Improved and enhanced test17 (SSE testing)

4 files changed, 144 insertions, 95 deletions

diff --git a/src/emu/x64run660f.c b/src/emu/x64run660f.c
index 0f7f252b..5d4463bd 100644
--- a/src/emu/x64run660f.c
+++ b/src/emu/x64run660f.c
@@ -367,7 +367,7 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 GETEX(0);

                 GETGX;

                 for (int i=0; i<16; ++i) {

-                    GX->sb[i] = abs(EX->sb[i]);

+                    GX->ub[i] = abs(EX->sb[i]);

                 }

                 break;

             case 0x1D:  /* PABSW Gx, Ex */

@@ -375,7 +375,7 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 GETEX(0);

                 GETGX;

                 for (int i=0; i<8; ++i) {

-                    GX->sw[i] = abs(EX->sw[i]);

+                    GX->uw[i] = abs(EX->sw[i]);

                 }

                 break;

             case 0x1E:  /* PABSD Gx, Ex */

@@ -383,7 +383,7 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 GETEX(0);

                 GETGX;

                 for (int i=0; i<4; ++i) {

-                    GX->sd[i] = abs(EX->sd[i]);

+                    GX->ud[i] = abs(EX->sd[i]);

                 }

                 break;

 

diff --git a/tests/ref17.txt b/tests/ref17.txt
index 9715adf0..2bae8c46 100644
--- a/tests/ref17.txt
+++ b/tests/ref17.txt
@@ -1,17 +1,17 @@
-ucomiss 1.000000, 2.000000 => 0x202
-ucomiss 2.000000, 1.000000 => 0x203
-ucomiss 1.000000, inf => 0x202
-ucomiss inf, 1.000000 => 0x203
-ucomiss 1.000000, -inf => 0x203
-ucomiss -inf, 1.000000 => 0x202
-ucomiss 1.000000, nan => 0x247
-ucomiss nan, 1.000000 => 0x247
+ucomiss 1.000000, 2.000000 => 0x203
+ucomiss 2.000000, 1.000000 => 0x202
+ucomiss 1.000000, inf => 0x203
+ucomiss inf, 1.000000 => 0x202
+ucomiss 1.000000, -inf => 0x202
+ucomiss -inf, 1.000000 => 0x203
+ucomiss 1.000000, nan => 0x203
+ucomiss nan, 1.000000 => 0x203
 ucomiss 1.000000, 1.000000 => 0x242
 ucomiss 1.000000, 1.000000 => 0x242
 ucomiss inf, inf => 0x242
-ucomiss -inf, inf => 0x202
-ucomiss inf, -inf => 0x203
-ucomiss nan, nan => 0x247
+ucomiss -inf, inf => 0x203
+ucomiss inf, -inf => 0x202
+ucomiss nan, nan => 0x203
 minss 1, 2 => 1
 minss 2, 1 => 1
 minss -inf, 2 => -inf
@@ -152,3 +152,9 @@ cmpss 7 inf, inf => 0xffffffff
 cmpss 7 -inf, inf => 0xffffffff
 cmpss 7 inf, -inf => 0xffffffff
 cmpss 7 nan, nan => 0x0
+pabsb(0xff 0x80 0x7f 0x0 0x1 0x2 0x3 0x81 0xfe 0x84 0x72 0x52 0xa5 0x0 0xc0 0x32 ) = 0x1 0x80 0x7f 0x0 0x1 0x2 0x3 0x7f 0x2 0x7c 0x72 0x52 0x5b 0x0 0x40 0x32 
+pabsw(0xffff 0x8000 0x7fff 0x0 0x1 0x2 0x3 0x8001 ) = 0x1 0x8000 0x7fff 0x0 0x1 0x2 0x3 0x7fff 
+pabsd(0xffffffff 0x80000000 0x7fffffff 0x0 ) = 0x1 0x80000000 0x7fffffff 0x0 
+pshuffleb(0xff 0x80 0x7f 0x0 0x1 0x2 0x3 0x81 0xfe 0x84 0x72 0x52 0xa5 0x0 0xc0 0x32 , 0x0 0x1 0x5 0x15 0x20 0x80 0xff 0x0 0x8 0x7 0x81 0x6 0xa 0xf 0x10 0x1 ) = 0xff 0x80 0x2 0x2 0xff 0x0 0x0 0xff 0xfe 0x81 0x0 0x3 0x72 0x32 0xff 0x80 
+phaddw(0xffff 0x8000 0x7fff 0x0 0x1 0x2 0x3 0x8001 , 0x8000 0x7fff 0xffff 0xffff 0x50 0x9000 0xfffe 0x8001 ) = 0x7fff 0x7fff 0x3 0x8004 0xffff 0xfffe 0x9050 0x7fff 
+phaddd(0xffffffff 0x80000000 0x7fffffff 0x0 , 0x1 0x80000000 0x5 0xfffffffe ) = 0x7fffffff 0x7fffffff 0x80000001 0x3 
diff --git a/tests/test17 b/tests/test17
index e56e6a46..0d0b89e4 100755
--- a/tests/test17
+++ b/tests/test17
Binary files differdiff --git a/tests/test17.c b/tests/test17.c
index 735e861d..e0c07685 100644
--- a/tests/test17.c
+++ b/tests/test17.c
@@ -4,102 +4,122 @@
 #include <stdlib.h>
 #include <stdint.h>
 #include <math.h>
+#include <pmmintrin.h>
+#include <immintrin.h> 
+
+typedef unsigned char u8x16 __attribute__ ((vector_size (16)));
+typedef unsigned short u16x8 __attribute__ ((vector_size (16)));
+typedef unsigned int  u32x4 __attribute__ ((vector_size (16)));
+typedef unsigned long int  u64x2 __attribute__ ((vector_size (16)));
+typedef float  f32x4 __attribute__ ((vector_size (16)));
+typedef double d64x2 __attribute__ ((vector_size (16)));
+
+typedef union {
+        __m128i mm;
+        __m128  mf;
+        __m128d md;
+        u8x16   u8;
+        u16x8   u16;
+        u32x4   u32;
+        u64x2   u64;
+        f32x4   f32;
+        d64x2   d64;
+} v128;
 
-#if defined(__x86_64__)
 uint64_t _ucomiss_(float a, float b)
 {
-    uint64_t ret;
-    asm volatile (
-    "ucomiss %%xmm0, %%xmm1\n"
-    "pushf\n"
-    "pop %%rax"
-    :"=a" (ret)::"xmm0","xmm1","cc");
+    uint64_t ret = 0x202;
+    v128 va, vb;
+    va.f32[0] = a;
+    vb.f32[0] = b;
+    if(_mm_ucomigt_ss(va.mf, vb.mf))
+        ret |= 0x000;
+    else if(_mm_ucomilt_ss(va.mf, vb.mf))
+        ret |= 0x001;
+    else if(_mm_ucomieq_ss(va.mf, vb.mf))
+        ret |= 0x040;
+    else
+        ret |= 0x045;
     return ret;
 }
+
 uint64_t _minss_(float a, float b)
 {
-    uint64_t ret;
-    asm volatile (
-    "minss %%xmm1, %%xmm0\n"
-    "movd %%xmm0, %%eax"
-    :"=a" (ret)::"xmm0","xmm1","cc");
-    return ret;
+    v128 va, vb, ret;
+    va.f32[0] = a;
+    vb.f32[0] = b;
+    ret.mf = _mm_min_ss(va.mf, vb.mf);
+    return ret.u64[0];
 }
 uint64_t _maxss_(float a, float b)
 {
-    uint64_t ret;
-    asm volatile (
-    "maxss %%xmm1, %%xmm0\n"
-    "movd %%xmm0, %%eax"
-    :"=a" (ret)::"xmm0","xmm1","cc");
-    return ret;
+    v128 va, vb, ret;
+    va.f32[0] = a;
+    vb.f32[0] = b;
+    ret.mf = _mm_max_ss(va.mf, vb.mf);
+    return ret.u64[0];
 }
-#define CMPSS(A)                        \
-uint64_t _cmpss_##A(float a, float b)   \
-{                                       \
-    uint64_t ret;                       \
-    asm volatile (                      \
-    "cmpss $" #A ", %%xmm1, %%xmm0\n"   \
-    "movd %%xmm0, %%eax"                \
-    :"=a" (ret)::"xmm0","xmm1","cc");   \
-    return ret;                         \
+
+#define CMPSS(A, B)                         \
+uint64_t _cmpss_##A(float a, float b)       \
+{                                           \
+    v128 va, vb, ret;                       \
+    va.f32[0] = a;                          \
+    vb.f32[0] = b;                          \
+    ret.mf = _mm_cmp##B##_ss(va.mf, vb.mf); \
+    return ret.u64[0];                      \
 }
-#else
-uint64_t _ucomiss_(float a, float b)
-{
-    uint32_t ret;
-    asm volatile (
-    "movss %1, %%xmm0\n"
-    "movss %2, %%xmm1\n"
-    "ucomiss %%xmm0, %%xmm1\n"
-    "pushf\n"
-    "pop %%eax"
-    :"=a" (ret):"m"(a), "m"(b):"xmm0", "xmm1", "cc");
-    return ret;
+CMPSS(0, eq)
+CMPSS(1, lt)
+CMPSS(2, le)
+CMPSS(3, unord)
+CMPSS(4, neq)
+CMPSS(5, nlt)
+CMPSS(6, nle)
+CMPSS(7, ord)
+#undef CMPSS
+
+const v128 a128_8 = {.u8 = {
+    0xff, 0x80, 0x7f, 0x00, 0x01, 0x02, 0x03, 0x81,
+    0xfe, 0x84, 0x72, 0x52, 0xa5, 0x00, 0xc0, 0x32
+}};
+const v128 a128_16 = {.u16 = {
+    0xffff, 0x8000, 0x7fff, 0x0000, 0x0001, 0x0002, 0x0003, 0x8001
+}};
+const v128 a128_32 = {.u32 = {
+    0xffffffff, 0x80000000, 0x7fffffff, 0x00000000
+}};
+const v128 a128_64 = {.u64 = {
+    0xffffffffffffffffLL, 0x8000000000000000LL
+}};
+
+const v128 b128_8 = {.u8 = {
+    0x00, 0x01, 0x05, 0x15, 0x20, 0x80, 0xff, 0x00,
+    0x08, 0x07, 0x81, 0x06, 0x0a, 0x0f, 0x10, 0x01
+}};
+const v128 b128_16 = {.u16 = {
+    0x8000, 0x7fff, 0xffff, 0xffff, 0x0050, 0x9000, 0xfffe, 0x8001
+}};
+const v128 b128_32 = {.u32 = {
+    0x00000001, 0x80000000, 0x00000005, 0xfffffffe
+}};
+
+void print_8(v128 v) {
+    for(int i=0; i<16; ++i)
+        printf("0x%x ", v.u8[i]);
 }
-uint64_t _minss_(float a, float b)
-{
-    uint32_t ret;
-    asm volatile (
-    "movss %1, %%xmm0\n"
-    "movss %2, %%xmm1\n"
-    "minss %%xmm1, %%xmm0\n"
-    "movd %%xmm0, %%eax"
-    :"=a" (ret):"m"(a), "m"(b):"xmm0", "xmm1", "cc");
-    return ret;
+void print_16(v128 v) {
+    for(int i=0; i<8; ++i)
+        printf("0x%x ", v.u16[i]);
 }
-uint64_t _maxss_(float a, float b)
-{
-    uint32_t ret;
-    asm volatile (
-    "movss %1, %%xmm0\n"
-    "movss %2, %%xmm1\n"
-    "maxss %%xmm1, %%xmm0\n"
-    "movd %%xmm0, %%eax"
-    :"=a" (ret):"m"(a), "m"(b):"xmm0", "xmm1", "cc");
-    return ret;
+void print_32(v128 v) {
+    for(int i=0; i<4; ++i)
+        printf("0x%x ", v.u32[i]);
 }
-#define CMPSS(A)                                            \
-uint64_t _cmpss_##A(float a, float b)                       \
-{                                                           \
-    uint32_t ret;                                           \
-    asm volatile (                                          \
-    "movss %1, %%xmm0\n"                                    \
-    "movss %2, %%xmm1\n"                                    \
-    "cmpss $" #A ", %%xmm1, %%xmm0\n"                       \
-    "movd %%xmm0, %%eax"                                    \
-    :"=a" (ret):"m"(a), "m"(b):"xmm0", "xmm1", "cc");       \
-    return ret;                                             \
+void print_64(v128 v) {
+    for(int i=0; i<2; ++i)
+        printf("0x%llx ", v.u64[i]);
 }
-#endif
-CMPSS(0)
-CMPSS(1)
-CMPSS(2)
-CMPSS(3)
-CMPSS(4)
-CMPSS(5)
-CMPSS(6)
-CMPSS(7)
 
 int main(int argc, const char** argv)
 {
@@ -195,6 +215,29 @@ printf(N " %g, %g => %g\n", b, a, *(float*)&r);
  GO1(_cmpss_5, "cmpss 5")
  GO1(_cmpss_6, "cmpss 6")
  GO1(_cmpss_7, "cmpss 7")
- 
+
+ #undef GO1
+ #undef GO2
+ v128 a128;
+
+ #define GO1(A, N, C)                               \
+ a128.mm = _mm_##A##_epi##N(a128_##N.mm);           \
+ printf("p%s%s(", #A, #C); print_##N(a128_##N);     \
+ printf(") = "); print_##N(a128); printf("\n");
+ #define GO2(A, N, C, A1, A2)                       \
+ a128.mm = _mm_##A##_epi##N(A1.mm, A2.mm);          \
+ printf("p%s%s(", #A, #C); print_##N(A1);           \
+ printf(", "); print_##N(A2);                       \
+ printf(") = "); print_##N(a128); printf("\n");
+
+
+ GO1(abs, 8, b)
+ GO1(abs, 16, w)
+ GO1(abs, 32, d)
+ GO2(shuffle, 8, b, a128_8, b128_8)
+ GO2(hadd, 16, w, a128_16, b128_16)
+ GO2(hadd, 32, d, a128_32, b128_32)
+
+
  return 0;
 }