| Issue |
180777
|
| Summary |
[WebAssembly][FastISel] emits inefficient shift sequence for sext instead of extend instructions
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
ParkHanbum
|
Description
When using FastISel (-fast-isel) for the WebAssembly target, sext (sign-extension) operations from i8 or i16 to i32 are lowered into a sequence of bitwise shift instructions (shl followed by shr_s).
WebAssembly has dedicated instructions for this purpose: i32.extend8_s and i32.extend16_s. Using these instructions is more efficient and results in smaller code size compared to the generic shift sequence.
Reproduction Steps
Create a file named sext.ll with the following content:
```
target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
target triple = "wasm32-unknown-unknown"
define i32 @sext_i16_i32(ptr %p) {
%v = load atomic i16, ptr %p seq_cst, align 2
%e = sext i16 %v to i32
ret i32 %e
}
define i32 @sext_i8_i32(ptr %p) {
%v = load atomic i8, ptr %p seq_cst, align 1
%e = sext i8 %v to i32
ret i32 %e
}
```
Current Behavior (Output)
FastISel generates a shift sequence (Left Shift + Arithmetic Right Shift) to sign-extend the value.
```
sext_i16_i32: # @sext_i16_i32
.functype sext_i16_i32 (i32) -> (i32)
local.get 0
i32.load16_u 0
; inefficient codes ===
i32.const 16
i32.shl
i32.const 16
i32.shr_s
; ===============
end_function
sext_i8_i32: # @sext_i8_i32
.functype sext_i8_i32 (i32) -> (i32)
local.get 0
i32.load8_u 0
; inefficient codes ===
i32.const 24
i32.shl
i32.const 24
i32.shr_s
; ===============
end_function
```
Expected Behavior
FastISel should utilize i32.extend8_s and i32.extend16_s. These instructions correctly sign-extend the lower 8 or 16 bits regardless of the upper bits' content (which matches the behavior needed after i32.load8_u or i32.load16_u).
```
test_sext_i8:
local.get 0
i32.load8_u 0
i32.extend8_s # <--- Optimized
end_function
test_sext_i16:
local.get 0
i32.load16_u 0
i32.extend16_s # <--- Optimized
end_function
```
Additional Context
Target: WebAssembly (wasm32)
Component: WebAssemblyFastISel.cpp
Function: SelectSExt
Currently, SelectSExt seems to fall back to the default expansion behavior. It should explicitly handle i8 and i16 source types by emitting WebAssembly::I32_EXTEND8_S and WebAssembly::I32_EXTEND16_S.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
