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

x64: Refactor vector_all_ones, and remove buggy sse_cmp_op (#4728)

Browse Source 
The sse_cmp_op rule had cases that would produce SseOperand values that aren't legal to use with MInst.XmmRmR, and was only used in vector_all_ones when constructing an XmmRmR value. Additionally, vector_all_ones always called sse_cmp_op with the same type, so the other cases were redundant.

The solution in this PR is to remove sse_cmp_op entirely and inline a call to x64_pcmpeqd directly in vector_all_ones, and remove the unused argument from vector_all_ones.
This commit is contained in:
Trevor Elliott
2022-08-17 14:30:52 -07:00
committed by GitHub
parent 1481721c9d
commit 0a71df6a37
2 changed files with 18 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
cranelift/codegen/src/isa/x64/inst.isle
View File

@@ -1496,35 +1496,19 @@
(decl sse_xor (Type Xmm XmmMem) Xmm)
(rule (sse_xor ty x y) (xmm_rm_r ty (sse_xor_op ty) x y))
;; Determine the appropriate operation to compare two vectors of the specified
;; type.
(decl sse_cmp_op (Type) SseOpcode)
(rule (sse_cmp_op (multi_lane 8 16)) (SseOpcode.Pcmpeqb))
(rule (sse_cmp_op (multi_lane 16 8)) (SseOpcode.Pcmpeqw))
(rule (sse_cmp_op (multi_lane 32 4)) (SseOpcode.Pcmpeqd))
(rule (sse_cmp_op (multi_lane 64 2)) (SseOpcode.Pcmpeqq))
(rule (sse_cmp_op $F32X4) (SseOpcode.Cmpps))
(rule (sse_cmp_op $F64X2) (SseOpcode.Cmppd))
;; Generates a register value which has an all-ones pattern of the specified
;; type.
;; Generates a register value which has an all-ones pattern.
;;
;; Note that this is accomplished by comparing a fresh register with itself,
;; which for integers is always true. Also note that the comparison is always
;; done for integers, it doesn't actually take the input `ty` into account. This
;; is because we're comparing a fresh register to itself and we don't know the
;; previous contents of the register. If a floating-point comparison is used
;; then it runs the risk of comparing NaN against NaN and not actually producing
;; an all-ones mask. By using integer comparision operations we're guaranteeed
;; that everything is equal to itself.
(decl vector_all_ones (Type) Xmm)
(rule (vector_all_ones ty)
(let ((r WritableXmm (temp_writable_xmm))
(_ Unit (emit (MInst.XmmRmR (sse_cmp_op $I32X4)
r
r
r))))
r))
;; done for integers. This is because we're comparing a fresh register to itself
;; and we don't know the previous contents of the register. If a floating-point
;; comparison is used then it runs the risk of comparing NaN against NaN and not
;; actually producing an all-ones mask. By using integer comparision operations
;; we're guaranteeed that everything is equal to itself.
(decl vector_all_ones () Xmm)
(rule (vector_all_ones)
(let ((r WritableXmm (temp_writable_xmm)))
(x64_pcmpeqd r r)))
;; Helper for creating an SSE register holding an `i64x2` from two `i64` values.
(decl make_i64x2_from_lanes (GprMem GprMem) Xmm)