xenia-project/xenia
1
0
Fork 0
mirror of https://github.com/xenia-project/xenia.git synced 2024-12-11 13:25:52 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Untested VECTOR_SHA 16 variable and VECTOR_SHL 16 variable.

Browse Source
This commit is contained in:
Ben Vanik 2014-08-05 11:49:59 -07:00
parent 810c256f21
commit 07afc58264
3 changed files with 37 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/alloy/backend/x64/x64_emitter.cc
View File

@ -539,6 +539,10 @@ Address X64Emitter::GetXmmConstPtr(XmmConst id) {
0xFFFFFF0Cu, 0xFFFFFF0Fu),
/* XMMOneOver255 */ vec128f(1.0f / 255.0f, 1.0f / 255.0f,
1.0f / 255.0f, 1.0f / 255.0f),
/* XMMMaskEvenPI16 */ vec128i(0x0000FFFFu, 0x0000FFFFu,
0x0000FFFFu, 0x0000FFFFu),
/* XMMShiftMaskEvenPI16 */ vec128i(0x0000000Fu, 0x0000000Fu,
0x0000000Fu, 0x0000000Fu),
/* XMMShiftMaskPS */ vec128i(0x0000001Fu, 0x0000001Fu,
0x0000001Fu, 0x0000001Fu),
/* XMMShiftByteMask */ vec128i(0x000000FFu, 0x000000FFu,

2
src/alloy/backend/x64/x64_emitter.h
View File

@ -54,6 +54,8 @@ enum XmmConst {
XMMPackD3DCOLOR,
XMMUnpackD3DCOLOR,
XMMOneOver255,
XMMMaskEvenPI16,
XMMShiftMaskEvenPI16,
XMMShiftMaskPS,
XMMShiftByteMask,
XMMUnsignedDwordMax,

32
src/alloy/backend/x64/x64_sequences.cc
View File

@ -4150,7 +4150,23 @@ EMITTER(VECTOR_SHL_V128, MATCH(I<OPCODE_VECTOR_SHL, V128<>, V128<>, V128<>>)) {
}
} else {
// Fully variable shift.
assert_always();
// TODO(benvanik): find a better sequence.
Xmm temp = i.dest;
if (i.dest == i.src1 || i.dest == i.src2) {
temp = e.xmm2;
}
// Even:
e.vpand(e.xmm0, i.src2, e.GetXmmConstPtr(XMMShiftMaskEvenPI16));
e.vpsllvd(e.xmm1, i.src1, e.xmm0);
e.vpand(e.xmm1, e.GetXmmConstPtr(XMMMaskEvenPI16));
// Odd:
e.vpsrld(e.xmm0, i.src2, 16);
e.vpand(e.xmm0, e.GetXmmConstPtr(XMMShiftMaskEvenPI16));
e.vpsrld(i.dest, i.src1, 16);
e.vpsllvd(i.dest, i.dest, e.xmm0);
e.vpslld(i.dest, 8);
// Merge:
e.vpor(i.dest, e.xmm1);
}
}
static void EmitInt32(X64Emitter& e, const EmitArgType& i) {
@ -4308,6 +4324,20 @@ EMITTER_OPCODE_TABLE(
EMITTER(VECTOR_SHA_V128, MATCH(I<OPCODE_VECTOR_SHA, V128<>, V128<>, V128<>>)) {
static void Emit(X64Emitter& e, const EmitArgType& i) {
switch (i.instr->flags) {
case INT16_TYPE:
// Even halfwords:
e.vpand(e.xmm0, i.src2, e.GetXmmConstPtr(XMMShiftMaskEvenPI16));
e.vpslld(e.xmm1, i.src1, 16);
e.vpsrad(e.xmm1, 8);
e.vpsravd(e.xmm1, e.xmm1, e.xmm0);
// Odd halfwords:
e.vpsrld(e.xmm0, i.src2, 16);
e.vpand(e.xmm0, e.GetXmmConstPtr(XMMShiftMaskEvenPI16));
e.vpslld(i.dest, i.src1, 16);
e.vpsravd(i.dest, i.dest, e.xmm0);
// Merge:
e.vpor(i.dest, e.xmm1);
break;
case INT32_TYPE:
// src shift mask may have values >31, and x86 sets to zero when
// that happens so we mask.