ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[CPU] Remove intrinsics from dot product constant propagation

This commit is contained in:
Triang3l 2022-07-06 21:32:56 +03:00
parent 326e718035
commit 6852e54937
1 changed files with 16 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
src/xenia/cpu/hir/value.cc
View File

@ -1384,14 +1384,17 @@ void Value::DotProduct3(Value* other) {
assert_true(this->type == VEC128_TYPE && other->type == VEC128_TYPE); assert_true(this->type == VEC128_TYPE && other->type == VEC128_TYPE);
switch (type) { switch (type) {
case VEC128_TYPE: { case VEC128_TYPE: {
alignas(16) float result[4];
__m128 src1 = _mm_load_ps(constant.v128.f32);
__m128 src2 = _mm_load_ps(other->constant.v128.f32);
__m128 dest = _mm_dp_ps(src1, src2, 0b01110001);
_mm_store_ps(result, dest);
// TODO(rick): is this sane? // TODO(rick): is this sane?
type = FLOAT32_TYPE; type = FLOAT32_TYPE;
constant.f32 = result[0]; // Using x86 DPPS ordering for consistency with x86-64 code generation:
// (X1 * X2 + Y1 * Y2) + (Z1 * Z2 + 0.0f)
// (+ 0.0f for zero sign, as zero imm8[4:7] bits result in zero terms,
// not in complete exclusion of them)
// TODO(Triang3l): NaN on overflow.
constant.f32 =
(constant.v128.f32[0] * other->constant.v128.f32[0] +
constant.v128.f32[1] * other->constant.v128.f32[1]) +
(constant.v128.f32[2] * other->constant.v128.f32[2] + 0.0f);
} break; } break;
default: default:
assert_unhandled_case(type); assert_unhandled_case(type);
@ -1403,14 +1406,15 @@ void Value::DotProduct4(Value* other) {
assert_true(this->type == VEC128_TYPE && other->type == VEC128_TYPE); assert_true(this->type == VEC128_TYPE && other->type == VEC128_TYPE);
switch (type) { switch (type) {
case VEC128_TYPE: { case VEC128_TYPE: {
alignas(16) float result[4];
__m128 src1 = _mm_load_ps(constant.v128.f32);
__m128 src2 = _mm_load_ps(other->constant.v128.f32);
__m128 dest = _mm_dp_ps(src1, src2, 0b11110001);
_mm_store_ps(result, dest);
// TODO(rick): is this sane? // TODO(rick): is this sane?
type = FLOAT32_TYPE; type = FLOAT32_TYPE;
constant.f32 = result[0]; // Using x86 DPPS ordering for consistency with x86-64 code generation:
// (X1 * X2 + Y1 * Y2) + (Z1 * Z2 + W1 * W2)
// TODO(Triang3l): NaN on overflow.
constant.f32 = (constant.v128.f32[0] * other->constant.v128.f32[0] +
constant.v128.f32[1] * other->constant.v128.f32[1]) +
(constant.v128.f32[2] * other->constant.v128.f32[2] +
constant.v128.f32[3] * other->constant.v128.f32[3]);
} break; } break;
default: default:
assert_unhandled_case(type); assert_unhandled_case(type);