Skip to content

Winch: Add splat instructions to x64 using AVX2 #10028

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
saulecabrera merged 7 commits into from
Jan 16, 2025
Merged

Winch: Add splat instructions to x64 using AVX2 #10028

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
a2b025c
Winch: Add splat instructions to x64 using AVX2
jeffcharles Jan 15, 2025
df4e387
Change vpbroadcast size to lane size
jeffcharles Jan 16, 2025
a7afe14
Create helper for vpshuf mask
jeffcharles Jan 16, 2025
8033edb
Change to single masm method for splatting
jeffcharles Jan 16, 2025
8fb5384
Split out SplatLoadKind from SplatKind
jeffcharles Jan 16, 2025
0791c99
Add comment about heap allocation
jeffcharles Jan 16, 2025
3295415
Merge branch 'main' into winch-simd-splat
jeffcharles Jan 16, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions crates/wast-util/src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -502,6 +502,7 @@ impl WastTest {
if !(std::is_x86_feature_detected!("avx") && std::is_x86_feature_detected!("avx2")) {
let unsupported = [
"misc_testsuite/winch/_simd_lane.wast",
"misc_testsuite/winch/_simd_splat.wast",
"spec_testsuite/simd_align.wast",
];

Expand Down
32 changes: 32 additions & 0 deletions tests/disas/winch/x64/f32x4_splat/const_avx2.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,32 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx2" ]

(module
(func (result v128)
(f32x4.splat (f32.const 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x10, %r11
;; cmpq %rsp, %r11
;; ja 0x3f
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
;; movq %rsi, (%rsp)
;; movss 0x14(%rip), %xmm0
;; vpbroadcastd %xmm0, %xmm0
;; addq $0x10, %rsp
;; popq %rbp
;; retq
;; 3f: ud2
;; 41: addb %al, (%rax)
;; 43: addb %al, (%rax)
;; 45: addb %al, (%rax)
;; 47: addb %al, (%rax)
;; 49: addb %al, (%rax)
28 changes: 28 additions & 0 deletions tests/disas/winch/x64/f32x4_splat/params_avx2.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx2" ]

(module
(func (param f32) (result v128)
(f32x4.splat (local.get 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x20, %r11
;; cmpq %rsp, %r11
;; ja 0x44
;; 1c: movq %rdi, %r14
;; subq $0x20, %rsp
;; movq %rdi, 0x18(%rsp)
;; movq %rsi, 0x10(%rsp)
;; movss %xmm0, 0xc(%rsp)
;; movss 0xc(%rsp), %xmm0
;; vpbroadcastd %xmm0, %xmm0
;; addq $0x20, %rsp
;; popq %rbp
;; retq
;; 44: ud2
34 changes: 34 additions & 0 deletions tests/disas/winch/x64/f64x2_splat/const_avx.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,34 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx" ]

(module
(func (result v128)
(f64x2.splat (f64.const 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x10, %r11
;; cmpq %rsp, %r11
;; ja 0x3f
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
;; movq %rsi, (%rsp)
;; movsd 0x14(%rip), %xmm0
;; vpshufd $0x44, %xmm0, %xmm0
;; addq $0x10, %rsp
;; popq %rbp
;; retq
;; 3f: ud2
;; 41: addb %al, (%rax)
;; 43: addb %al, (%rax)
;; 45: addb %al, (%rax)
;; 47: addb %al, (%rax)
;; 49: addb %al, (%rax)
;; 4b: addb %al, (%rax)
;; 4d: addb %al, (%rax)
28 changes: 28 additions & 0 deletions tests/disas/winch/x64/f64x2_splat/param_avx.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx" ]

(module
(func (param f64) (result v128)
(f64x2.splat (local.get 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x20, %r11
;; cmpq %rsp, %r11
;; ja 0x44
;; 1c: movq %rdi, %r14
;; subq $0x20, %rsp
;; movq %rdi, 0x18(%rsp)
;; movq %rsi, 0x10(%rsp)
;; movsd %xmm0, 8(%rsp)
;; movsd 8(%rsp), %xmm0
;; vpshufd $0x44, %xmm0, %xmm0
;; addq $0x20, %rsp
;; popq %rbp
;; retq
;; 44: ud2
29 changes: 29 additions & 0 deletions tests/disas/winch/x64/i16x8_splat/const_avx2.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx2" ]

(module
(func (result v128)
(i16x8.splat (i32.const 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x10, %r11
;; cmpq %rsp, %r11
;; ja 0x3b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
;; movq %rsi, (%rsp)
;; vpbroadcastw 0xb(%rip), %xmm0
;; addq $0x10, %rsp
;; popq %rbp
;; retq
;; 3b: ud2
;; 3d: addb %al, (%rax)
;; 3f: addb %al, (%rax)
;; 41: addb %al, (%rax)
29 changes: 29 additions & 0 deletions tests/disas/winch/x64/i16x8_splat/param_avx2.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx2" ]

(module
(func (param i32) (result v128)
(i16x8.splat (local.get 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x20, %r11
;; cmpq %rsp, %r11
;; ja 0x44
;; 1c: movq %rdi, %r14
;; subq $0x20, %rsp
;; movq %rdi, 0x18(%rsp)
;; movq %rsi, 0x10(%rsp)
;; movl %edx, 0xc(%rsp)
;; movl 0xc(%rsp), %eax
;; movd %eax, %xmm0
;; vpbroadcastw %xmm0, %xmm0
;; addq $0x20, %rsp
;; popq %rbp
;; retq
;; 44: ud2
29 changes: 29 additions & 0 deletions tests/disas/winch/x64/i32x4_splat/const_avx2.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx2" ]

(module
(func (result v128)
(i32x4.splat (i32.const 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x10, %r11
;; cmpq %rsp, %r11
;; ja 0x3b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
;; movq %rsi, (%rsp)
;; vpbroadcastd 0xb(%rip), %xmm0
;; addq $0x10, %rsp
;; popq %rbp
;; retq
;; 3b: ud2
;; 3d: addb %al, (%rax)
;; 3f: addb %al, (%rax)
;; 41: addb %al, (%rax)
29 changes: 29 additions & 0 deletions tests/disas/winch/x64/i32x4_splat/param_avx2.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx2" ]

(module
(func (param i32) (result v128)
(i32x4.splat (local.get 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x20, %r11
;; cmpq %rsp, %r11
;; ja 0x44
;; 1c: movq %rdi, %r14
;; subq $0x20, %rsp
;; movq %rdi, 0x18(%rsp)
;; movq %rsi, 0x10(%rsp)
;; movl %edx, 0xc(%rsp)
;; movl 0xc(%rsp), %eax
;; movd %eax, %xmm0
;; vpbroadcastd %xmm0, %xmm0
;; addq $0x20, %rsp
;; popq %rbp
;; retq
;; 44: ud2
31 changes: 31 additions & 0 deletions tests/disas/winch/x64/i64x2_splat/const_avx.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx" ]

(module
(func (result v128)
(i64x2.splat (i64.const 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x10, %r11
;; cmpq %rsp, %r11
;; ja 0x3b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
;; movq %rsi, (%rsp)
;; vpshufd $0x44, 0xb(%rip), %xmm0
;; addq $0x10, %rsp
;; popq %rbp
;; retq
;; 3b: ud2
;; 3d: addb %al, (%rax)
;; 3f: addb %al, (%rax)
;; 41: addb %al, (%rax)
;; 43: addb %al, (%rax)
;; 45: addb %al, (%rax)
29 changes: 29 additions & 0 deletions tests/disas/winch/x64/i64x2_splat/param_avx.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx" ]

(module
(func (param i64) (result v128)
(i64x2.splat (local.get 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x20, %r11
;; cmpq %rsp, %r11
;; ja 0x47
;; 1c: movq %rdi, %r14
;; subq $0x20, %rsp
;; movq %rdi, 0x18(%rsp)
;; movq %rsi, 0x10(%rsp)
;; movq %rdx, 8(%rsp)
;; movq 8(%rsp), %rax
;; movq %rax, %xmm0
;; vpshufd $0x44, %xmm0, %xmm0
;; addq $0x20, %rsp
;; popq %rbp
;; retq
;; 47: ud2
29 changes: 29 additions & 0 deletions tests/disas/winch/x64/i8x16_splat/const_avx2.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx2" ]

(module
(func (result v128)
(i8x16.splat (i32.const 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x10, %r11
;; cmpq %rsp, %r11
;; ja 0x3b
;; 1c: movq %rdi, %r14
;; subq $0x10, %rsp
;; movq %rdi, 8(%rsp)
;; movq %rsi, (%rsp)
;; vpbroadcastb 0xb(%rip), %xmm0
;; addq $0x10, %rsp
;; popq %rbp
;; retq
;; 3b: ud2
;; 3d: addb %al, (%rax)
;; 3f: addb %al, (%rax)
;; 41: addb %al, (%rax)
29 changes: 29 additions & 0 deletions tests/disas/winch/x64/i8x16_splat/param_avx2.wat
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
;;! target = "x86_64"
;;! test = "winch"
;;! flags = [ "-Ccranelift-has-avx2" ]

(module
(func (param i32) (result v128)
(i8x16.splat (local.get 0))
)
)
;; wasm[0]::function[0]:
;; pushq %rbp
;; movq %rsp, %rbp
;; movq 8(%rdi), %r11
;; movq 0x10(%r11), %r11
;; addq $0x20, %r11
;; cmpq %rsp, %r11
;; ja 0x44
;; 1c: movq %rdi, %r14
;; subq $0x20, %rsp
;; movq %rdi, 0x18(%rsp)
;; movq %rsi, 0x10(%rsp)
;; movl %edx, 0xc(%rsp)
;; movl 0xc(%rsp), %eax
;; movd %eax, %xmm0
;; vpbroadcastb %xmm0, %xmm0
;; addq $0x20, %rsp
;; popq %rbp
;; retq
;; 44: ud2
Loading