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 2 ), align 2
+  %a3  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 3 ), align 2
+  %a4  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 4 ), align 2
+  %a5  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 5 ), align 2
+  %a6  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 6 ), align 2
+  %a7  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 7 ), align 2
+  %a8  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 8 ), align 2
+  %a9  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 9 ), align 2
+  %a10 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 10), align 2
+  %a11 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 11), align 2
+  %a12 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 12), align 2
+  %a13 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 13), align 2
+  %a14 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 14), align 2
+  %a15 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 15), align 2
+  %a16 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 16), align 2
+  %a17 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 17), align 2
+  %a18 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 18), align 2
+  %a19 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 19), align 2
+  %a20 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 20), align 2
+  %a21 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 21), align 2
+  %a22 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 22), align 2
+  %a23 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 23), align 2
+  %a24 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 24), align 2
+  %a25 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 25), align 2
+  %a26 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 26), align 2
+  %a27 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 27), align 2
+  %a28 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 28), align 2
+  %a29 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 29), align 2
+  %a30 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 30), align 2
+  %a31 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @a16, i32 0, i64 31), align 2
+  %b0  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 0 ), align 2
+  %b1  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 1 ), align 2
+  %b2  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 2 ), align 2
+  %b3  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 3 ), align 2
+  %b4  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 4 ), align 2
+  %b5  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 5 ), align 2
+  %b6  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 6 ), align 2
+  %b7  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 7 ), align 2
+  %b8  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 8 ), align 2
+  %b9  = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 9 ), align 2
+  %b10 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 10), align 2
+  %b11 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 11), align 2
+  %b12 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 12), align 2
+  %b13 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 13), align 2
+  %b14 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 14), align 2
+  %b15 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 15), align 2
+  %b16 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 16), align 2
+  %b17 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 17), align 2
+  %b18 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 18), align 2
+  %b19 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 19), align 2
+  %b20 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 20), align 2
+  %b21 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 21), align 2
+  %b22 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 22), align 2
+  %b23 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 23), align 2
+  %b24 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 24), align 2
+  %b25 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 25), align 2
+  %b26 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 26), align 2
+  %b27 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 27), align 2
+  %b28 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 28), align 2
+  %b29 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 29), align 2
+  %b30 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 30), align 2
+  %b31 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @b16, i32 0, i64 31), align 2
+  %r0  = call i16 @llvm.umul.fix.i16(i16 %a0 , i16 %b0 , i32 3)
+  %r1  = call i16 @llvm.umul.fix.i16(i16 %a1 , i16 %b1 , i32 3)
+  %r2  = call i16 @llvm.umul.fix.i16(i16 %a2 , i16 %b2 , i32 3)
+  %r3  = call i16 @llvm.umul.fix.i16(i16 %a3 , i16 %b3 , i32 3)
+  %r4  = call i16 @llvm.umul.fix.i16(i16 %a4 , i16 %b4 , i32 3)
+  %r5  = call i16 @llvm.umul.fix.i16(i16 %a5 , i16 %b5 , i32 3)
+  %r6  = call i16 @llvm.umul.fix.i16(i16 %a6 , i16 %b6 , i32 3)
+  %r7  = call i16 @llvm.umul.fix.i16(i16 %a7 , i16 %b7 , i32 3)
+  %r8  = call i16 @llvm.umul.fix.i16(i16 %a8 , i16 %b8 , i32 3)
+  %r9  = call i16 @llvm.umul.fix.i16(i16 %a9 , i16 %b9 , i32 3)
+  %r10 = call i16 @llvm.umul.fix.i16(i16 %a10, i16 %b10, i32 3)
+  %r11 = call i16 @llvm.umul.fix.i16(i16 %a11, i16 %b11, i32 3)
+  %r12 = call i16 @llvm.umul.fix.i16(i16 %a12, i16 %b12, i32 3)
+  %r13 = call i16 @llvm.umul.fix.i16(i16 %a13, i16 %b13, i32 3)
+  %r14 = call i16 @llvm.umul.fix.i16(i16 %a14, i16 %b14, i32 3)
+  %r15 = call i16 @llvm.umul.fix.i16(i16 %a15, i16 %b15, i32 3)
+  %r16 = call i16 @llvm.umul.fix.i16(i16 %a16, i16 %b16, i32 3)
+  %r17 = call i16 @llvm.umul.fix.i16(i16 %a17, i16 %b17, i32 3)
+  %r18 = call i16 @llvm.umul.fix.i16(i16 %a18, i16 %b18, i32 3)
+  %r19 = call i16 @llvm.umul.fix.i16(i16 %a19, i16 %b19, i32 3)
+  %r20 = call i16 @llvm.umul.fix.i16(i16 %a20, i16 %b20, i32 3)
+  %r21 = call i16 @llvm.umul.fix.i16(i16 %a21, i16 %b21, i32 3)
+  %r22 = call i16 @llvm.umul.fix.i16(i16 %a22, i16 %b22, i32 3)
+  %r23 = call i16 @llvm.umul.fix.i16(i16 %a23, i16 %b23, i32 3)
+  %r24 = call i16 @llvm.umul.fix.i16(i16 %a24, i16 %b24, i32 3)
+  %r25 = call i16 @llvm.umul.fix.i16(i16 %a25, i16 %b25, i32 3)
+  %r26 = call i16 @llvm.umul.fix.i16(i16 %a26, i16 %b26, i32 3)
+  %r27 = call i16 @llvm.umul.fix.i16(i16 %a27, i16 %b27, i32 3)
+  %r28 = call i16 @llvm.umul.fix.i16(i16 %a28, i16 %b28, i32 3)
+  %r29 = call i16 @llvm.umul.fix.i16(i16 %a29, i16 %b29, i32 3)
+  %r30 = call i16 @llvm.umul.fix.i16(i16 %a30, i16 %b30, i32 3)
+  %r31 = call i16 @llvm.umul.fix.i16(i16 %a31, i16 %b31, i32 3)
+  store i16 %r0 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 0 ), align 2
+  store i16 %r1 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 1 ), align 2
+  store i16 %r2 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 2 ), align 2
+  store i16 %r3 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 3 ), align 2
+  store i16 %r4 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 4 ), align 2
+  store i16 %r5 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 5 ), align 2
+  store i16 %r6 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 6 ), align 2
+  store i16 %r7 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 7 ), align 2
+  store i16 %r8 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 8 ), align 2
+  store i16 %r9 , i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 9 ), align 2
+  store i16 %r10, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 10), align 2
+  store i16 %r11, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 11), align 2
+  store i16 %r12, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 12), align 2
+  store i16 %r13, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 13), align 2
+  store i16 %r14, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 14), align 2
+  store i16 %r15, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 15), align 2
+  store i16 %r16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 16), align 2
+  store i16 %r17, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 17), align 2
+  store i16 %r18, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 18), align 2
+  store i16 %r19, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 19), align 2
+  store i16 %r20, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 20), align 2
+  store i16 %r21, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 21), align 2
+  store i16 %r22, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 22), align 2
+  store i16 %r23, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 23), align 2
+  store i16 %r24, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 24), align 2
+  store i16 %r25, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 25), align 2
+  store i16 %r26, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 26), align 2
+  store i16 %r27, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 27), align 2
+  store i16 %r28, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 28), align 2
+  store i16 %r29, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 29), align 2
+  store i16 %r30, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 30), align 2
+  store i16 %r31, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @c16, i32 0, i64 31), align 2
+  ret void
+}
+
+define void @umul_v64i8() {
+; CHECK-LABEL: @umul_v64i8(
+; CHECK-NEXT:    [[A0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 0), align 1
+; CHECK-NEXT:    [[A1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 1), align 1
+; CHECK-NEXT:    [[A2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 2), align 1
+; CHECK-NEXT:    [[A3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 3), align 1
+; CHECK-NEXT:    [[A4:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 4), align 1
+; CHECK-NEXT:    [[A5:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 5), align 1
+; CHECK-NEXT:    [[A6:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 6), align 1
+; CHECK-NEXT:    [[A7:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 7), align 1
+; CHECK-NEXT:    [[A8:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 8), align 1
+; CHECK-NEXT:    [[A9:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 9), align 1
+; CHECK-NEXT:    [[A10:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 10), align 1
+; CHECK-NEXT:    [[A11:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 11), align 1
+; CHECK-NEXT:    [[A12:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 12), align 1
+; CHECK-NEXT:    [[A13:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 13), align 1
+; CHECK-NEXT:    [[A14:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 14), align 1
+; CHECK-NEXT:    [[A15:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 15), align 1
+; CHECK-NEXT:    [[A16:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 16), align 1
+; CHECK-NEXT:    [[A17:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 17), align 1
+; CHECK-NEXT:    [[A18:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 18), align 1
+; CHECK-NEXT:    [[A19:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 19), align 1
+; CHECK-NEXT:    [[A20:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 20), align 1
+; CHECK-NEXT:    [[A21:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 21), align 1
+; CHECK-NEXT:    [[A22:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 22), align 1
+; CHECK-NEXT:    [[A23:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 23), align 1
+; CHECK-NEXT:    [[A24:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 24), align 1
+; CHECK-NEXT:    [[A25:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 25), align 1
+; CHECK-NEXT:    [[A26:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 26), align 1
+; CHECK-NEXT:    [[A27:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 27), align 1
+; CHECK-NEXT:    [[A28:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 28), align 1
+; CHECK-NEXT:    [[A29:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 29), align 1
+; CHECK-NEXT:    [[A30:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 30), align 1
+; CHECK-NEXT:    [[A31:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 31), align 1
+; CHECK-NEXT:    [[A32:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 32), align 1
+; CHECK-NEXT:    [[A33:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 33), align 1
+; CHECK-NEXT:    [[A34:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 34), align 1
+; CHECK-NEXT:    [[A35:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 35), align 1
+; CHECK-NEXT:    [[A36:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 36), align 1
+; CHECK-NEXT:    [[A37:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 37), align 1
+; CHECK-NEXT:    [[A38:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 38), align 1
+; CHECK-NEXT:    [[A39:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 39), align 1
+; CHECK-NEXT:    [[A40:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 40), align 1
+; CHECK-NEXT:    [[A41:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 41), align 1
+; CHECK-NEXT:    [[A42:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 42), align 1
+; CHECK-NEXT:    [[A43:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 43), align 1
+; CHECK-NEXT:    [[A44:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 44), align 1
+; CHECK-NEXT:    [[A45:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 45), align 1
+; CHECK-NEXT:    [[A46:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 46), align 1
+; CHECK-NEXT:    [[A47:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 47), align 1
+; CHECK-NEXT:    [[A48:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 48), align 1
+; CHECK-NEXT:    [[A49:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 49), align 1
+; CHECK-NEXT:    [[A50:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 50), align 1
+; CHECK-NEXT:    [[A51:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 51), align 1
+; CHECK-NEXT:    [[A52:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 52), align 1
+; CHECK-NEXT:    [[A53:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 53), align 1
+; CHECK-NEXT:    [[A54:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 54), align 1
+; CHECK-NEXT:    [[A55:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 55), align 1
+; CHECK-NEXT:    [[A56:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 56), align 1
+; CHECK-NEXT:    [[A57:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 57), align 1
+; CHECK-NEXT:    [[A58:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 58), align 1
+; CHECK-NEXT:    [[A59:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 59), align 1
+; CHECK-NEXT:    [[A60:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 60), align 1
+; CHECK-NEXT:    [[A61:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 61), align 1
+; CHECK-NEXT:    [[A62:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 62), align 1
+; CHECK-NEXT:    [[A63:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 63), align 1
+; CHECK-NEXT:    [[B0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 0), align 1
+; CHECK-NEXT:    [[B1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 1), align 1
+; CHECK-NEXT:    [[B2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 2), align 1
+; CHECK-NEXT:    [[B3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 3), align 1
+; CHECK-NEXT:    [[B4:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 4), align 1
+; CHECK-NEXT:    [[B5:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 5), align 1
+; CHECK-NEXT:    [[B6:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 6), align 1
+; CHECK-NEXT:    [[B7:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 7), align 1
+; CHECK-NEXT:    [[B8:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 8), align 1
+; CHECK-NEXT:    [[B9:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 9), align 1
+; CHECK-NEXT:    [[B10:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 10), align 1
+; CHECK-NEXT:    [[B11:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 11), align 1
+; CHECK-NEXT:    [[B12:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 12), align 1
+; CHECK-NEXT:    [[B13:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 13), align 1
+; CHECK-NEXT:    [[B14:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 14), align 1
+; CHECK-NEXT:    [[B15:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 15), align 1
+; CHECK-NEXT:    [[B16:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 16), align 1
+; CHECK-NEXT:    [[B17:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 17), align 1
+; CHECK-NEXT:    [[B18:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 18), align 1
+; CHECK-NEXT:    [[B19:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 19), align 1
+; CHECK-NEXT:    [[B20:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 20), align 1
+; CHECK-NEXT:    [[B21:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 21), align 1
+; CHECK-NEXT:    [[B22:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 22), align 1
+; CHECK-NEXT:    [[B23:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 23), align 1
+; CHECK-NEXT:    [[B24:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 24), align 1
+; CHECK-NEXT:    [[B25:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 25), align 1
+; CHECK-NEXT:    [[B26:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 26), align 1
+; CHECK-NEXT:    [[B27:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 27), align 1
+; CHECK-NEXT:    [[B28:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 28), align 1
+; CHECK-NEXT:    [[B29:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 29), align 1
+; CHECK-NEXT:    [[B30:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 30), align 1
+; CHECK-NEXT:    [[B31:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 31), align 1
+; CHECK-NEXT:    [[B32:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 32), align 1
+; CHECK-NEXT:    [[B33:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 33), align 1
+; CHECK-NEXT:    [[B34:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 34), align 1
+; CHECK-NEXT:    [[B35:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 35), align 1
+; CHECK-NEXT:    [[B36:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 36), align 1
+; CHECK-NEXT:    [[B37:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 37), align 1
+; CHECK-NEXT:    [[B38:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 38), align 1
+; CHECK-NEXT:    [[B39:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 39), align 1
+; CHECK-NEXT:    [[B40:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 40), align 1
+; CHECK-NEXT:    [[B41:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 41), align 1
+; CHECK-NEXT:    [[B42:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 42), align 1
+; CHECK-NEXT:    [[B43:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 43), align 1
+; CHECK-NEXT:    [[B44:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 44), align 1
+; CHECK-NEXT:    [[B45:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 45), align 1
+; CHECK-NEXT:    [[B46:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 46), align 1
+; CHECK-NEXT:    [[B47:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 47), align 1
+; CHECK-NEXT:    [[B48:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 48), align 1
+; CHECK-NEXT:    [[B49:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 49), align 1
+; CHECK-NEXT:    [[B50:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 50), align 1
+; CHECK-NEXT:    [[B51:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 51), align 1
+; CHECK-NEXT:    [[B52:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 52), align 1
+; CHECK-NEXT:    [[B53:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 53), align 1
+; CHECK-NEXT:    [[B54:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 54), align 1
+; CHECK-NEXT:    [[B55:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 55), align 1
+; CHECK-NEXT:    [[B56:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 56), align 1
+; CHECK-NEXT:    [[B57:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 57), align 1
+; CHECK-NEXT:    [[B58:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 58), align 1
+; CHECK-NEXT:    [[B59:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 59), align 1
+; CHECK-NEXT:    [[B60:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 60), align 1
+; CHECK-NEXT:    [[B61:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 61), align 1
+; CHECK-NEXT:    [[B62:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 62), align 1
+; CHECK-NEXT:    [[B63:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 63), align 1
+; CHECK-NEXT:    [[R0:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A0]], i8 [[B0]], i32 3)
+; CHECK-NEXT:    [[R1:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A1]], i8 [[B1]], i32 3)
+; CHECK-NEXT:    [[R2:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A2]], i8 [[B2]], i32 3)
+; CHECK-NEXT:    [[R3:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A3]], i8 [[B3]], i32 3)
+; CHECK-NEXT:    [[R4:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A4]], i8 [[B4]], i32 3)
+; CHECK-NEXT:    [[R5:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A5]], i8 [[B5]], i32 3)
+; CHECK-NEXT:    [[R6:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A6]], i8 [[B6]], i32 3)
+; CHECK-NEXT:    [[R7:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A7]], i8 [[B7]], i32 3)
+; CHECK-NEXT:    [[R8:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A8]], i8 [[B8]], i32 3)
+; CHECK-NEXT:    [[R9:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A9]], i8 [[B9]], i32 3)
+; CHECK-NEXT:    [[R10:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A10]], i8 [[B10]], i32 3)
+; CHECK-NEXT:    [[R11:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A11]], i8 [[B11]], i32 3)
+; CHECK-NEXT:    [[R12:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A12]], i8 [[B12]], i32 3)
+; CHECK-NEXT:    [[R13:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A13]], i8 [[B13]], i32 3)
+; CHECK-NEXT:    [[R14:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A14]], i8 [[B14]], i32 3)
+; CHECK-NEXT:    [[R15:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A15]], i8 [[B15]], i32 3)
+; CHECK-NEXT:    [[R16:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A16]], i8 [[B16]], i32 3)
+; CHECK-NEXT:    [[R17:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A17]], i8 [[B17]], i32 3)
+; CHECK-NEXT:    [[R18:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A18]], i8 [[B18]], i32 3)
+; CHECK-NEXT:    [[R19:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A19]], i8 [[B19]], i32 3)
+; CHECK-NEXT:    [[R20:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A20]], i8 [[B20]], i32 3)
+; CHECK-NEXT:    [[R21:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A21]], i8 [[B21]], i32 3)
+; CHECK-NEXT:    [[R22:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A22]], i8 [[B22]], i32 3)
+; CHECK-NEXT:    [[R23:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A23]], i8 [[B23]], i32 3)
+; CHECK-NEXT:    [[R24:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A24]], i8 [[B24]], i32 3)
+; CHECK-NEXT:    [[R25:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A25]], i8 [[B25]], i32 3)
+; CHECK-NEXT:    [[R26:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A26]], i8 [[B26]], i32 3)
+; CHECK-NEXT:    [[R27:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A27]], i8 [[B27]], i32 3)
+; CHECK-NEXT:    [[R28:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A28]], i8 [[B28]], i32 3)
+; CHECK-NEXT:    [[R29:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A29]], i8 [[B29]], i32 3)
+; CHECK-NEXT:    [[R30:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A30]], i8 [[B30]], i32 3)
+; CHECK-NEXT:    [[R31:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A31]], i8 [[B31]], i32 3)
+; CHECK-NEXT:    [[R32:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A32]], i8 [[B32]], i32 3)
+; CHECK-NEXT:    [[R33:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A33]], i8 [[B33]], i32 3)
+; CHECK-NEXT:    [[R34:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A34]], i8 [[B34]], i32 3)
+; CHECK-NEXT:    [[R35:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A35]], i8 [[B35]], i32 3)
+; CHECK-NEXT:    [[R36:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A36]], i8 [[B36]], i32 3)
+; CHECK-NEXT:    [[R37:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A37]], i8 [[B37]], i32 3)
+; CHECK-NEXT:    [[R38:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A38]], i8 [[B38]], i32 3)
+; CHECK-NEXT:    [[R39:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A39]], i8 [[B39]], i32 3)
+; CHECK-NEXT:    [[R40:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A40]], i8 [[B40]], i32 3)
+; CHECK-NEXT:    [[R41:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A41]], i8 [[B41]], i32 3)
+; CHECK-NEXT:    [[R42:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A42]], i8 [[B42]], i32 3)
+; CHECK-NEXT:    [[R43:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A43]], i8 [[B43]], i32 3)
+; CHECK-NEXT:    [[R44:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A44]], i8 [[B44]], i32 3)
+; CHECK-NEXT:    [[R45:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A45]], i8 [[B45]], i32 3)
+; CHECK-NEXT:    [[R46:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A46]], i8 [[B46]], i32 3)
+; CHECK-NEXT:    [[R47:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A47]], i8 [[B47]], i32 3)
+; CHECK-NEXT:    [[R48:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A48]], i8 [[B48]], i32 3)
+; CHECK-NEXT:    [[R49:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A49]], i8 [[B49]], i32 3)
+; CHECK-NEXT:    [[R50:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A50]], i8 [[B50]], i32 3)
+; CHECK-NEXT:    [[R51:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A51]], i8 [[B51]], i32 3)
+; CHECK-NEXT:    [[R52:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A52]], i8 [[B52]], i32 3)
+; CHECK-NEXT:    [[R53:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A53]], i8 [[B53]], i32 3)
+; CHECK-NEXT:    [[R54:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A54]], i8 [[B54]], i32 3)
+; CHECK-NEXT:    [[R55:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A55]], i8 [[B55]], i32 3)
+; CHECK-NEXT:    [[R56:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A56]], i8 [[B56]], i32 3)
+; CHECK-NEXT:    [[R57:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A57]], i8 [[B57]], i32 3)
+; CHECK-NEXT:    [[R58:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A58]], i8 [[B58]], i32 3)
+; CHECK-NEXT:    [[R59:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A59]], i8 [[B59]], i32 3)
+; CHECK-NEXT:    [[R60:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A60]], i8 [[B60]], i32 3)
+; CHECK-NEXT:    [[R61:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A61]], i8 [[B61]], i32 3)
+; CHECK-NEXT:    [[R62:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A62]], i8 [[B62]], i32 3)
+; CHECK-NEXT:    [[R63:%.*]] = call i8 @llvm.umul.fix.i8(i8 [[A63]], i8 [[B63]], i32 3)
+; CHECK-NEXT:    store i8 [[R0]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 0), align 1
+; CHECK-NEXT:    store i8 [[R1]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 1), align 1
+; CHECK-NEXT:    store i8 [[R2]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 2), align 1
+; CHECK-NEXT:    store i8 [[R3]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 3), align 1
+; CHECK-NEXT:    store i8 [[R4]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 4), align 1
+; CHECK-NEXT:    store i8 [[R5]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 5), align 1
+; CHECK-NEXT:    store i8 [[R6]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 6), align 1
+; CHECK-NEXT:    store i8 [[R7]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 7), align 1
+; CHECK-NEXT:    store i8 [[R8]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 8), align 1
+; CHECK-NEXT:    store i8 [[R9]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 9), align 1
+; CHECK-NEXT:    store i8 [[R10]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 10), align 1
+; CHECK-NEXT:    store i8 [[R11]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 11), align 1
+; CHECK-NEXT:    store i8 [[R12]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 12), align 1
+; CHECK-NEXT:    store i8 [[R13]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 13), align 1
+; CHECK-NEXT:    store i8 [[R14]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 14), align 1
+; CHECK-NEXT:    store i8 [[R15]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 15), align 1
+; CHECK-NEXT:    store i8 [[R16]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 16), align 1
+; CHECK-NEXT:    store i8 [[R17]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 17), align 1
+; CHECK-NEXT:    store i8 [[R18]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 18), align 1
+; CHECK-NEXT:    store i8 [[R19]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 19), align 1
+; CHECK-NEXT:    store i8 [[R20]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 20), align 1
+; CHECK-NEXT:    store i8 [[R21]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 21), align 1
+; CHECK-NEXT:    store i8 [[R22]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 22), align 1
+; CHECK-NEXT:    store i8 [[R23]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 23), align 1
+; CHECK-NEXT:    store i8 [[R24]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 24), align 1
+; CHECK-NEXT:    store i8 [[R25]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 25), align 1
+; CHECK-NEXT:    store i8 [[R26]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 26), align 1
+; CHECK-NEXT:    store i8 [[R27]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 27), align 1
+; CHECK-NEXT:    store i8 [[R28]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 28), align 1
+; CHECK-NEXT:    store i8 [[R29]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 29), align 1
+; CHECK-NEXT:    store i8 [[R30]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 30), align 1
+; CHECK-NEXT:    store i8 [[R31]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 31), align 1
+; CHECK-NEXT:    store i8 [[R32]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 32), align 1
+; CHECK-NEXT:    store i8 [[R33]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 33), align 1
+; CHECK-NEXT:    store i8 [[R34]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 34), align 1
+; CHECK-NEXT:    store i8 [[R35]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 35), align 1
+; CHECK-NEXT:    store i8 [[R36]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 36), align 1
+; CHECK-NEXT:    store i8 [[R37]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 37), align 1
+; CHECK-NEXT:    store i8 [[R38]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 38), align 1
+; CHECK-NEXT:    store i8 [[R39]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 39), align 1
+; CHECK-NEXT:    store i8 [[R40]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 40), align 1
+; CHECK-NEXT:    store i8 [[R41]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 41), align 1
+; CHECK-NEXT:    store i8 [[R42]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 42), align 1
+; CHECK-NEXT:    store i8 [[R43]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 43), align 1
+; CHECK-NEXT:    store i8 [[R44]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 44), align 1
+; CHECK-NEXT:    store i8 [[R45]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 45), align 1
+; CHECK-NEXT:    store i8 [[R46]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 46), align 1
+; CHECK-NEXT:    store i8 [[R47]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 47), align 1
+; CHECK-NEXT:    store i8 [[R48]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 48), align 1
+; CHECK-NEXT:    store i8 [[R49]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 49), align 1
+; CHECK-NEXT:    store i8 [[R50]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 50), align 1
+; CHECK-NEXT:    store i8 [[R51]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 51), align 1
+; CHECK-NEXT:    store i8 [[R52]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 52), align 1
+; CHECK-NEXT:    store i8 [[R53]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 53), align 1
+; CHECK-NEXT:    store i8 [[R54]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 54), align 1
+; CHECK-NEXT:    store i8 [[R55]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 55), align 1
+; CHECK-NEXT:    store i8 [[R56]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 56), align 1
+; CHECK-NEXT:    store i8 [[R57]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 57), align 1
+; CHECK-NEXT:    store i8 [[R58]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 58), align 1
+; CHECK-NEXT:    store i8 [[R59]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 59), align 1
+; CHECK-NEXT:    store i8 [[R60]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 60), align 1
+; CHECK-NEXT:    store i8 [[R61]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 61), align 1
+; CHECK-NEXT:    store i8 [[R62]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 62), align 1
+; CHECK-NEXT:    store i8 [[R63]], i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 63), align 1
+; CHECK-NEXT:    ret void
+;
+  %a0  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 0 ), align 1
+  %a1  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 1 ), align 1
+  %a2  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 2 ), align 1
+  %a3  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 3 ), align 1
+  %a4  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 4 ), align 1
+  %a5  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 5 ), align 1
+  %a6  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 6 ), align 1
+  %a7  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 7 ), align 1
+  %a8  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 8 ), align 1
+  %a9  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 9 ), align 1
+  %a10 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 10), align 1
+  %a11 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 11), align 1
+  %a12 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 12), align 1
+  %a13 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 13), align 1
+  %a14 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 14), align 1
+  %a15 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 15), align 1
+  %a16 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 16), align 1
+  %a17 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 17), align 1
+  %a18 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 18), align 1
+  %a19 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 19), align 1
+  %a20 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 20), align 1
+  %a21 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 21), align 1
+  %a22 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 22), align 1
+  %a23 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 23), align 1
+  %a24 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 24), align 1
+  %a25 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 25), align 1
+  %a26 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 26), align 1
+  %a27 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 27), align 1
+  %a28 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 28), align 1
+  %a29 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 29), align 1
+  %a30 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 30), align 1
+  %a31 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 31), align 1
+  %a32 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 32), align 1
+  %a33 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 33), align 1
+  %a34 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 34), align 1
+  %a35 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 35), align 1
+  %a36 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 36), align 1
+  %a37 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 37), align 1
+  %a38 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 38), align 1
+  %a39 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 39), align 1
+  %a40 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 40), align 1
+  %a41 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 41), align 1
+  %a42 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 42), align 1
+  %a43 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 43), align 1
+  %a44 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 44), align 1
+  %a45 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 45), align 1
+  %a46 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 46), align 1
+  %a47 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 47), align 1
+  %a48 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 48), align 1
+  %a49 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 49), align 1
+  %a50 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 50), align 1
+  %a51 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 51), align 1
+  %a52 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 52), align 1
+  %a53 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 53), align 1
+  %a54 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 54), align 1
+  %a55 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 55), align 1
+  %a56 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 56), align 1
+  %a57 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 57), align 1
+  %a58 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 58), align 1
+  %a59 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 59), align 1
+  %a60 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 60), align 1
+  %a61 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 61), align 1
+  %a62 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 62), align 1
+  %a63 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @a8, i32 0, i64 63), align 1
+  %b0  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 0 ), align 1
+  %b1  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 1 ), align 1
+  %b2  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 2 ), align 1
+  %b3  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 3 ), align 1
+  %b4  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 4 ), align 1
+  %b5  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 5 ), align 1
+  %b6  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 6 ), align 1
+  %b7  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 7 ), align 1
+  %b8  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 8 ), align 1
+  %b9  = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 9 ), align 1
+  %b10 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 10), align 1
+  %b11 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 11), align 1
+  %b12 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 12), align 1
+  %b13 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 13), align 1
+  %b14 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 14), align 1
+  %b15 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 15), align 1
+  %b16 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 16), align 1
+  %b17 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 17), align 1
+  %b18 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 18), align 1
+  %b19 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 19), align 1
+  %b20 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 20), align 1
+  %b21 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 21), align 1
+  %b22 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 22), align 1
+  %b23 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 23), align 1
+  %b24 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 24), align 1
+  %b25 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 25), align 1
+  %b26 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 26), align 1
+  %b27 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 27), align 1
+  %b28 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 28), align 1
+  %b29 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 29), align 1
+  %b30 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 30), align 1
+  %b31 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 31), align 1
+  %b32 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 32), align 1
+  %b33 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 33), align 1
+  %b34 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 34), align 1
+  %b35 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 35), align 1
+  %b36 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 36), align 1
+  %b37 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 37), align 1
+  %b38 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 38), align 1
+  %b39 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 39), align 1
+  %b40 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 40), align 1
+  %b41 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 41), align 1
+  %b42 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 42), align 1
+  %b43 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 43), align 1
+  %b44 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 44), align 1
+  %b45 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 45), align 1
+  %b46 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 46), align 1
+  %b47 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 47), align 1
+  %b48 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 48), align 1
+  %b49 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 49), align 1
+  %b50 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 50), align 1
+  %b51 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 51), align 1
+  %b52 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 52), align 1
+  %b53 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 53), align 1
+  %b54 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 54), align 1
+  %b55 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 55), align 1
+  %b56 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 56), align 1
+  %b57 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 57), align 1
+  %b58 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 58), align 1
+  %b59 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 59), align 1
+  %b60 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 60), align 1
+  %b61 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 61), align 1
+  %b62 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 62), align 1
+  %b63 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @b8, i32 0, i64 63), align 1
+  %r0  = call i8 @llvm.umul.fix.i8(i8 %a0 , i8 %b0 , i32 3)
+  %r1  = call i8 @llvm.umul.fix.i8(i8 %a1 , i8 %b1 , i32 3)
+  %r2  = call i8 @llvm.umul.fix.i8(i8 %a2 , i8 %b2 , i32 3)
+  %r3  = call i8 @llvm.umul.fix.i8(i8 %a3 , i8 %b3 , i32 3)
+  %r4  = call i8 @llvm.umul.fix.i8(i8 %a4 , i8 %b4 , i32 3)
+  %r5  = call i8 @llvm.umul.fix.i8(i8 %a5 , i8 %b5 , i32 3)
+  %r6  = call i8 @llvm.umul.fix.i8(i8 %a6 , i8 %b6 , i32 3)
+  %r7  = call i8 @llvm.umul.fix.i8(i8 %a7 , i8 %b7 , i32 3)
+  %r8  = call i8 @llvm.umul.fix.i8(i8 %a8 , i8 %b8 , i32 3)
+  %r9  = call i8 @llvm.umul.fix.i8(i8 %a9 , i8 %b9 , i32 3)
+  %r10 = call i8 @llvm.umul.fix.i8(i8 %a10, i8 %b10, i32 3)
+  %r11 = call i8 @llvm.umul.fix.i8(i8 %a11, i8 %b11, i32 3)
+  %r12 = call i8 @llvm.umul.fix.i8(i8 %a12, i8 %b12, i32 3)
+  %r13 = call i8 @llvm.umul.fix.i8(i8 %a13, i8 %b13, i32 3)
+  %r14 = call i8 @llvm.umul.fix.i8(i8 %a14, i8 %b14, i32 3)
+  %r15 = call i8 @llvm.umul.fix.i8(i8 %a15, i8 %b15, i32 3)
+  %r16 = call i8 @llvm.umul.fix.i8(i8 %a16, i8 %b16, i32 3)
+  %r17 = call i8 @llvm.umul.fix.i8(i8 %a17, i8 %b17, i32 3)
+  %r18 = call i8 @llvm.umul.fix.i8(i8 %a18, i8 %b18, i32 3)
+  %r19 = call i8 @llvm.umul.fix.i8(i8 %a19, i8 %b19, i32 3)
+  %r20 = call i8 @llvm.umul.fix.i8(i8 %a20, i8 %b20, i32 3)
+  %r21 = call i8 @llvm.umul.fix.i8(i8 %a21, i8 %b21, i32 3)
+  %r22 = call i8 @llvm.umul.fix.i8(i8 %a22, i8 %b22, i32 3)
+  %r23 = call i8 @llvm.umul.fix.i8(i8 %a23, i8 %b23, i32 3)
+  %r24 = call i8 @llvm.umul.fix.i8(i8 %a24, i8 %b24, i32 3)
+  %r25 = call i8 @llvm.umul.fix.i8(i8 %a25, i8 %b25, i32 3)
+  %r26 = call i8 @llvm.umul.fix.i8(i8 %a26, i8 %b26, i32 3)
+  %r27 = call i8 @llvm.umul.fix.i8(i8 %a27, i8 %b27, i32 3)
+  %r28 = call i8 @llvm.umul.fix.i8(i8 %a28, i8 %b28, i32 3)
+  %r29 = call i8 @llvm.umul.fix.i8(i8 %a29, i8 %b29, i32 3)
+  %r30 = call i8 @llvm.umul.fix.i8(i8 %a30, i8 %b30, i32 3)
+  %r31 = call i8 @llvm.umul.fix.i8(i8 %a31, i8 %b31, i32 3)
+  %r32 = call i8 @llvm.umul.fix.i8(i8 %a32, i8 %b32, i32 3)
+  %r33 = call i8 @llvm.umul.fix.i8(i8 %a33, i8 %b33, i32 3)
+  %r34 = call i8 @llvm.umul.fix.i8(i8 %a34, i8 %b34, i32 3)
+  %r35 = call i8 @llvm.umul.fix.i8(i8 %a35, i8 %b35, i32 3)
+  %r36 = call i8 @llvm.umul.fix.i8(i8 %a36, i8 %b36, i32 3)
+  %r37 = call i8 @llvm.umul.fix.i8(i8 %a37, i8 %b37, i32 3)
+  %r38 = call i8 @llvm.umul.fix.i8(i8 %a38, i8 %b38, i32 3)
+  %r39 = call i8 @llvm.umul.fix.i8(i8 %a39, i8 %b39, i32 3)
+  %r40 = call i8 @llvm.umul.fix.i8(i8 %a40, i8 %b40, i32 3)
+  %r41 = call i8 @llvm.umul.fix.i8(i8 %a41, i8 %b41, i32 3)
+  %r42 = call i8 @llvm.umul.fix.i8(i8 %a42, i8 %b42, i32 3)
+  %r43 = call i8 @llvm.umul.fix.i8(i8 %a43, i8 %b43, i32 3)
+  %r44 = call i8 @llvm.umul.fix.i8(i8 %a44, i8 %b44, i32 3)
+  %r45 = call i8 @llvm.umul.fix.i8(i8 %a45, i8 %b45, i32 3)
+  %r46 = call i8 @llvm.umul.fix.i8(i8 %a46, i8 %b46, i32 3)
+  %r47 = call i8 @llvm.umul.fix.i8(i8 %a47, i8 %b47, i32 3)
+  %r48 = call i8 @llvm.umul.fix.i8(i8 %a48, i8 %b48, i32 3)
+  %r49 = call i8 @llvm.umul.fix.i8(i8 %a49, i8 %b49, i32 3)
+  %r50 = call i8 @llvm.umul.fix.i8(i8 %a50, i8 %b50, i32 3)
+  %r51 = call i8 @llvm.umul.fix.i8(i8 %a51, i8 %b51, i32 3)
+  %r52 = call i8 @llvm.umul.fix.i8(i8 %a52, i8 %b52, i32 3)
+  %r53 = call i8 @llvm.umul.fix.i8(i8 %a53, i8 %b53, i32 3)
+  %r54 = call i8 @llvm.umul.fix.i8(i8 %a54, i8 %b54, i32 3)
+  %r55 = call i8 @llvm.umul.fix.i8(i8 %a55, i8 %b55, i32 3)
+  %r56 = call i8 @llvm.umul.fix.i8(i8 %a56, i8 %b56, i32 3)
+  %r57 = call i8 @llvm.umul.fix.i8(i8 %a57, i8 %b57, i32 3)
+  %r58 = call i8 @llvm.umul.fix.i8(i8 %a58, i8 %b58, i32 3)
+  %r59 = call i8 @llvm.umul.fix.i8(i8 %a59, i8 %b59, i32 3)
+  %r60 = call i8 @llvm.umul.fix.i8(i8 %a60, i8 %b60, i32 3)
+  %r61 = call i8 @llvm.umul.fix.i8(i8 %a61, i8 %b61, i32 3)
+  %r62 = call i8 @llvm.umul.fix.i8(i8 %a62, i8 %b62, i32 3)
+  %r63 = call i8 @llvm.umul.fix.i8(i8 %a63, i8 %b63, i32 3)
+  store i8 %r0 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 0 ), align 1
+  store i8 %r1 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 1 ), align 1
+  store i8 %r2 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 2 ), align 1
+  store i8 %r3 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 3 ), align 1
+  store i8 %r4 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 4 ), align 1
+  store i8 %r5 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 5 ), align 1
+  store i8 %r6 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 6 ), align 1
+  store i8 %r7 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 7 ), align 1
+  store i8 %r8 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 8 ), align 1
+  store i8 %r9 , i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 9 ), align 1
+  store i8 %r10, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 10), align 1
+  store i8 %r11, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 11), align 1
+  store i8 %r12, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 12), align 1
+  store i8 %r13, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 13), align 1
+  store i8 %r14, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 14), align 1
+  store i8 %r15, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 15), align 1
+  store i8 %r16, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 16), align 1
+  store i8 %r17, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 17), align 1
+  store i8 %r18, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 18), align 1
+  store i8 %r19, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 19), align 1
+  store i8 %r20, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 20), align 1
+  store i8 %r21, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 21), align 1
+  store i8 %r22, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 22), align 1
+  store i8 %r23, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 23), align 1
+  store i8 %r24, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 24), align 1
+  store i8 %r25, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 25), align 1
+  store i8 %r26, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 26), align 1
+  store i8 %r27, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 27), align 1
+  store i8 %r28, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 28), align 1
+  store i8 %r29, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 29), align 1
+  store i8 %r30, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 30), align 1
+  store i8 %r31, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 31), align 1
+  store i8 %r32, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 32), align 1
+  store i8 %r33, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 33), align 1
+  store i8 %r34, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 34), align 1
+  store i8 %r35, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 35), align 1
+  store i8 %r36, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 36), align 1
+  store i8 %r37, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 37), align 1
+  store i8 %r38, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 38), align 1
+  store i8 %r39, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 39), align 1
+  store i8 %r40, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 40), align 1
+  store i8 %r41, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 41), align 1
+  store i8 %r42, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 42), align 1
+  store i8 %r43, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 43), align 1
+  store i8 %r44, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 44), align 1
+  store i8 %r45, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 45), align 1
+  store i8 %r46, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 46), align 1
+  store i8 %r47, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 47), align 1
+  store i8 %r48, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 48), align 1
+  store i8 %r49, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 49), align 1
+  store i8 %r50, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 50), align 1
+  store i8 %r51, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 51), align 1
+  store i8 %r52, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 52), align 1
+  store i8 %r53, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 53), align 1
+  store i8 %r54, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 54), align 1
+  store i8 %r55, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 55), align 1
+  store i8 %r56, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 56), align 1
+  store i8 %r57, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 57), align 1
+  store i8 %r58, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 58), align 1
+  store i8 %r59, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 59), align 1
+  store i8 %r60, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 60), align 1
+  store i8 %r61, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 61), align 1
+  store i8 %r62, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 62), align 1
+  store i8 %r63, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @c8, i32 0, i64 63), align 1
+  ret void
+}




More information about the llvm-commits mailing list