SuperVector operators fixes and simd_utils low/high64 functions implementations added

diff --git a/src/util/arch/common/simd_utils.h b/src/util/arch/common/simd_utils.h
index 65e7b69ab012ea13f065800d264af533e0b4aba3..5bf846f94b4888138c1c7f059656753c46d3d9e1 100644 (file)
--- a/src/util/arch/common/simd_utils.h
+++ b/src/util/arch/common/simd_utils.h
@@ -49,8 +49,8 @@
 static inline void print_m128_16x8(const char *label, m128 vector) {
     uint8_t ALIGN_ATTR(16) data[16];
     store128(data, vector);
-    DEBUG_PRINTF("%s: ", label);
-    for(int i=0; i < 16; i++)
+    DEBUG_PRINTF("%12s: ", label);
+    for(int i=15; i >=0; i--)
         printf("%02x ", data[i]);
     printf("\n");
 }
@@ -58,8 +58,8 @@ static inline void print_m128_16x8(const char *label, m128 vector) {
 static inline void print_m128_8x16(const char *label, m128 vector) {
     uint16_t ALIGN_ATTR(16) data[8];
     store128(data, vector);
-    DEBUG_PRINTF("%s: ", label);
-    for(int i=0; i < 8; i++)
+    DEBUG_PRINTF("%12s: ", label);
+    for(int i=7; i >= 0; i--)
         printf("%04x ", data[i]);
     printf("\n");
 }
@@ -67,8 +67,8 @@ static inline void print_m128_8x16(const char *label, m128 vector) {
 static inline void print_m128_4x32(const char *label, m128 vector) {
     uint32_t ALIGN_ATTR(16) data[4];
     store128(data, vector);
-    DEBUG_PRINTF("%s: ", label);
-    for(int i=0; i < 4; i++)
+    DEBUG_PRINTF("%12s: ", label);
+    for(int i=3; i >= 0; i--)
         printf("%08x ", data[i]);
     printf("\n");
 }
@@ -76,8 +76,8 @@ static inline void print_m128_4x32(const char *label, m128 vector) {
 static inline void print_m128_2x64(const char *label, m128 vector) {
     uint64_t ALIGN_ATTR(16) data[2];
     store128(data, vector);
-    DEBUG_PRINTF("%s: ", label);
-    for(int i=0; i < 2; i++)
+    DEBUG_PRINTF("%12s: ", label);
+    for(int i=1; i >= 0; i--)
         printf("%016lx ", data[i]);
     printf("\n");
 }

diff --git a/src/util/arch/ppc64el/simd_utils.h b/src/util/arch/ppc64el/simd_utils.h
index f4b97ffb471ffe3e84038dc2149569300b07b26d..a54012aafbde3757ca40ba972312ac1742649df1 100644 (file)
--- a/src/util/arch/ppc64el/simd_utils.h
+++ b/src/util/arch/ppc64el/simd_utils.h
@@ -270,7 +270,7 @@ switch (imm) {
     }
 }
 
-static really_inline u64a extract64from128(const m128 in, unsigned UNUSED imm) {
+static really_inline u64a extract64from128(const m128 in, unsigned imm) {
 u64a ALIGN_ATTR(16) a[2];
 vec_xst((uint64x2_t) in, 0, a);
 switch (imm) {
@@ -285,19 +285,11 @@ switch (imm) {
 }
 
 static really_inline m128 low64from128(const m128 in) {
-    //u64a ALIGN_ATTR(16) a[2];
-    //vec_xst((uint64x2_t) in, 0, a);
-    //return a[1];
-    // #warning FIXME
-    return vec_add(in, in);
+    return (m128) vec_perm((int64x2_t)in, (int64x2_t)vec_splats((uint64_t)0), (uint8x16_t)vec_splat_u8(1));
 }
 
 static really_inline m128 high64from128(const m128 in) {
-    //u64a ALIGN_ATTR(16) a[2];
-    //vec_xst((uint64x2_t) in, 0, a);
-    //return a[0];
-    // #warning FIXME
-    return vec_add(in, in);
+    return (m128) vec_perm((int64x2_t)in, (int64x2_t)vec_splats((uint64_t)0), (uint8x16_t)vec_splat_u8(0));
 }
 
 

diff --git a/src/util/supervector/arch/ppc64el/impl.cpp b/src/util/supervector/arch/ppc64el/impl.cpp
index 8628c66218f6cdbd4137e7d72f0e76fca7d6f7a4..93cc4d632db79f266f030a806f80fd966372fc46 100644 (file)
--- a/src/util/supervector/arch/ppc64el/impl.cpp
+++ b/src/util/supervector/arch/ppc64el/impl.cpp
@@ -186,29 +186,25 @@ really_inline SuperVector<16> SuperVector<16>::operator!=(SuperVector<16> const
 template <>
 really_inline SuperVector<16> SuperVector<16>::operator>(SuperVector<16> const &b) const
 { 
-    int32x4_t v = {u.s32[0] > b.u.s32[0], u.s32[1] > b.u.s32[1], u.s32[2] > b.u.s32[2], u.s32[3] > b.u.s32[3]};
-    return (m128) v; 
+    return {(m128) vec_cmpgt(u.v128[0], b.u.v128[0])}; 
 }
 
 template <>
 really_inline SuperVector<16> SuperVector<16>::operator>=(SuperVector<16> const &b) const
 {
-    int32x4_t v = {u.s32[0] >= b.u.s32[0], u.s32[1] >= b.u.s32[1], u.s32[2] >= b.u.s32[2], u.s32[3] >= b.u.s32[3]};
-    return (m128) v; 
+    return {(m128) vec_cmpge(u.v128[0], b.u.v128[0])};  
 }
 
 template <>
 really_inline SuperVector<16> SuperVector<16>::operator<(SuperVector<16> const &b) const
 {
-    int32x4_t v = {u.s32[0] < b.u.s32[0], u.s32[1] < b.u.s32[1], u.s32[2] < b.u.s32[2], u.s32[3] < b.u.s32[3]};
-    return (m128) v; 
+    return {(m128) vec_cmpgt(b.u.v128[0], u.v128[0])};  
 }
 
 template <>
 really_inline SuperVector<16> SuperVector<16>::operator<=(SuperVector<16> const &b) const
 {   
-    int32x4_t v = {u.s32[0] <= b.u.s32[0], u.s32[1] <= b.u.s32[1], u.s32[2] <= b.u.s32[2], u.s32[3] <= b.u.s32[3]};
-    return (m128) v; 
+    return {(m128) vec_cmpge(b.u.v128[0], u.v128[0])};   
 }
 
 
@@ -222,9 +218,21 @@ template <>
 really_inline typename SuperVector<16>::movemask_type SuperVector<16>::movemask(void)const
 { 
     uint8x16_t s1 = vec_sr((uint8x16_t)u.v128[0], vec_splat_u8(7));
+    //printf("s1:");
+    //for(int i=15; i>=0; i--) {printf("%02x, ",s1[i]);}
+    //printf("\n");
     uint16x8_t ss = vec_sr((uint16x8_t)s1, vec_splat_u16(7));
+    //printf("ss:");
+    //for(int i=7; i>=0; i--) {printf("%04x, ",ss[i]);}
+    //printf("\n");
     uint16x8_t res_and = vec_and((uint16x8_t)s1, vec_splats((uint16_t)0xff));
+    //printf("res_and:");
+    //for(int i=7; i>=0; i--) {printf("%04x, ",res_and[i]);}
+    //printf("\n");
     uint16x8_t s2 = vec_or((uint16x8_t)ss, res_and);
+    //printf("s2:");
+    //for(int i=7; i>=0; i--) {printf("%04x, ",s2[i]);}
+    //printf("\n");
 
     uint32x4_t ss2 = vec_sr((uint32x4_t)s2 , vec_splat_u32(14));
     uint32x4_t res_and2 = vec_and((uint32x4_t)s2, vec_splats((uint32_t)0xff));
@@ -238,6 +246,9 @@ really_inline typename SuperVector<16>::movemask_type SuperVector<16>::movemask(
     uint64x2_t res_and4 = vec_and((uint64x2_t)s4, vec_splats((uint64_t)0xff));
     uint64x2_t s5 = vec_or((uint64x2_t)ss4, res_and4);
     
+    //printf("s5:");
+    //for(int i=1; i>=0; i--) {printf("%016llx, ",s5[i]);}
+    //printf("\n");
     return s5[0];
 }