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

Improve bitselect codegen with knowledge of operand origin (#1783)

Browse Source 
* Encode vselect using BLEND instructions on x86

* Legalize vselect to bitselect

* Optimize bitselect to vselect for some operands

* Add run tests for bitselect-vselect optimization

* Address review feedback
This commit is contained in:
teapotd
2020-05-30 04:53:11 +02:00
committed by GitHub
parent 16afca4451
commit e430984ac4
11 changed files with 341 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
cranelift/codegen/meta/src/isa/x86/opcodes.rs
View File

@@ -54,6 +54,14 @@ pub static BIT_SCAN_FORWARD: [u8; 2] = [0x0f, 0xbc];
/// Bit scan reverse (stores index of first encountered 1 from the back).
pub static BIT_SCAN_REVERSE: [u8; 2] = [0x0f, 0xbd];
/// Select packed single-precision floating-point values from xmm1 and xmm2/m128
/// from mask specified in XMM0 and store the values into xmm1 (SSE4.1).
pub static BLENDVPS: [u8; 4] = [0x66, 0x0f, 0x38, 0x14];
/// Select packed double-precision floating-point values from xmm1 and xmm2/m128
/// from mask specified in XMM0 and store the values into xmm1 (SSE4.1).
pub static BLENDVPD: [u8; 4] = [0x66, 0x0f, 0x38, 0x15];
/// Call near, relative, displacement relative to next instruction (sign-extended).
pub static CALL_RELATIVE: [u8; 1] = [0xe8];
@@ -335,6 +343,10 @@ pub static PAVGB: [u8; 3] = [0x66, 0x0f, 0xE0];
/// Average packed unsigned word integers from xmm2/m128 and xmm1 with rounding (SSE2).
pub static PAVGW: [u8; 3] = [0x66, 0x0f, 0xE3];
/// Select byte values from xmm1 and xmm2/m128 from mask specified in the high bit of each byte
/// in XMM0 and store the values into xmm1 (SSE4.1).
pub static PBLENDVB: [u8; 4] = [0x66, 0x0f, 0x38, 0x10];
/// Compare packed data for equal (SSE2).
pub static PCMPEQB: [u8; 3] = [0x66, 0x0f, 0x74];