[llvm] c52280b - [InstCombine] Tests for D156026 (Contracting x^2 + 2*x*y + y^2 to (x + y)^2) (NFC)
Nikita Popov via llvm-commits
llvm-commits at lists.llvm.org
Wed Aug 9 05:13:48 PDT 2023
Author: Christoph Stiller
Date: 2023-08-09T14:13:41+02:00
New Revision: c52280b6103a26b8e33749e687323571beebc0ba
URL: https://github.com/llvm/llvm-project/commit/c52280b6103a26b8e33749e687323571beebc0ba
DIFF: https://github.com/llvm/llvm-project/commit/c52280b6103a26b8e33749e687323571beebc0ba.diff
LOG: [InstCombine] Tests for D156026 (Contracting x^2 + 2*x*y + y^2 to (x + y)^2) (NFC)
Differential Revision: https://reviews.llvm.org/D156347
Added:
Modified:
llvm/test/Transforms/InstCombine/add.ll
Removed:
################################################################################
diff --git a/llvm/test/Transforms/InstCombine/add.ll b/llvm/test/Transforms/InstCombine/add.ll
index 3ba01fbce0ba6a..5dada159d543da 100644
--- a/llvm/test/Transforms/InstCombine/add.ll
+++ b/llvm/test/Transforms/InstCombine/add.ll
@@ -3096,4 +3096,879 @@ define i32 @add_zext_sext_i1_
diff erent_values(i1 %a, i1 %b) {
ret i32 %add
}
+define i32 @add_reduce_sqr_sum_nsw(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_nsw(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %two_a = shl i32 %a, 1
+ %two_a_plus_b = add i32 %two_a, %b
+ %mul = mul i32 %two_a_plus_b, %b
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_u(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_u(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul i32 %a, %a
+ %two_a = shl i32 %a, 1
+ %two_a_plus_b = add i32 %two_a, %b
+ %mul = mul i32 %two_a_plus_b, %b
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_nuw(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_nuw(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nuw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul nuw i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nuw i32 %a, %a
+ %two_a = mul i32 %a, 2
+ %two_a_plus_b = add i32 %two_a, %b
+ %mul = mul nuw i32 %two_a_plus_b, %b
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_flipped(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_flipped(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[A_SQ]], [[MUL]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %two_a = shl i32 %a, 1
+ %two_a_plus_b = add i32 %two_a, %b
+ %mul = mul i32 %two_a_plus_b, %b
+ %add = add i32 %a_sq, %mul
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_flipped2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_flipped2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %two_a = shl i32 %a, 1
+ %two_a_plus_b = add i32 %two_a, %b
+ %mul = mul i32 %b, %two_a_plus_b
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_flipped3(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_flipped3(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %two_a = shl i32 %a, 1
+ %two_a_plus_b = add i32 %b, %two_a
+ %mul = mul i32 %two_a_plus_b, %b
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_order2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB1:%.*]] = add i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB_B2:%.*]] = mul i32 [[TWOAB1]], [[B]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[A_SQ]], [[TWOAB_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %twoab_b2 = add i32 %twoab, %b_sq
+ %ab2 = add i32 %a_sq, %twoab_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order2_flipped(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order2_flipped(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB1:%.*]] = add i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB_B2:%.*]] = mul i32 [[TWOAB1]], [[B]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB_B2]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %twoab_b2 = add i32 %twoab, %b_sq
+ %ab2 = add i32 %twoab_b2, %a_sq
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order2_flipped2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order2_flipped2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB1:%.*]] = add i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB_B2:%.*]] = mul i32 [[TWOAB1]], [[B]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[A_SQ]], [[TWOAB_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %twoab_b2 = add i32 %b_sq, %twoab
+ %ab2 = add i32 %a_sq, %twoab_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order2_flipped3(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order2_flipped3(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB1:%.*]] = add i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB_B2:%.*]] = mul i32 [[TWOAB1]], [[B]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[A_SQ]], [[TWOAB_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %b, %twoa
+ %b_sq = mul i32 %b, %b
+ %twoab_b2 = add i32 %twoab, %b_sq
+ %ab2 = add i32 %a_sq, %twoab_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order3(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order3(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order3_flipped(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order3_flipped(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[A2_B2]], [[TWOAB]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %a2_b2, %twoab
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order3_flipped2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order3_flipped2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[B_SQ]], [[A_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %b_sq, %a_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order3_flipped3(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order3_flipped3(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %b, %twoa
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order4(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order4(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[AB:%.*]] = mul i32 [[A]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[AB]], 1
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %ab = mul i32 %a, %b
+ %twoab = mul i32 %ab, 2
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order4_flipped(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order4_flipped(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[AB:%.*]] = mul i32 [[A]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[AB]], 1
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[A2_B2]], [[TWOAB]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %ab = mul i32 %a, %b
+ %twoab = mul i32 %ab, 2
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %a2_b2, %twoab
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order4_flipped2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order4_flipped2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[AB:%.*]] = mul i32 [[A]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[AB]], 1
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[B_SQ]], [[A_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %ab = mul i32 %a, %b
+ %twoab = mul i32 %ab, 2
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %b_sq, %a_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order4_flipped3(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order4_flipped3(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[AB:%.*]] = mul i32 [[A]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[AB]], 1
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %ab = mul i32 %a, %b
+ %twoab = mul i32 2, %ab
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order4_flipped4(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order4_flipped4(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[AB:%.*]] = mul i32 [[B:%.*]], [[A]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[AB]], 1
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %ab = mul i32 %b, %a
+ %twoab = mul i32 %ab, 2
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order5(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order5(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOB:%.*]] = shl i32 [[B:%.*]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOB]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twob = mul i32 %b, 2
+ %twoab = mul i32 %twob, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order5_flipped(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order5_flipped(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOB:%.*]] = shl i32 [[B:%.*]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOB]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[A2_B2]], [[TWOAB]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twob = mul i32 %b, 2
+ %twoab = mul i32 %twob, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %a2_b2, %twoab
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order5_flipped2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order5_flipped2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOB:%.*]] = shl i32 [[B:%.*]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOB]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[B_SQ]], [[A_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twob = mul i32 %b, 2
+ %twoab = mul i32 %twob, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %b_sq, %a_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order5_flipped3(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order5_flipped3(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOB:%.*]] = shl i32 [[B:%.*]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOB]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twob = mul i32 %b, 2
+ %twoab = mul i32 %a, %twob
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order5_flipped4(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order5_flipped4(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOB:%.*]] = shl i32 [[B:%.*]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOB]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twob = mul i32 2, %b
+ %twoab = mul i32 %twob, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_not_one_use(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_not_one_use(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[MUL]])
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %two_a = shl i32 %a, 1
+ %two_a_plus_b = add i32 %two_a, %b
+ %mul = mul i32 %two_a_plus_b, %b
+ tail call void @fake_func (i32 %mul)
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_not_one_use2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_not_one_use2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[A_SQ]])
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %two_a = shl i32 %a, 1
+ %two_a_plus_b = add i32 %two_a, %b
+ %mul = mul i32 %two_a_plus_b, %b
+ tail call void @fake_func (i32 %a_sq)
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_order2_not_one_use(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order2_not_one_use(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB1:%.*]] = add i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB_B2:%.*]] = mul i32 [[TWOAB1]], [[B]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[TWOAB_B2]])
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[A_SQ]], [[TWOAB_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %twoab_b2 = add i32 %twoab, %b_sq
+ tail call void @fake_func (i32 %twoab_b2)
+ %ab2 = add i32 %a_sq, %twoab_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order2_not_one_use2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order2_not_one_use2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB1:%.*]] = add i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB_B2:%.*]] = mul i32 [[TWOAB1]], [[B]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[A_SQ]])
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[A_SQ]], [[TWOAB_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %twoab_b2 = add i32 %twoab, %b_sq
+ tail call void @fake_func (i32 %a_sq)
+ %ab2 = add i32 %a_sq, %twoab_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order3_not_one_use(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order3_not_one_use(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[TWOAB]])
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ tail call void @fake_func (i32 %twoab)
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order3_not_one_use2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order3_not_one_use2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOA:%.*]] = shl i32 [[A]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[A2_B2]])
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twoa = mul i32 %a, 2
+ %twoab = mul i32 %twoa, %b
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ tail call void @fake_func (i32 %a2_b2)
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order4_not_one_use(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order4_not_one_use(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[AB:%.*]] = mul i32 [[A]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[AB]], 1
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[TWOAB]])
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %ab = mul i32 %a, %b
+ %twoab = mul i32 %ab, 2
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ tail call void @fake_func (i32 %twoab)
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order4_not_one_use2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order4_not_one_use2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[AB:%.*]] = mul i32 [[A]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[AB]], 1
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[A2_B2]])
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %ab = mul i32 %a, %b
+ %twoab = mul i32 %ab, 2
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ tail call void @fake_func (i32 %a2_b2)
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order5_not_one_use(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order5_not_one_use(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOB:%.*]] = shl i32 [[B:%.*]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOB]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[TWOAB]])
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twob = mul i32 %b, 2
+ %twoab = mul i32 %twob, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ tail call void @fake_func (i32 %twoab)
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_order5_not_one_use2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_order5_not_one_use2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOB:%.*]] = shl i32 [[B:%.*]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[TWOB]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: tail call void @fake_func(i32 [[A2_B2]])
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %twob = mul i32 %b, 2
+ %twoab = mul i32 %twob, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ tail call void @fake_func (i32 %a2_b2)
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_invalid0(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_invalid0(
+; CHECK-NEXT: [[TWO_A:%.*]] = shl i32 [[A:%.*]], 1
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL1:%.*]] = add i32 [[TWO_A_PLUS_B]], [[A]]
+; CHECK-NEXT: [[ADD:%.*]] = mul i32 [[MUL1]], [[B]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %not_a_sq = mul nsw i32 %a, %b
+ %two_a = shl i32 %a, 1
+ %two_a_plus_b = add i32 %two_a, %b
+ %mul = mul i32 %two_a_plus_b, %b
+ %add = add i32 %mul, %not_a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_invalid1(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_invalid1(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[NOT_TWO_A_PLUS_B:%.*]] = mul i32 [[A]], 3
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[NOT_TWO_A_PLUS_B]], [[B:%.*]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %two_a = shl i32 %a, 1
+ %not_two_a_plus_b = add i32 %two_a, %a
+ %mul = mul i32 %not_two_a_plus_b, %b
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_invalid2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_invalid2(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[NOT_TWO_A:%.*]] = shl i32 [[A]], 2
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = add i32 [[NOT_TWO_A]], [[B:%.*]]
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %not_two_a = shl i32 %a, 2
+ %two_a_plus_b = add i32 %not_two_a, %b
+ %mul = mul i32 %two_a_plus_b, %b
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_invalid3(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_invalid3(
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = mul i32 [[B:%.*]], 3
+; CHECK-NEXT: [[MUL1:%.*]] = add i32 [[TWO_A_PLUS_B]], [[A:%.*]]
+; CHECK-NEXT: [[ADD:%.*]] = mul i32 [[MUL1]], [[A]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %not_two_a = shl i32 %b, 1
+ %two_a_plus_b = add i32 %not_two_a, %b
+ %mul = mul i32 %two_a_plus_b, %a
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_invalid4(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_invalid4(
+; CHECK-NEXT: [[A_SQ:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWO_A_PLUS_B:%.*]] = mul i32 [[B:%.*]], 3
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[TWO_A_PLUS_B]], [[B]]
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[A_SQ]]
+; CHECK-NEXT: ret i32 [[ADD]]
+;
+ %a_sq = mul nsw i32 %a, %a
+ %not_two_a = shl i32 %b, 1
+ %two_a_plus_b = add i32 %not_two_a, %b
+ %mul = mul i32 %two_a_plus_b, %b
+ %add = add i32 %mul, %a_sq
+ ret i32 %add
+}
+
+define i32 @add_reduce_sqr_sum_varB_invalid0(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_varB_invalid0(
+; CHECK-NEXT: [[NOT_A_B:%.*]] = mul nsw i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl nuw i32 [[NOT_A_B]], 1
+; CHECK-NEXT: [[A_SQ:%.*]] = mul i32 [[A]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B:%.*]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %not_a_b = mul nsw i32 %a, %a
+ %twoab = mul i32 %not_a_b, 2
+ %a_sq = mul i32 %a, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_varB_invalid1(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_varB_invalid1(
+; CHECK-NEXT: [[NOT_A_B:%.*]] = mul nsw i32 [[B:%.*]], [[B]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl nuw i32 [[NOT_A_B]], 1
+; CHECK-NEXT: [[A_SQ:%.*]] = mul i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %not_a_b = mul nsw i32 %b, %b
+ %twoab = mul i32 %not_a_b, 2
+ %a_sq = mul i32 %a, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_varB_invalid2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_varB_invalid2(
+; CHECK-NEXT: [[A_B:%.*]] = mul nsw i32 [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[NOT_TWOAB:%.*]] = shl i32 [[A_B]], 2
+; CHECK-NEXT: [[A_SQ:%.*]] = mul i32 [[A]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[NOT_TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_b = mul nsw i32 %a, %b
+ %not_twoab = mul i32 %a_b, 4
+ %a_sq = mul i32 %a, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %not_twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_varB_invalid3(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_varB_invalid3(
+; CHECK-NEXT: [[A_B:%.*]] = mul nsw i32 [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[A_B]], 1
+; CHECK-NEXT: [[B_SQ1:%.*]] = add i32 [[A]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = mul i32 [[B_SQ1]], [[B]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_b = mul nsw i32 %a, %b
+ %twoab = mul i32 %a_b, 2
+ %not_a_sq = mul i32 %b, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %not_a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_varB_invalid4(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_varB_invalid4(
+; CHECK-NEXT: [[A_B:%.*]] = mul nsw i32 [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT: [[TWOAB:%.*]] = shl i32 [[A_B]], 1
+; CHECK-NEXT: [[NOT_B_SQ1:%.*]] = add i32 [[A]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = mul i32 [[NOT_B_SQ1]], [[A]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %a_b = mul nsw i32 %a, %b
+ %twoab = mul i32 %a_b, 2
+ %a_sq = mul i32 %a, %a
+ %not_b_sq = mul i32 %b, %a
+ %a2_b2 = add i32 %a_sq, %not_b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_varC_invalid0(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_varC_invalid0(
+; CHECK-NEXT: [[NOT_TWOA:%.*]] = shl nsw i32 [[B:%.*]], 1
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[NOT_TWOA]], [[B]]
+; CHECK-NEXT: [[A_SQ:%.*]] = mul i32 [[A:%.*]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %not_twoa = mul nsw i32 %b, 2
+ %twoab = mul i32 %not_twoa, %b
+ %a_sq = mul i32 %a, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_varC_invalid1(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_varC_invalid1(
+; CHECK-NEXT: [[NOT_TWOA:%.*]] = shl nsw i32 [[A:%.*]], 2
+; CHECK-NEXT: [[TWOAB:%.*]] = mul i32 [[NOT_TWOA]], [[B:%.*]]
+; CHECK-NEXT: [[A_SQ:%.*]] = mul i32 [[A]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %not_twoa = mul nsw i32 %a, 4
+ %twoab = mul i32 %not_twoa, %b
+ %a_sq = mul i32 %a, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %twoab, %a2_b2
+ ret i32 %ab2
+}
+
+define i32 @add_reduce_sqr_sum_varC_invalid2(i32 %a, i32 %b) {
+; CHECK-LABEL: @add_reduce_sqr_sum_varC_invalid2(
+; CHECK-NEXT: [[TWOA:%.*]] = shl nsw i32 [[A:%.*]], 1
+; CHECK-NEXT: [[NOT_TWOAB:%.*]] = mul i32 [[TWOA]], [[A]]
+; CHECK-NEXT: [[A_SQ:%.*]] = mul i32 [[A]], [[A]]
+; CHECK-NEXT: [[B_SQ:%.*]] = mul i32 [[B:%.*]], [[B]]
+; CHECK-NEXT: [[A2_B2:%.*]] = add i32 [[A_SQ]], [[B_SQ]]
+; CHECK-NEXT: [[AB2:%.*]] = add i32 [[NOT_TWOAB]], [[A2_B2]]
+; CHECK-NEXT: ret i32 [[AB2]]
+;
+ %twoa = mul nsw i32 %a, 2
+ %not_twoab = mul i32 %twoa, %a
+ %a_sq = mul i32 %a, %a
+ %b_sq = mul i32 %b, %b
+ %a2_b2 = add i32 %a_sq, %b_sq
+ %ab2 = add i32 %not_twoab, %a2_b2
+ ret i32 %ab2
+}
+
declare void @llvm.assume(i1)
+declare void @fake_func(i32)
More information about the llvm-commits
mailing list