Adding vector add tests and fixing bugs found (still broken).

This commit is contained in:
Ben Vanik
2014-08-23 22:09:30 -07:00
parent 96007049d2
commit 28bae464c2
4 changed files with 297 additions and 3 deletions

View File

@@ -2630,7 +2630,7 @@ uint32_t Translate_VECTOR_ADD_I32(IntCodeState& ics, const IntCode* i) {
if (arithmetic_flags & ARITHMETIC_SATURATE) {
if (arithmetic_flags & ARITHMETIC_UNSIGNED) {
for (int n = 0; n < 4; n++) {
uint64_t v = VECI4(src1, n) + VECI4(src2, n);
uint64_t v = (uint64_t)VECI4(src1, n) + (uint64_t)VECI4(src2, n);
if (v > 0xFFFFFFFF) {
VECI4(dest, n) = 0xFFFFFFFF;
ics.did_saturate = 1;
@@ -2640,7 +2640,7 @@ uint32_t Translate_VECTOR_ADD_I32(IntCodeState& ics, const IntCode* i) {
}
} else {
for (int n = 0; n < 4; n++) {
int64_t v = (int32_t)VECI4(src1, n) + (int32_t)VECI4(src2, n);
int64_t v = (int64_t)(int32_t)VECI4(src1, n) + (int64_t)(int32_t)VECI4(src2, n);
if (v > 0x7FFFFFFF) {
VECI4(dest, n) = 0x7FFFFFFF;
ics.did_saturate = 1;

View File

@@ -2749,7 +2749,7 @@ EMITTER(VECTOR_ADD, MATCH(I<OPCODE_VECTOR_ADD, V128<>, V128<>, V128<>>)) {
e.vblendvps(e.xmm2, e.xmm0, e.GetXmmConstPtr(XMMSignMaskPS), e.xmm2);
e.vpor(e.xmm1, src1, src2); // sign_or
e.vpandn(e.xmm1, e.xmm0); // max_sat_mask
e.vblendvps(e.xmm2, e.GetXmmConstPtr(XMMAbsMaskPS), e.xmm1);
e.vblendvps(dest, e.GetXmmConstPtr(XMMAbsMaskPS), e.xmm1);
}
} else {
e.vpaddd(dest, src1, src2);

View File

@@ -42,6 +42,13 @@ typedef struct alignas(16) vec128_s {
return low == b.low && high == b.high;
}
} vec128_t;
static inline vec128_t vec128i(uint32_t src) {
vec128_t v;
for (auto i = 0; i < 4; ++i) {
v.i4[i] = src;
}
return v;
}
static inline vec128_t vec128i(uint32_t x, uint32_t y, uint32_t z, uint32_t w) {
vec128_t v;
v.i4[0] = x;
@@ -50,6 +57,13 @@ static inline vec128_t vec128i(uint32_t x, uint32_t y, uint32_t z, uint32_t w) {
v.i4[3] = w;
return v;
}
static inline vec128_t vec128f(float src) {
vec128_t v;
for (auto i = 0; i < 4; ++i) {
v.f4[i] = src;
}
return v;
}
static inline vec128_t vec128f(float x, float y, float z, float w) {
vec128_t v;
v.f4[0] = x;