From a0686462c34f247ab3b5f07f3dc414c9a331cb3d Mon Sep 17 00:00:00 2001
From: Sanjay Patel <spatel@rotateright.com>
Date: Wed, 28 Jul 2021 14:53:47 -0400
Subject: [PATCH] [AArch64][x86] add tests for add-with-overflow folds; NFC

There's a generic combine for these, but no test coverage.
It's not clear if this is actually a good fold.
The combine was added with D58874, but it has a bug that
can cause crashing ( https://llvm.org/PR51238 ).

(cherry picked from commit e427077ec10ea18ac21f5065342183481d87783a)
---
 test/CodeGen/AArch64/addsub.ll  | 44 ++++++++++++++++++++
 test/CodeGen/X86/combine-add.ll | 72 ++++++++++++++++++++++++---------
 2 files changed, 96 insertions(+), 20 deletions(-)

diff --git a/test/CodeGen/AArch64/addsub.ll b/test/CodeGen/AArch64/addsub.ll
index b62b2fe3cb1..880325c3584 100644
--- a/test/CodeGen/AArch64/addsub.ll
+++ b/test/CodeGen/AArch64/addsub.ll
@@ -223,4 +223,48 @@ test6:
 ret:
   ret void
 }
+
+declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
+
+define i1 @sadd_add(i32 %a, i32 %b, i32* %p) {
+; CHECK-LABEL: sadd_add:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mvn w8, w0
+; CHECK-NEXT:    cmn w8, w1
+; CHECK-NEXT:    cset w8, vs
+; CHECK-NEXT:    sub w9, w1, w0
+; CHECK-NEXT:    mov w0, w8
+; CHECK-NEXT:    str w9, [x2]
+; CHECK-NEXT:    ret
+  %nota = xor i32 %a, -1
+  %a0 = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %nota, i32 %b)
+  %e0 = extractvalue {i32, i1} %a0, 0
+  %e1 = extractvalue {i32, i1} %a0, 1
+  %res = add i32 %e0, 1
+  store i32 %res, i32* %p
+  ret i1 %e1
+}
+
+declare {i8, i1} @llvm.uadd.with.overflow.i8(i8 %a, i8 %b)
+
+define i1 @uadd_add(i8 %a, i8 %b, i8* %p) {
+; CHECK-LABEL: uadd_add:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mvn w8, w0
+; CHECK-NEXT:    and w8, w8, #0xff
+; CHECK-NEXT:    add w8, w8, w1, uxtb
+; CHECK-NEXT:    lsr w8, w8, #8
+; CHECK-NEXT:    sub w9, w1, w0
+; CHECK-NEXT:    mov w0, w8
+; CHECK-NEXT:    strb w9, [x2]
+; CHECK-NEXT:    ret
+  %nota = xor i8 %a, -1
+  %a0 = call {i8, i1} @llvm.uadd.with.overflow.i8(i8 %nota, i8 %b)
+  %e0 = extractvalue {i8, i1} %a0, 0
+  %e1 = extractvalue {i8, i1} %a0, 1
+  %res = add i8 %e0, 1
+  store i8 %res, i8* %p
+  ret i1 %e1
+}
+
 ; TODO: adds/subs
diff --git a/test/CodeGen/X86/combine-add.ll b/test/CodeGen/X86/combine-add.ll
index 81b38334cac..3bff9ea8b84 100644
--- a/test/CodeGen/X86/combine-add.ll
+++ b/test/CodeGen/X86/combine-add.ll
@@ -1,16 +1,12 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2   | FileCheck %s --check-prefixes=CHECK,AVX
 
 ; fold (add x, 0) -> x
 define <4 x i32> @combine_vec_add_to_zero(<4 x i32> %a) {
-; SSE-LABEL: combine_vec_add_to_zero:
-; SSE:       # %bb.0:
-; SSE-NEXT:    retq
-;
-; AVX-LABEL: combine_vec_add_to_zero:
-; AVX:       # %bb.0:
-; AVX-NEXT:    retq
+; CHECK-LABEL: combine_vec_add_to_zero:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    retq
   %1 = add <4 x i32> %a, zeroinitializer
   ret <4 x i32> %1
 }
@@ -352,17 +348,11 @@ define <4 x i32> @combine_vec_add_sextinreg(<4 x i32> %a0, <4 x i32> %a1) {
 
 ; (add (add (xor a, -1), b), 1) -> (sub b, a)
 define i32 @combine_add_add_not(i32 %a, i32 %b) {
-; SSE-LABEL: combine_add_add_not:
-; SSE:       # %bb.0:
-; SSE-NEXT:    movl %esi, %eax
-; SSE-NEXT:    subl %edi, %eax
-; SSE-NEXT:    retq
-;
-; AVX-LABEL: combine_add_add_not:
-; AVX:       # %bb.0:
-; AVX-NEXT:    movl %esi, %eax
-; AVX-NEXT:    subl %edi, %eax
-; AVX-NEXT:    retq
+; CHECK-LABEL: combine_add_add_not:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movl %esi, %eax
+; CHECK-NEXT:    subl %edi, %eax
+; CHECK-NEXT:    retq
   %nota = xor i32 %a, -1
   %add = add i32 %nota, %b
   %r = add i32 %add, 1
@@ -385,3 +375,45 @@ define <4 x i32> @combine_vec_add_add_not(<4 x i32> %a, <4 x i32> %b) {
   %r = add <4 x i32> %add, <i32 1, i32 1, i32 1, i32 1>
   ret <4 x i32> %r
 }
+
+declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
+
+define i1 @sadd_add(i32 %a, i32 %b, i32* %p) {
+; CHECK-LABEL: sadd_add:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    notl %eax
+; CHECK-NEXT:    addl %esi, %eax
+; CHECK-NEXT:    seto %al
+; CHECK-NEXT:    subl %edi, %esi
+; CHECK-NEXT:    movl %esi, (%rdx)
+; CHECK-NEXT:    retq
+  %nota = xor i32 %a, -1
+  %a0 = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %nota, i32 %b)
+  %e0 = extractvalue {i32, i1} %a0, 0
+  %e1 = extractvalue {i32, i1} %a0, 1
+  %res = add i32 %e0, 1
+  store i32 %res, i32* %p
+  ret i1 %e1
+}
+
+declare {i8, i1} @llvm.uadd.with.overflow.i8(i8 %a, i8 %b)
+
+define i1 @uadd_add(i8 %a, i8 %b, i8* %p) {
+; CHECK-LABEL: uadd_add:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    notb %al
+; CHECK-NEXT:    addb %sil, %al
+; CHECK-NEXT:    setb %al
+; CHECK-NEXT:    subb %dil, %sil
+; CHECK-NEXT:    movb %sil, (%rdx)
+; CHECK-NEXT:    retq
+  %nota = xor i8 %a, -1
+  %a0 = call {i8, i1} @llvm.uadd.with.overflow.i8(i8 %nota, i8 %b)
+  %e0 = extractvalue {i8, i1} %a0, 0
+  %e1 = extractvalue {i8, i1} %a0, 1
+  %res = add i8 %e0, 1
+  store i8 %res, i8* %p
+  ret i1 %e1
+}