[InstCombine] Canonicalize sadd.with.overflow with clamp to sadd.sat

This adds to D69245, adding extra signed patterns for folding from a
sadd_with_overflow to a sadd_sat. These are more complex than the
unsigned patterns, as the overflow can occur in either direction.

For the add case, the positive overflow can only occur if both of the
values are positive (same for both the values being negative). So there
is an extra select on whether to use the positive or negative overflow
limit.

Differential Revision: https://reviews.llvm.org/D69252

1 files changed, 9 insertions, 54 deletions

diff --git a/llvm/test/Transforms/InstCombine/overflow_to_sat.ll b/llvm/test/Transforms/InstCombine/overflow_to_sat.ll
index 12ee430ecfc..474c8fe6b3c 100644
--- a/llvm/test/Transforms/InstCombine/overflow_to_sat.ll
+++ b/llvm/test/Transforms/InstCombine/overflow_to_sat.ll
@@ -47,12 +47,7 @@ define i8 @sadd_x_lt_min(i8 %x, i8 %y) {
 
 define i8 @sadd_x_lt_max(i8 %x, i8 %y) {
 ; CHECK-LABEL: @sadd_x_lt_max(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i8, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i8, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp slt i8 [[X]], 0
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i8 -128, i8 127
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i8 [[S]], i8 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i8 @llvm.sadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %ao = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %x, i8 %y)
@@ -85,12 +80,7 @@ define i8 @sadd_x_le_min(i8 %x, i8 %y) {
 
 define i8 @sadd_x_le_max(i8 %x, i8 %y) {
 ; CHECK-LABEL: @sadd_x_le_max(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i8, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i8, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp slt i8 [[X]], 1
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i8 -128, i8 127
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i8 [[S]], i8 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i8 @llvm.sadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %ao = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %x, i8 %y)
@@ -104,12 +94,7 @@ define i8 @sadd_x_le_max(i8 %x, i8 %y) {
 
 define i8 @sadd_x_gt_min(i8 %x, i8 %y) {
 ; CHECK-LABEL: @sadd_x_gt_min(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i8, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i8, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp sgt i8 [[X]], 0
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i8 127, i8 -128
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i8 [[S]], i8 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i8 @llvm.sadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %ao = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %x, i8 %y)
@@ -142,12 +127,7 @@ define i8 @sadd_x_gt_max(i8 %x, i8 %y) {
 
 define i8 @sadd_x_ge_min(i8 %x, i8 %y) {
 ; CHECK-LABEL: @sadd_x_ge_min(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i8, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i8, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp sgt i8 [[X]], -1
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i8 127, i8 -128
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i8 [[S]], i8 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i8 @llvm.sadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %ao = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %x, i8 %y)
@@ -200,12 +180,7 @@ define i8 @sadd_y_lt_min(i8 %x, i8 %y) {
 
 define i8 @sadd_y_lt_max(i8 %x, i8 %y) {
 ; CHECK-LABEL: @sadd_y_lt_max(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i8, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i8, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp slt i8 [[Y]], 0
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i8 -128, i8 127
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i8 [[S]], i8 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i8 @llvm.sadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %ao = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %x, i8 %y)
@@ -238,12 +213,7 @@ define i8 @sadd_y_le_min(i8 %x, i8 %y) {
 
 define i8 @sadd_y_le_max(i8 %x, i8 %y) {
 ; CHECK-LABEL: @sadd_y_le_max(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i8, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i8, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp slt i8 [[Y]], 1
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i8 -128, i8 127
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i8 [[S]], i8 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i8 @llvm.sadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %ao = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %x, i8 %y)
@@ -257,12 +227,7 @@ define i8 @sadd_y_le_max(i8 %x, i8 %y) {
 
 define i8 @sadd_y_gt_min(i8 %x, i8 %y) {
 ; CHECK-LABEL: @sadd_y_gt_min(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i8, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i8, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp sgt i8 [[Y]], 0
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i8 127, i8 -128
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i8 [[S]], i8 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i8 @llvm.sadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %ao = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %x, i8 %y)
@@ -295,12 +260,7 @@ define i8 @sadd_y_gt_max(i8 %x, i8 %y) {
 
 define i8 @sadd_y_ge_min(i8 %x, i8 %y) {
 ; CHECK-LABEL: @sadd_y_ge_min(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i8, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i8, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp sgt i8 [[Y]], -1
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i8 127, i8 -128
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i8 [[S]], i8 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i8 @llvm.sadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %ao = tail call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 %x, i8 %y)
@@ -718,12 +678,7 @@ define i8 @ssub_y_ge_max(i8 %x, i8 %y) {
 
 define i32 @sadd_i32(i32 %x, i32 %y) {
 ; CHECK-LABEL: @sadd_i32(
-; CHECK-NEXT:    [[AO:%.*]] = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 [[X:%.*]], i32 [[Y:%.*]])
-; CHECK-NEXT:    [[O:%.*]] = extractvalue { i32, i1 } [[AO]], 1
-; CHECK-NEXT:    [[A:%.*]] = extractvalue { i32, i1 } [[AO]], 0
-; CHECK-NEXT:    [[C:%.*]] = icmp slt i32 [[X]], 0
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[C]], i32 -2147483648, i32 2147483647
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[O]], i32 [[S]], i32 [[A]]
+; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.sadd.sat.i32(i32 [[X:%.*]], i32 [[Y:%.*]])
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
   %ao = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %x, i32 %y)