danog/ir
1
0
Fork 0
mirror of https://github.com/danog/ir.git synced 2024-11-30 04:39:43 +01:00
Code Issues Packages Projects Releases Wiki Activity

Clear destination regeister before INT to FP conversion to avoid partial register stall

Browse Source
This commit is contained in:
Dmitry Stogov 2022-12-28 00:05:23 +03:00
parent cc8f3fe987
commit 54597bc862
3 changed files with 18 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
ir_x86.dasc
View File

@ -4595,15 +4595,19 @@ static void ir_emit_int2fp(ir_ctx *ctx, ir_ref def, ir_insn *insn)
if (!src64) {
if (dst_type == IR_DOUBLE) {
if (ctx->flags & IR_AVX) {
| vxorps xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| vcvtsi2sd xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), Rd(op1_reg)
} else {
| pxor xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| cvtsi2sd xmm(def_reg-IR_REG_FP_FIRST), Rd(op1_reg)
}
} else {
IR_ASSERT(dst_type == IR_FLOAT);
if (ctx->flags & IR_AVX) {
| vxorps xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| vcvtsi2ss xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), Rd(op1_reg)
} else {
| pxor xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| cvtsi2ss xmm(def_reg-IR_REG_FP_FIRST), Rd(op1_reg)
}
}
@ -4612,15 +4616,19 @@ static void ir_emit_int2fp(ir_ctx *ctx, ir_ref def, ir_insn *insn)
|.if X64
if (dst_type == IR_DOUBLE) {
if (ctx->flags & IR_AVX) {
| vxorps xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| vcvtsi2sd xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), Rq(op1_reg)
} else {
| pxor xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| cvtsi2sd xmm(def_reg-IR_REG_FP_FIRST), Rq(op1_reg)
}
} else {
IR_ASSERT(dst_type == IR_FLOAT);
if (ctx->flags & IR_AVX) {
| vxorps xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| vcvtsi2ss xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), Rq(op1_reg)
} else {
| pxor xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| cvtsi2ss xmm(def_reg-IR_REG_FP_FIRST), Rq(op1_reg)
}
}
@ -4639,15 +4647,19 @@ static void ir_emit_int2fp(ir_ctx *ctx, ir_ref def, ir_insn *insn)
if (!src64) {
if (dst_type == IR_DOUBLE) {
if (ctx->flags & IR_AVX) {
| vxorps xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| vcvtsi2sd xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), dword [Ra(op1_reg)+offset]
} else {
| pxor xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| cvtsi2sd xmm(def_reg-IR_REG_FP_FIRST), dword [Ra(op1_reg)+offset]
}
} else {
IR_ASSERT(dst_type == IR_FLOAT);
if (ctx->flags & IR_AVX) {
| vxorps xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| vcvtsi2ss xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), dword [Ra(op1_reg)+offset]
} else {
| pxor xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| cvtsi2ss xmm(def_reg-IR_REG_FP_FIRST), dword [Ra(op1_reg)+offset]
}
}
@ -4656,15 +4668,19 @@ static void ir_emit_int2fp(ir_ctx *ctx, ir_ref def, ir_insn *insn)
|.if X64
if (dst_type == IR_DOUBLE) {
if (ctx->flags & IR_AVX) {
| vxorps xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| vcvtsi2sd xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), qword [Ra(op1_reg)+offset]
} else {
| pxor xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| cvtsi2sd xmm(def_reg-IR_REG_FP_FIRST), qword [Ra(op1_reg)+offset]
}
} else {
IR_ASSERT(dst_type == IR_FLOAT);
if (ctx->flags & IR_AVX) {
| vxorps xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| vcvtsi2ss xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST), qword [Ra(op1_reg)+offset]
} else {
| pxor xmm(def_reg-IR_REG_FP_FIRST), xmm(def_reg-IR_REG_FP_FIRST)
| cvtsi2ss xmm(def_reg-IR_REG_FP_FIRST), qword [Ra(op1_reg)+offset]
}
}

1
tests/x86/conv_006.irt
View File

@ -15,6 +15,7 @@ x86
test:
subl $8, %esp
movl 0xc(%esp), %eax
pxor %xmm0, %xmm0
cvtsi2sdl %eax, %xmm0
movsd %xmm0, (%esp)
fldl (%esp)

1
tests/x86_64/conv_006.irt
View File

@ -13,5 +13,6 @@ x86_64
}
--EXPECT--
test:
pxor %xmm0, %xmm0
cvtsi2sdl %edi, %xmm0
retq