[llvm] c81d5b1 - [X86] Add scalar test coverage for ISD::AVG nodes on 32 and 64-bit targets

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Fri Jun 14 06:28:56 PDT 2024


Author: Simon Pilgrim
Date: 2024-06-14T14:28:29+01:00
New Revision: c81d5b11cf7caf82749638752d819a061fdf4d9e

URL: https://github.com/llvm/llvm-project/commit/c81d5b11cf7caf82749638752d819a061fdf4d9e
DIFF: https://github.com/llvm/llvm-project/commit/c81d5b11cf7caf82749638752d819a061fdf4d9e.diff

LOG: [X86] Add scalar test coverage for ISD::AVG nodes on 32 and 64-bit targets

Added: 
    llvm/test/CodeGen/X86/avgceils-scalar.ll
    llvm/test/CodeGen/X86/avgceilu-scalar.ll
    llvm/test/CodeGen/X86/avgfloors-scalar.ll
    llvm/test/CodeGen/X86/avgflooru-scalar.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/X86/avgceils-scalar.ll b/llvm/test/CodeGen/X86/avgceils-scalar.ll
new file mode 100644
index 0000000000000..86de35d36f076
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avgceils-scalar.ll
@@ -0,0 +1,267 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-- | FileCheck %s --check-prefixes=X86
+; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s --check-prefixes=X64
+
+;
+; fixed avg(x,y) = sub(or(x,y),ashr(xor(x,y),1))
+;
+; ext avg(x,y) = trunc(ashr(add(sext(x),sext(y),1),1))
+;
+
+define i8 @test_fixed_i8(i8 %a0, i8 %a1) nounwind {
+; X86-LABEL: test_fixed_i8:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orb %cl, %al
+; X86-NEXT:    xorb %cl, %dl
+; X86-NEXT:    sarb %dl
+; X86-NEXT:    subb %dl, %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i8:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orb %sil, %al
+; X64-NEXT:    xorb %sil, %dil
+; X64-NEXT:    sarb %dil
+; X64-NEXT:    subb %dil, %al
+; X64-NEXT:    retq
+  %or = or i8 %a0, %a1
+  %xor = xor i8 %a0, %a1
+  %shift = ashr i8 %xor, 1
+  %res = sub i8 %or, %shift
+  ret i8 %res
+}
+
+define i8 @test_ext_i8(i8 %a0, i8 %a1) nounwind {
+; X86-LABEL: test_ext_i8:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orb %cl, %al
+; X86-NEXT:    xorb %cl, %dl
+; X86-NEXT:    sarb %dl
+; X86-NEXT:    subb %dl, %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i8:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orb %sil, %al
+; X64-NEXT:    xorb %sil, %dil
+; X64-NEXT:    sarb %dil
+; X64-NEXT:    subb %dil, %al
+; X64-NEXT:    retq
+  %x0 = sext i8 %a0 to i16
+  %x1 = sext i8 %a1 to i16
+  %sum = add i16 %x0, %x1
+  %sum1 = add i16 %sum, 1
+  %shift = ashr i16 %sum1, 1
+  %res = trunc i16 %shift to i8
+  ret i8 %res
+}
+
+define i16 @test_fixed_i16(i16 %a0, i16 %a1) nounwind {
+; X86-LABEL: test_fixed_i16:
+; X86:       # %bb.0:
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    xorl %ecx, %edx
+; X86-NEXT:    movswl %dx, %ecx
+; X86-NEXT:    sarl %ecx
+; X86-NEXT:    subl %ecx, %eax
+; X86-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i16:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    movswl %di, %ecx
+; X64-NEXT:    sarl %ecx
+; X64-NEXT:    subl %ecx, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %or = or i16 %a0, %a1
+  %xor = xor i16 %a0, %a1
+  %shift = ashr i16 %xor, 1
+  %res = sub i16 %or, %shift
+  ret i16 %res
+}
+
+define i16 @test_ext_i16(i16 %a0, i16 %a1) nounwind {
+; X86-LABEL: test_ext_i16:
+; X86:       # %bb.0:
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    xorl %ecx, %edx
+; X86-NEXT:    movswl %dx, %ecx
+; X86-NEXT:    sarl %ecx
+; X86-NEXT:    subl %ecx, %eax
+; X86-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i16:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    movswl %di, %ecx
+; X64-NEXT:    sarl %ecx
+; X64-NEXT:    subl %ecx, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %x0 = sext i16 %a0 to i32
+  %x1 = sext i16 %a1 to i32
+  %sum = add i32 %x0, %x1
+  %sum1 = add i32 %sum, 1
+  %shift = ashr i32 %sum1, 1
+  %res = trunc i32 %shift to i16
+  ret i16 %res
+}
+
+define i32 @test_fixed_i32(i32 %a0, i32 %a1) nounwind {
+; X86-LABEL: test_fixed_i32:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    xorl %ecx, %edx
+; X86-NEXT:    sarl %edx
+; X86-NEXT:    subl %edx, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i32:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    sarl %edi
+; X64-NEXT:    subl %edi, %eax
+; X64-NEXT:    retq
+  %or = or i32 %a0, %a1
+  %xor = xor i32 %a1, %a0
+  %shift = ashr i32 %xor, 1
+  %res = sub i32 %or, %shift
+  ret i32 %res
+}
+
+define i32 @test_ext_i32(i32 %a0, i32 %a1) nounwind {
+; X86-LABEL: test_ext_i32:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    xorl %ecx, %edx
+; X86-NEXT:    sarl %edx
+; X86-NEXT:    subl %edx, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i32:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    sarl %edi
+; X64-NEXT:    subl %edi, %eax
+; X64-NEXT:    retq
+  %x0 = sext i32 %a0 to i64
+  %x1 = sext i32 %a1 to i64
+  %sum = add i64 %x0, %x1
+  %sum1 = add i64 %sum, 1
+  %shift = ashr i64 %sum1, 1
+  %res = trunc i64 %shift to i32
+  ret i32 %res
+}
+
+define i64 @test_fixed_i64(i64 %a0, i64 %a1) nounwind {
+; X86-LABEL: test_fixed_i64:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl %ecx, %edi
+; X86-NEXT:    movl %edx, %ebx
+; X86-NEXT:    xorl %esi, %ebx
+; X86-NEXT:    shrdl $1, %ebx, %edi
+; X86-NEXT:    orl %esi, %edx
+; X86-NEXT:    sarl %ebx
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    subl %edi, %eax
+; X86-NEXT:    sbbl %ebx, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i64:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    orq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    sarq %rdi
+; X64-NEXT:    subq %rdi, %rax
+; X64-NEXT:    retq
+  %or = or i64 %a0, %a1
+  %xor = xor i64 %a1, %a0
+  %shift = ashr i64 %xor, 1
+  %res = sub i64 %or, %shift
+  ret i64 %res
+}
+
+define i64 @test_ext_i64(i64 %a0, i64 %a1) nounwind {
+; X86-LABEL: test_ext_i64:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl %ecx, %edi
+; X86-NEXT:    movl %edx, %ebx
+; X86-NEXT:    xorl %esi, %ebx
+; X86-NEXT:    shrdl $1, %ebx, %edi
+; X86-NEXT:    orl %esi, %edx
+; X86-NEXT:    sarl %ebx
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    subl %edi, %eax
+; X86-NEXT:    sbbl %ebx, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i64:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    orq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    sarq %rdi
+; X64-NEXT:    subq %rdi, %rax
+; X64-NEXT:    retq
+  %x0 = sext i64 %a0 to i128
+  %x1 = sext i64 %a1 to i128
+  %sum = add i128 %x0, %x1
+  %sum1 = add i128 %sum, 1
+  %shift = ashr i128 %sum1, 1
+  %res = trunc i128 %shift to i64
+  ret i64 %res
+}

diff  --git a/llvm/test/CodeGen/X86/avgceilu-scalar.ll b/llvm/test/CodeGen/X86/avgceilu-scalar.ll
new file mode 100644
index 0000000000000..014c984528141
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avgceilu-scalar.ll
@@ -0,0 +1,267 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-- | FileCheck %s --check-prefixes=X86
+; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s --check-prefixes=X64
+
+;
+; fixed avg(x,y) = sub(or(x,y),lshr(xor(x,y),1))
+;
+; ext avg(x,y) = trunc(lshr(add(zext(x),zext(y),1),1))
+;
+
+define i8 @test_fixed_i8(i8 %a0, i8 %a1) nounwind {
+; X86-LABEL: test_fixed_i8:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orb %cl, %al
+; X86-NEXT:    xorb %cl, %dl
+; X86-NEXT:    shrb %dl
+; X86-NEXT:    subb %dl, %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i8:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orb %sil, %al
+; X64-NEXT:    xorb %sil, %dil
+; X64-NEXT:    shrb %dil
+; X64-NEXT:    subb %dil, %al
+; X64-NEXT:    retq
+  %or = or i8 %a0, %a1
+  %xor = xor i8 %a0, %a1
+  %shift = lshr i8 %xor, 1
+  %res = sub i8 %or, %shift
+  ret i8 %res
+}
+
+define i8 @test_ext_i8(i8 %a0, i8 %a1) nounwind {
+; X86-LABEL: test_ext_i8:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orb %cl, %al
+; X86-NEXT:    xorb %cl, %dl
+; X86-NEXT:    shrb %dl
+; X86-NEXT:    subb %dl, %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i8:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orb %sil, %al
+; X64-NEXT:    xorb %sil, %dil
+; X64-NEXT:    shrb %dil
+; X64-NEXT:    subb %dil, %al
+; X64-NEXT:    retq
+  %x0 = zext i8 %a0 to i16
+  %x1 = zext i8 %a1 to i16
+  %sum = add i16 %x0, %x1
+  %sum1 = add i16 %sum, 1
+  %shift = lshr i16 %sum1, 1
+  %res = trunc i16 %shift to i8
+  ret i8 %res
+}
+
+define i16 @test_fixed_i16(i16 %a0, i16 %a1) nounwind {
+; X86-LABEL: test_fixed_i16:
+; X86:       # %bb.0:
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    xorl %ecx, %edx
+; X86-NEXT:    movzwl %dx, %ecx
+; X86-NEXT:    shrl %ecx
+; X86-NEXT:    subl %ecx, %eax
+; X86-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i16:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    movzwl %di, %ecx
+; X64-NEXT:    shrl %ecx
+; X64-NEXT:    subl %ecx, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %or = or i16 %a0, %a1
+  %xor = xor i16 %a0, %a1
+  %shift = lshr i16 %xor, 1
+  %res = sub i16 %or, %shift
+  ret i16 %res
+}
+
+define i16 @test_ext_i16(i16 %a0, i16 %a1) nounwind {
+; X86-LABEL: test_ext_i16:
+; X86:       # %bb.0:
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    xorl %ecx, %edx
+; X86-NEXT:    movzwl %dx, %ecx
+; X86-NEXT:    shrl %ecx
+; X86-NEXT:    subl %ecx, %eax
+; X86-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i16:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    movzwl %di, %ecx
+; X64-NEXT:    shrl %ecx
+; X64-NEXT:    subl %ecx, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %x0 = zext i16 %a0 to i32
+  %x1 = zext i16 %a1 to i32
+  %sum = add i32 %x0, %x1
+  %sum1 = add i32 %sum, 1
+  %shift = lshr i32 %sum1, 1
+  %res = trunc i32 %shift to i16
+  ret i16 %res
+}
+
+define i32 @test_fixed_i32(i32 %a0, i32 %a1) nounwind {
+; X86-LABEL: test_fixed_i32:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    xorl %ecx, %edx
+; X86-NEXT:    shrl %edx
+; X86-NEXT:    subl %edx, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i32:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    shrl %edi
+; X64-NEXT:    subl %edi, %eax
+; X64-NEXT:    retq
+  %or = or i32 %a0, %a1
+  %xor = xor i32 %a1, %a0
+  %shift = lshr i32 %xor, 1
+  %res = sub i32 %or, %shift
+  ret i32 %res
+}
+
+define i32 @test_ext_i32(i32 %a0, i32 %a1) nounwind {
+; X86-LABEL: test_ext_i32:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    xorl %ecx, %edx
+; X86-NEXT:    shrl %edx
+; X86-NEXT:    subl %edx, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i32:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    orl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    shrl %edi
+; X64-NEXT:    subl %edi, %eax
+; X64-NEXT:    retq
+  %x0 = zext i32 %a0 to i64
+  %x1 = zext i32 %a1 to i64
+  %sum = add i64 %x0, %x1
+  %sum1 = add i64 %sum, 1
+  %shift = lshr i64 %sum1, 1
+  %res = trunc i64 %shift to i32
+  ret i32 %res
+}
+
+define i64 @test_fixed_i64(i64 %a0, i64 %a1) nounwind {
+; X86-LABEL: test_fixed_i64:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl %ecx, %edi
+; X86-NEXT:    movl %edx, %ebx
+; X86-NEXT:    xorl %esi, %ebx
+; X86-NEXT:    shrdl $1, %ebx, %edi
+; X86-NEXT:    orl %esi, %edx
+; X86-NEXT:    shrl %ebx
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    subl %edi, %eax
+; X86-NEXT:    sbbl %ebx, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i64:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    orq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    shrq %rdi
+; X64-NEXT:    subq %rdi, %rax
+; X64-NEXT:    retq
+  %or = or i64 %a0, %a1
+  %xor = xor i64 %a1, %a0
+  %shift = lshr i64 %xor, 1
+  %res = sub i64 %or, %shift
+  ret i64 %res
+}
+
+define i64 @test_ext_i64(i64 %a0, i64 %a1) nounwind {
+; X86-LABEL: test_ext_i64:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl %ecx, %edi
+; X86-NEXT:    movl %edx, %ebx
+; X86-NEXT:    xorl %esi, %ebx
+; X86-NEXT:    shrdl $1, %ebx, %edi
+; X86-NEXT:    orl %esi, %edx
+; X86-NEXT:    shrl %ebx
+; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    subl %edi, %eax
+; X86-NEXT:    sbbl %ebx, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i64:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    orq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    shrq %rdi
+; X64-NEXT:    subq %rdi, %rax
+; X64-NEXT:    retq
+  %x0 = zext i64 %a0 to i128
+  %x1 = zext i64 %a1 to i128
+  %sum = add i128 %x0, %x1
+  %sum1 = add i128 %sum, 1
+  %shift = lshr i128 %sum1, 1
+  %res = trunc i128 %shift to i64
+  ret i64 %res
+}

diff  --git a/llvm/test/CodeGen/X86/avgfloors-scalar.ll b/llvm/test/CodeGen/X86/avgfloors-scalar.ll
new file mode 100644
index 0000000000000..4c591d4079040
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avgfloors-scalar.ll
@@ -0,0 +1,263 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-- | FileCheck %s --check-prefixes=X86
+; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s --check-prefixes=X64
+
+;
+; fixed avg(x,y) = add(and(x,y),ashr(xor(x,y),1))
+;
+; ext avg(x,y) = trunc(ashr(add(sext(x),sext(y)),1))
+;
+
+define i8 @test_fixed_i8(i8 %a0, i8 %a1) nounwind {
+; X86-LABEL: test_fixed_i8:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andb %cl, %dl
+; X86-NEXT:    xorb %cl, %al
+; X86-NEXT:    sarb %al
+; X86-NEXT:    addb %dl, %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i8:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andb %sil, %al
+; X64-NEXT:    xorb %sil, %dil
+; X64-NEXT:    sarb %dil
+; X64-NEXT:    addb %dil, %al
+; X64-NEXT:    retq
+  %and = and i8 %a0, %a1
+  %xor = xor i8 %a0, %a1
+  %shift = ashr i8 %xor, 1
+  %res = add i8 %and, %shift
+  ret i8 %res
+}
+
+define i8 @test_ext_i8(i8 %a0, i8 %a1) nounwind {
+; X86-LABEL: test_ext_i8:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andb %cl, %dl
+; X86-NEXT:    xorb %cl, %al
+; X86-NEXT:    sarb %al
+; X86-NEXT:    addb %dl, %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i8:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andb %sil, %al
+; X64-NEXT:    xorb %sil, %dil
+; X64-NEXT:    sarb %dil
+; X64-NEXT:    addb %dil, %al
+; X64-NEXT:    retq
+  %x0 = sext i8 %a0 to i16
+  %x1 = sext i8 %a1 to i16
+  %sum = add i16 %x0, %x1
+  %shift = ashr i16 %sum, 1
+  %res = trunc i16 %shift to i8
+  ret i8 %res
+}
+
+define i16 @test_fixed_i16(i16 %a0, i16 %a1) nounwind {
+; X86-LABEL: test_fixed_i16:
+; X86:       # %bb.0:
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    andl %eax, %edx
+; X86-NEXT:    xorl %eax, %ecx
+; X86-NEXT:    movswl %cx, %eax
+; X86-NEXT:    sarl %eax
+; X86-NEXT:    addl %edx, %eax
+; X86-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i16:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %ecx
+; X64-NEXT:    andl %esi, %ecx
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    movswl %di, %eax
+; X64-NEXT:    sarl %eax
+; X64-NEXT:    addl %ecx, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %and = and i16 %a0, %a1
+  %xor = xor i16 %a0, %a1
+  %shift = ashr i16 %xor, 1
+  %res = add i16 %and, %shift
+  ret i16 %res
+}
+
+define i16 @test_ext_i16(i16 %a0, i16 %a1) nounwind {
+; X86-LABEL: test_ext_i16:
+; X86:       # %bb.0:
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    andl %eax, %edx
+; X86-NEXT:    xorl %eax, %ecx
+; X86-NEXT:    movswl %cx, %eax
+; X86-NEXT:    sarl %eax
+; X86-NEXT:    addl %edx, %eax
+; X86-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i16:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %ecx
+; X64-NEXT:    andl %esi, %ecx
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    movswl %di, %eax
+; X64-NEXT:    sarl %eax
+; X64-NEXT:    addl %ecx, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %x0 = sext i16 %a0 to i32
+  %x1 = sext i16 %a1 to i32
+  %sum = add i32 %x0, %x1
+  %shift = ashr i32 %sum, 1
+  %res = trunc i32 %shift to i16
+  ret i16 %res
+}
+
+define i32 @test_fixed_i32(i32 %a0, i32 %a1) nounwind {
+; X86-LABEL: test_fixed_i32:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl %ecx, %edx
+; X86-NEXT:    xorl %ecx, %eax
+; X86-NEXT:    sarl %eax
+; X86-NEXT:    addl %edx, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i32:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    sarl %edi
+; X64-NEXT:    addl %edi, %eax
+; X64-NEXT:    retq
+  %and = and i32 %a0, %a1
+  %xor = xor i32 %a1, %a0
+  %shift = ashr i32 %xor, 1
+  %res = add i32 %and, %shift
+  ret i32 %res
+}
+
+define i32 @test_ext_i32(i32 %a0, i32 %a1) nounwind {
+; X86-LABEL: test_ext_i32:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl %ecx, %edx
+; X86-NEXT:    xorl %ecx, %eax
+; X86-NEXT:    sarl %eax
+; X86-NEXT:    addl %edx, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i32:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    sarl %edi
+; X64-NEXT:    addl %edi, %eax
+; X64-NEXT:    retq
+  %x0 = sext i32 %a0 to i64
+  %x1 = sext i32 %a1 to i64
+  %sum = add i64 %x0, %x1
+  %shift = ashr i64 %sum, 1
+  %res = trunc i64 %shift to i32
+  ret i32 %res
+}
+
+define i64 @test_fixed_i64(i64 %a0, i64 %a1) nounwind {
+; X86-LABEL: test_fixed_i64:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl %esi, %ebx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    xorl %edi, %edx
+; X86-NEXT:    shrdl $1, %edx, %ebx
+; X86-NEXT:    andl %edi, %ecx
+; X86-NEXT:    sarl %edx
+; X86-NEXT:    andl %esi, %eax
+; X86-NEXT:    addl %ebx, %eax
+; X86-NEXT:    adcl %ecx, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i64:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    andq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    sarq %rdi
+; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    retq
+  %and = and i64 %a0, %a1
+  %xor = xor i64 %a1, %a0
+  %shift = ashr i64 %xor, 1
+  %res = add i64 %and, %shift
+  ret i64 %res
+}
+
+define i64 @test_ext_i64(i64 %a0, i64 %a1) nounwind {
+; X86-LABEL: test_ext_i64:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl %esi, %ebx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    xorl %edi, %edx
+; X86-NEXT:    shrdl $1, %edx, %ebx
+; X86-NEXT:    andl %edi, %ecx
+; X86-NEXT:    sarl %edx
+; X86-NEXT:    andl %esi, %eax
+; X86-NEXT:    addl %ebx, %eax
+; X86-NEXT:    adcl %ecx, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i64:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    andq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    sarq %rdi
+; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    retq
+  %x0 = sext i64 %a0 to i128
+  %x1 = sext i64 %a1 to i128
+  %sum = add i128 %x0, %x1
+  %shift = ashr i128 %sum, 1
+  %res = trunc i128 %shift to i64
+  ret i64 %res
+}

diff  --git a/llvm/test/CodeGen/X86/avgflooru-scalar.ll b/llvm/test/CodeGen/X86/avgflooru-scalar.ll
new file mode 100644
index 0000000000000..592e5e15b936a
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avgflooru-scalar.ll
@@ -0,0 +1,263 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-- | FileCheck %s --check-prefixes=X86
+; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s --check-prefixes=X64
+
+;
+; fixed avg(x,y) = add(and(x,y),lshr(xor(x,y),1))
+;
+; ext avg(x,y) = trunc(lshr(add(zext(x),zext(y)),1))
+;
+
+define i8 @test_fixed_i8(i8 %a0, i8 %a1) nounwind {
+; X86-LABEL: test_fixed_i8:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andb %cl, %dl
+; X86-NEXT:    xorb %cl, %al
+; X86-NEXT:    shrb %al
+; X86-NEXT:    addb %dl, %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i8:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andb %sil, %al
+; X64-NEXT:    xorb %sil, %dil
+; X64-NEXT:    shrb %dil
+; X64-NEXT:    addb %dil, %al
+; X64-NEXT:    retq
+  %and = and i8 %a0, %a1
+  %xor = xor i8 %a0, %a1
+  %shift = lshr i8 %xor, 1
+  %res = add i8 %and, %shift
+  ret i8 %res
+}
+
+define i8 @test_ext_i8(i8 %a0, i8 %a1) nounwind {
+; X86-LABEL: test_ext_i8:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andb %cl, %dl
+; X86-NEXT:    xorb %cl, %al
+; X86-NEXT:    shrb %al
+; X86-NEXT:    addb %dl, %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i8:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andb %sil, %al
+; X64-NEXT:    xorb %sil, %dil
+; X64-NEXT:    shrb %dil
+; X64-NEXT:    addb %dil, %al
+; X64-NEXT:    retq
+  %x0 = zext i8 %a0 to i16
+  %x1 = zext i8 %a1 to i16
+  %sum = add i16 %x0, %x1
+  %shift = lshr i16 %sum, 1
+  %res = trunc i16 %shift to i8
+  ret i8 %res
+}
+
+define i16 @test_fixed_i16(i16 %a0, i16 %a1) nounwind {
+; X86-LABEL: test_fixed_i16:
+; X86:       # %bb.0:
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    andl %eax, %edx
+; X86-NEXT:    xorl %eax, %ecx
+; X86-NEXT:    movzwl %cx, %eax
+; X86-NEXT:    shrl %eax
+; X86-NEXT:    addl %edx, %eax
+; X86-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i16:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %ecx
+; X64-NEXT:    andl %esi, %ecx
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    movzwl %di, %eax
+; X64-NEXT:    shrl %eax
+; X64-NEXT:    addl %ecx, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %and = and i16 %a0, %a1
+  %xor = xor i16 %a0, %a1
+  %shift = lshr i16 %xor, 1
+  %res = add i16 %and, %shift
+  ret i16 %res
+}
+
+define i16 @test_ext_i16(i16 %a0, i16 %a1) nounwind {
+; X86-LABEL: test_ext_i16:
+; X86:       # %bb.0:
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    andl %eax, %edx
+; X86-NEXT:    xorl %eax, %ecx
+; X86-NEXT:    movzwl %cx, %eax
+; X86-NEXT:    shrl %eax
+; X86-NEXT:    addl %edx, %eax
+; X86-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i16:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %ecx
+; X64-NEXT:    andl %esi, %ecx
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    movzwl %di, %eax
+; X64-NEXT:    shrl %eax
+; X64-NEXT:    addl %ecx, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %x0 = zext i16 %a0 to i32
+  %x1 = zext i16 %a1 to i32
+  %sum = add i32 %x0, %x1
+  %shift = lshr i32 %sum, 1
+  %res = trunc i32 %shift to i16
+  ret i16 %res
+}
+
+define i32 @test_fixed_i32(i32 %a0, i32 %a1) nounwind {
+; X86-LABEL: test_fixed_i32:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl %ecx, %edx
+; X86-NEXT:    xorl %ecx, %eax
+; X86-NEXT:    shrl %eax
+; X86-NEXT:    addl %edx, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i32:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    shrl %edi
+; X64-NEXT:    addl %edi, %eax
+; X64-NEXT:    retq
+  %and = and i32 %a0, %a1
+  %xor = xor i32 %a1, %a0
+  %shift = lshr i32 %xor, 1
+  %res = add i32 %and, %shift
+  ret i32 %res
+}
+
+define i32 @test_ext_i32(i32 %a0, i32 %a1) nounwind {
+; X86-LABEL: test_ext_i32:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl %ecx, %edx
+; X86-NEXT:    xorl %ecx, %eax
+; X86-NEXT:    shrl %eax
+; X86-NEXT:    addl %edx, %eax
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i32:
+; X64:       # %bb.0:
+; X64-NEXT:    movl %edi, %eax
+; X64-NEXT:    andl %esi, %eax
+; X64-NEXT:    xorl %esi, %edi
+; X64-NEXT:    shrl %edi
+; X64-NEXT:    addl %edi, %eax
+; X64-NEXT:    retq
+  %x0 = zext i32 %a0 to i64
+  %x1 = zext i32 %a1 to i64
+  %sum = add i64 %x0, %x1
+  %shift = lshr i64 %sum, 1
+  %res = trunc i64 %shift to i32
+  ret i32 %res
+}
+
+define i64 @test_fixed_i64(i64 %a0, i64 %a1) nounwind {
+; X86-LABEL: test_fixed_i64:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl %esi, %ebx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    xorl %edi, %edx
+; X86-NEXT:    shrdl $1, %edx, %ebx
+; X86-NEXT:    andl %edi, %ecx
+; X86-NEXT:    shrl %edx
+; X86-NEXT:    andl %esi, %eax
+; X86-NEXT:    addl %ebx, %eax
+; X86-NEXT:    adcl %ecx, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_fixed_i64:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    andq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    shrq %rdi
+; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    retq
+  %and = and i64 %a0, %a1
+  %xor = xor i64 %a1, %a0
+  %shift = lshr i64 %xor, 1
+  %res = add i64 %and, %shift
+  ret i64 %res
+}
+
+define i64 @test_ext_i64(i64 %a0, i64 %a1) nounwind {
+; X86-LABEL: test_ext_i64:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl %esi, %ebx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    xorl %edi, %edx
+; X86-NEXT:    shrdl $1, %edx, %ebx
+; X86-NEXT:    andl %edi, %ecx
+; X86-NEXT:    shrl %edx
+; X86-NEXT:    andl %esi, %eax
+; X86-NEXT:    addl %ebx, %eax
+; X86-NEXT:    adcl %ecx, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_ext_i64:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    andq %rsi, %rax
+; X64-NEXT:    xorq %rsi, %rdi
+; X64-NEXT:    shrq %rdi
+; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    retq
+  %x0 = zext i64 %a0 to i128
+  %x1 = zext i64 %a1 to i128
+  %sum = add i128 %x0, %x1
+  %shift = lshr i128 %sum, 1
+  %res = trunc i128 %shift to i64
+  ret i64 %res
+}


        


More information about the llvm-commits mailing list