From 1eb3b19f63f05bad1cb5776bb5ca39b8f192bc23 Mon Sep 17 00:00:00 2001
From: Apostolos Tapsas <apostolos@vm01.debian11>
Date: Sun, 24 Oct 2021 16:52:12 +0000
Subject: [PATCH] Shuffle simd and SuperVector implementetions as well as their
 test realy fixed

---
 src/nfa/limex_shuffle.h                    |  1 -
 src/util/arch/ppc64el/simd_utils.h         |  4 +++-
 src/util/supervector/arch/ppc64el/impl.cpp |  4 +++-
 unit/internal/shuffle.cpp                  |  6 +++---
 unit/internal/simd_utils.cpp               | 17 ++++++++++++-----
 unit/internal/supervector.cpp              |  6 +++++-
 6 files changed, 26 insertions(+), 12 deletions(-)

diff --git a/src/nfa/limex_shuffle.h b/src/nfa/limex_shuffle.h
index a1728e6a..365d4729 100644
--- a/src/nfa/limex_shuffle.h
+++ b/src/nfa/limex_shuffle.h
@@ -45,7 +45,6 @@
 static really_inline
 u32 packedExtract128(m128 s, const m128 permute, const m128 compare) {
     m128 shuffled = pshufb_m128(s, permute);
-    print_m128_16x8("shufled", shuffled);
     m128 compared = and128(shuffled, compare);
     u16 rv = ~movemask128(eq128(compared, shuffled));
     return (u32)rv;
diff --git a/src/util/arch/ppc64el/simd_utils.h b/src/util/arch/ppc64el/simd_utils.h
index 9e8c59bf..107ca110 100644
--- a/src/util/arch/ppc64el/simd_utils.h
+++ b/src/util/arch/ppc64el/simd_utils.h
@@ -462,7 +462,9 @@ char testbit128(m128 val, unsigned int n) {
 
 static really_inline
 m128 pshufb_m128(m128 a, m128 b) {
-    return (m128) vec_perm((uint8x16_t)a, (uint8x16_t)a, (uint8x16_t)b);
+    uint8x16_t mask =(uint8x16_t)vec_cmpge((uint8x16_t)b, (uint8x16_t)vec_splats((uint8_t)0x80));
+    uint8x16_t res = vec_perm ((uint8x16_t)a, (uint8x16_t)a, (uint8x16_t)b);
+    return (m128) vec_sel((uint8x16_t)res, (uint8x16_t)zeroes128(), (uint8x16_t)mask);
 }
 
 static really_inline
diff --git a/src/util/supervector/arch/ppc64el/impl.cpp b/src/util/supervector/arch/ppc64el/impl.cpp
index dc318c82..0af136a5 100644
--- a/src/util/supervector/arch/ppc64el/impl.cpp
+++ b/src/util/supervector/arch/ppc64el/impl.cpp
@@ -603,7 +603,9 @@ template<>
 template<>
 really_inline SuperVector<16> SuperVector<16>::pshufb<false>(SuperVector<16> b)
 {
-    return (m128) vec_perm((uint8x16_t)u.v128[0], (uint8x16_t)u.v128[0], (uint8x16_t)b.u.v128[0]);
+    uint8x16_t mask =(uint8x16_t)vec_cmpge((uint8x16_t)b.u.v128[0], (uint8x16_t)vec_splats((uint8_t)0x80));
+    uint8x16_t res = vec_perm ((uint8x16_t)u.v128[0], (uint8x16_t)u.v128[0], (uint8x16_t)b.u.v128[0]);
+    return (m128) vec_sel((uint8x16_t)res, (uint8x16_t)vec_splat_s8(0), (uint8x16_t)mask);
 }
 
 template<>
diff --git a/unit/internal/shuffle.cpp b/unit/internal/shuffle.cpp
index b7c1b4f5..038c6193 100644
--- a/unit/internal/shuffle.cpp
+++ b/unit/internal/shuffle.cpp
@@ -187,7 +187,7 @@ TEST(Shuffle, PackedExtract128_1) {
         // shuffle a single 1 bit to the front
         m128 permute, compare;
         build_pshufb_masks_onebit(i, &permute, &compare);
-	    EXPECT_EQ(1U, packedExtract128(setbit<m128>(i), permute, compare));
+	EXPECT_EQ(1U, packedExtract128(setbit<m128>(i), permute, compare));
         EXPECT_EQ(1U, packedExtract128(ones128(), permute, compare));
         // we should get zero out of these cases
         EXPECT_EQ(0U, packedExtract128(zeroes128(), permute, compare));
@@ -199,7 +199,7 @@ TEST(Shuffle, PackedExtract128_1) {
     }
 }
 
-/*
+
 TEST(Shuffle, PackedExtract_templatized_128_1) {
     // Try all possible one-bit masks
     for (unsigned int i = 0; i < 128; i++) {
@@ -218,7 +218,7 @@ TEST(Shuffle, PackedExtract_templatized_128_1) {
         }
     }
 }
-*/
+
 
 
 #if defined(HAVE_AVX2)
diff --git a/unit/internal/simd_utils.cpp b/unit/internal/simd_utils.cpp
index 2085c9df..037230d0 100644
--- a/unit/internal/simd_utils.cpp
+++ b/unit/internal/simd_utils.cpp
@@ -849,15 +849,22 @@ TEST(SimdUtilsTest, pshufb_m128) {
     }
     u8 vec2[16];
     for (int i=0; i<16; i++) {
-        vec2[i]=i + (rand() % 16 + 0);
-    } 
+        vec2[i]=i + (rand() % 15 + 0);
+    }
+
     m128 v1 = loadu128(vec);
     m128 v2 = loadu128(vec2);
-    m128 vres = pshufb_m128(v1, v2);
+    m128 vres = pshufb_m128(v1, v2); 
+    
     u8 res[16];
-    store128(res, vres);
+    storeu128(res, vres);
+
     for (int i=0; i<16; i++) {
-        ASSERT_EQ(vec[vec2[i] % 16 ], res[i]);
+	if(vec2[i] & 0x80){
+	   ASSERT_EQ(res[i], 0);
+        }else{	   
+           ASSERT_EQ(vec[vec2[i] % 16 ], res[i]);
+	}
     }
 }
 
diff --git a/unit/internal/supervector.cpp b/unit/internal/supervector.cpp
index 4be93aa8..9c5f8f3a 100644
--- a/unit/internal/supervector.cpp
+++ b/unit/internal/supervector.cpp
@@ -286,7 +286,11 @@ TEST(SuperVectorUtilsTest,pshufb128c) {
     auto SP2 = SuperVector<16>::loadu(vec2);
     auto SResult = SP1.template pshufb<true>(SP2);
     for (int i=0; i<16; i++) {
-        ASSERT_EQ(vec[vec2[i] % 16 ],SResult.u.u8[i]);
+	if(vec2[i] & 0x80){
+	   ASSERT_EQ(SResult.u.u8[i], 0);
+	}else{
+           ASSERT_EQ(vec[vec2[i] % 16 ],SResult.u.u8[i]);
+	}
     }
 }