1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=bdver1 | FileCheck %s 3 4; clang -Oz -c test1.cpp -emit-llvm -S -o 5; Verify that we generate shld insruction when we are optimizing for size, 6; even for X86_64 processors that are known to have poor latency double 7; precision shift instructions. 8; uint64_t lshift10(uint64_t a, uint64_t b) 9; { 10; return (a << 10) | (b >> 54); 11; } 12 13; Function Attrs: minsize nounwind readnone uwtable 14define i64 @_Z8lshift10mm(i64 %a, i64 %b) #0 { 15; CHECK-LABEL: _Z8lshift10mm: 16; CHECK: # %bb.0: # %entry 17; CHECK-NEXT: shldq $10, %rsi, %rdi 18; CHECK-NEXT: movq %rdi, %rax 19; CHECK-NEXT: retq 20entry: 21 %shl = shl i64 %a, 10 22 %shr = lshr i64 %b, 54 23 %or = or i64 %shr, %shl 24 ret i64 %or 25} 26 27attributes #0 = { minsize nounwind readnone uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } 28 29 30; clang -Os -c test2.cpp -emit-llvm -S 31; Verify that we generate shld insruction when we are optimizing for size, 32; even for X86_64 processors that are known to have poor latency double 33; precision shift instructions. 34; uint64_t lshift11(uint64_t a, uint64_t b) 35; { 36; return (a << 11) | (b >> 53); 37; } 38 39; Function Attrs: nounwind optsize readnone uwtable 40define i64 @_Z8lshift11mm(i64 %a, i64 %b) #1 { 41; CHECK-LABEL: _Z8lshift11mm: 42; CHECK: # %bb.0: # %entry 43; CHECK-NEXT: shldq $11, %rsi, %rdi 44; CHECK-NEXT: movq %rdi, %rax 45; CHECK-NEXT: retq 46entry: 47 %shl = shl i64 %a, 11 48 %shr = lshr i64 %b, 53 49 %or = or i64 %shr, %shl 50 ret i64 %or 51} 52 53attributes #1 = { nounwind optsize readnone uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } 54 55; clang -O2 -c test2.cpp -emit-llvm -S 56; Verify that we do not generate shld insruction when we are not optimizing 57; for size for X86_64 processors that are known to have poor latency double 58; precision shift instructions. 59; uint64_t lshift12(uint64_t a, uint64_t b) 60; { 61; return (a << 12) | (b >> 52); 62; } 63 64; Function Attrs: nounwind optsize readnone uwtable 65define i64 @_Z8lshift12mm(i64 %a, i64 %b) #2 { 66; CHECK-LABEL: _Z8lshift12mm: 67; CHECK: # %bb.0: # %entry 68; CHECK-NEXT: shlq $12, %rdi 69; CHECK-NEXT: shrq $52, %rsi 70; CHECK-NEXT: leaq (%rsi,%rdi), %rax 71; CHECK-NEXT: retq 72entry: 73 %shl = shl i64 %a, 12 74 %shr = lshr i64 %b, 52 75 %or = or i64 %shr, %shl 76 ret i64 %or 77} 78 79attributes #2= { nounwind readnone uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } 80 81