T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[machinst x64]: refactor using added RegMem::from(Writable<Reg>)

Browse Source
This commit is contained in:
Andrew Brown
2020-09-23 09:24:06 -07:00
parent e3eb098c99
commit f50d905152
2 changed files with 21 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
cranelift/codegen/src/isa/x64/inst/args.rs
View File

@@ -5,7 +5,7 @@ use super::EmitState;
use crate::ir::condcodes::{FloatCC, IntCC}; use crate::ir::condcodes::{FloatCC, IntCC};
use crate::machinst::*; use crate::machinst::*;
use core::fmt::Debug; use core::fmt::Debug;
use regalloc::{RealRegUniverse, Reg, RegClass, RegUsageCollector, RegUsageMapper}; use regalloc::{RealRegUniverse, Reg, RegClass, RegUsageCollector, RegUsageMapper, Writable};
use std::fmt; use std::fmt;
use std::string::{String, ToString}; use std::string::{String, ToString};
@@ -265,6 +265,12 @@ impl RegMem {
} }
} }
impl From<Writable<Reg>> for RegMem {
fn from(r: Writable<Reg>) -> Self {
RegMem::reg(r.to_reg())
}
}
impl ShowWithRRU for RegMem { impl ShowWithRRU for RegMem {
fn show_rru(&self, mb_rru: Option<&RealRegUniverse>) -> String { fn show_rru(&self, mb_rru: Option<&RealRegUniverse>) -> String {
self.show_rru_sized(mb_rru, 8) self.show_rru_sized(mb_rru, 8)

26
cranelift/codegen/src/isa/x64/lower.rs
View File

@@ -2685,8 +2685,7 @@ fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
// After loading the constructed mask in a temporary register, we use this to // After loading the constructed mask in a temporary register, we use this to
// shuffle the `dst` register (remember that, in this case, it is the same as // shuffle the `dst` register (remember that, in this case, it is the same as
// `src` so we disregard this register). // `src` so we disregard this register).
let tmp = RegMem::reg(tmp.to_reg()); ctx.emit(Inst::xmm_rm_r(SseOpcode::Pshufb, RegMem::from(tmp), dst));
ctx.emit(Inst::xmm_rm_r(SseOpcode::Pshufb, tmp, dst));
} else { } else {
// If `lhs` and `rhs` are different, we must shuffle each separately and then OR // If `lhs` and `rhs` are different, we must shuffle each separately and then OR
// them together. This is necessary due to PSHUFB semantics. As in the case above, // them together. This is necessary due to PSHUFB semantics. As in the case above,
@@ -2698,8 +2697,7 @@ fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
let constructed_mask = mask.iter().cloned().map(zero_unknown_lane_index).collect(); let constructed_mask = mask.iter().cloned().map(zero_unknown_lane_index).collect();
let tmp1 = ctx.alloc_tmp(RegClass::V128, types::I8X16); let tmp1 = ctx.alloc_tmp(RegClass::V128, types::I8X16);
ctx.emit(Inst::xmm_load_const_seq(constructed_mask, tmp1, ty)); ctx.emit(Inst::xmm_load_const_seq(constructed_mask, tmp1, ty));
let tmp1 = RegMem::reg(tmp1.to_reg()); ctx.emit(Inst::xmm_rm_r(SseOpcode::Pshufb, RegMem::from(tmp1), tmp0));
ctx.emit(Inst::xmm_rm_r(SseOpcode::Pshufb, tmp1, tmp0));
// PSHUFB the second argument, placing zeroes for unused lanes. // PSHUFB the second argument, placing zeroes for unused lanes.
let constructed_mask = mask let constructed_mask = mask
@@ -2709,13 +2707,11 @@ fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
.collect(); .collect();
let tmp2 = ctx.alloc_tmp(RegClass::V128, types::I8X16); let tmp2 = ctx.alloc_tmp(RegClass::V128, types::I8X16);
ctx.emit(Inst::xmm_load_const_seq(constructed_mask, tmp2, ty)); ctx.emit(Inst::xmm_load_const_seq(constructed_mask, tmp2, ty));
let tmp2 = RegMem::reg(tmp2.to_reg()); ctx.emit(Inst::xmm_rm_r(SseOpcode::Pshufb, RegMem::from(tmp2), dst));
ctx.emit(Inst::xmm_rm_r(SseOpcode::Pshufb, tmp2, dst));
// OR the shuffled registers (the mechanism and lane-size for OR-ing the registers // OR the shuffled registers (the mechanism and lane-size for OR-ing the registers
// is not important). // is not important).
let tmp0 = RegMem::reg(tmp0.to_reg()); ctx.emit(Inst::xmm_rm_r(SseOpcode::Orps, RegMem::from(tmp0), dst));
ctx.emit(Inst::xmm_rm_r(SseOpcode::Orps, tmp0, dst));
// TODO when AVX512 is enabled we should replace this sequence with a single VPERMB // TODO when AVX512 is enabled we should replace this sequence with a single VPERMB
} }
@@ -2744,13 +2740,19 @@ fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
ctx.emit(Inst::xmm_load_const_seq(zero_mask_value, zero_mask, ty)); ctx.emit(Inst::xmm_load_const_seq(zero_mask_value, zero_mask, ty));
// Use the `zero_mask` on a writable `swizzle_mask`. // Use the `zero_mask` on a writable `swizzle_mask`.
let zero_mask = RegMem::reg(zero_mask.to_reg());
let swizzle_mask = Writable::from_reg(swizzle_mask); let swizzle_mask = Writable::from_reg(swizzle_mask);
ctx.emit(Inst::xmm_rm_r(SseOpcode::Paddusb, zero_mask, swizzle_mask)); ctx.emit(Inst::xmm_rm_r(
SseOpcode::Paddusb,
RegMem::from(zero_mask),
swizzle_mask,
));
// Shuffle `dst` using the fixed-up `swizzle_mask`. // Shuffle `dst` using the fixed-up `swizzle_mask`.
let swizzle_mask = RegMem::reg(swizzle_mask.to_reg()); ctx.emit(Inst::xmm_rm_r(
ctx.emit(Inst::xmm_rm_r(SseOpcode::Pshufb, swizzle_mask, dst)); SseOpcode::Pshufb,
RegMem::from(swizzle_mask),
dst,
));
} }
Opcode::Insertlane => { Opcode::Insertlane => {