From c27a8c8c8b867e6812b905f6154762802a498dbd Mon Sep 17 00:00:00 2001
From: Emanuel Peter <epeter@openjdk.org>
Date: Tue, 13 Aug 2024 05:52:58 +0000
Subject: [PATCH] 8338124: C2 SuperWord: MulAddS2I input permutation still
 partially broken after JDK-8333840

Reviewed-by: chagedorn, thartmann, kvn
---
 src/hotspot/share/opto/superword.cpp           |  2 +-
 .../loopopts/superword/TestMulAddS2I.java      | 18 +++++++++++++++++-
 2 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/src/hotspot/share/opto/superword.cpp b/src/hotspot/share/opto/superword.cpp
index 2a43d2ec0af4a..bb5fed78b0274 100644
--- a/src/hotspot/share/opto/superword.cpp
+++ b/src/hotspot/share/opto/superword.cpp
@@ -2270,7 +2270,7 @@ Node_List* PackSet::strided_pack_input_at_index_or_null(const Node_List* pack, c
     return nullptr; // size mismatch
   }
 
-  for (uint i = 1; i < pack->size(); i++) {
+  for (uint i = 0; i < pack->size(); i++) {
     if (pack->at(i)->in(index) != pack_in->at(i * stride + offset)) {
       return nullptr; // use-def mismatch
     }
diff --git a/test/hotspot/jtreg/compiler/loopopts/superword/TestMulAddS2I.java b/test/hotspot/jtreg/compiler/loopopts/superword/TestMulAddS2I.java
index e0f073c7f8f7d..fb99fc5983a16 100644
--- a/test/hotspot/jtreg/compiler/loopopts/superword/TestMulAddS2I.java
+++ b/test/hotspot/jtreg/compiler/loopopts/superword/TestMulAddS2I.java
@@ -53,6 +53,7 @@ public class TestMulAddS2I {
     static final int[] GOLDEN_J;
     static final int[] GOLDEN_K;
     static final int[] GOLDEN_L;
+    static final int[] GOLDEN_M;
 
     static {
         for (int i = 0; i < RANGE; i++) {
@@ -71,6 +72,7 @@ public class TestMulAddS2I {
         GOLDEN_J = testj(new int[ITER]);
         GOLDEN_K = testk(new int[ITER]);
         GOLDEN_L = testl(new int[ITER]);
+        GOLDEN_M = testm(new int[ITER]);
     }
 
 
@@ -80,7 +82,7 @@ public static void main(String[] args) {
     }
 
     @Run(test = {"testa", "testb", "testc", "testd", "teste", "testf", "testg", "testh",
-                 "testi", "testj", "testk", "testl"})
+                 "testi", "testj", "testk", "testl", "testm"})
     @Warmup(0)
     public static void run() {
         compare(testa(), GOLDEN_A, "testa");
@@ -95,6 +97,7 @@ public static void run() {
         compare(testj(new int[ITER]), GOLDEN_J, "testj");
         compare(testk(new int[ITER]), GOLDEN_K, "testk");
         compare(testl(new int[ITER]), GOLDEN_L, "testl");
+        compare(testm(new int[ITER]), GOLDEN_M, "testm");
     }
 
     public static void compare(int[] out, int[] golden, String name) {
@@ -299,4 +302,17 @@ public static int[] testl(int[] out) {
         return out;
     }
 
+    @Test
+    @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
+        applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
+    @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
+    public static int[] testm(int[] out) {
+        for (int i = 0; i < ITER-4; i+=4) {
+            // Unrolled, with some swaps that prevent vectorization.
+            out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+1]) + (sArr1[2*i+1] * sArr2[2*i+0])); // bad
+            out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+2]) + (sArr1[2*i+3] * sArr2[2*i+3])); // ok
+            // 2-element gap
+        }
+        return out;
+    }
 }