skia2/resources/SkVMTest.expected
Herb Derby 43f7641e1b Arrange instruction to reduce register pressure
When converting from Instructions to OptimizedInstructions
place instructions that reduce register pressure earlier in
the instruction list.

This change reduces some register pressure in SkVM, and
improves the bitmap_RGBA_8888_A_scale_bilerp benchmark by
about 5%.

Change-Id: If5f6385bd2f7720701d1c827265062b35491a790
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/276485
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
2020-03-17 21:37:18 +00:00

777 lines
18 KiB
Plaintext

A8 over A8
14 values (originally 17):
v0 = load8 arg(0)
v1 = to_f32 v0
↑ v2 = splat 3F800000 (1)
↑ v3 = splat 3B808081 (0.0039215689)
v4 = fnma_f32 v1 v3 v2
v5 = load8 arg(1)
v6 = to_f32 v5
v7 = mul_f32 v6 v3
v8 = mul_f32 v1 v3
v9 = fma_f32 v7 v4 v8
↑ v10 = splat 437F0000 (255)
v11 = mul_f32 v9 v10
v12 = round v11
store8 arg(1) v12
6 registers, 14 instructions:
0 r0 = splat 3F800000 (1)
1 r1 = splat 3B808081 (0.0039215689)
2 r2 = splat 437F0000 (255)
loop:
3 r3 = load8 arg(0)
4 r3 = to_f32 r3
5 r4 = fnma_f32 r3 r1 r0
6 r5 = load8 arg(1)
7 r5 = to_f32 r5
8 r5 = mul_f32 r5 r1
9 r3 = mul_f32 r3 r1
10 r3 = fma_f32 r5 r4 r3
11 r3 = mul_f32 r3 r2
12 r3 = round r3
13 store8 arg(1) r3
A8 over G8
19 values (originally 24):
v0 = load8 arg(0)
v1 = to_f32 v0
↑ v2 = splat 3F800000 (1)
↑ v3 = splat 3B808081 (0.0039215689)
v4 = fnma_f32 v1 v3 v2
v5 = load8 arg(1)
v6 = to_f32 v5
v7 = mul_f32 v6 v3
v8 = mul_f32 v7 v4
↑ v9 = splat 3D93DD98 (0.0722)
v10 = mul_f32 v8 v9
↑ v11 = splat 3F371759 (0.71520001)
v12 = fma_f32 v8 v11 v10
↑ v13 = splat 3E59B3D0 (0.21259999)
v14 = fma_f32 v8 v13 v12
↑ v15 = splat 437F0000 (255)
v16 = mul_f32 v14 v15
v17 = round v16
store8 arg(1) v17
8 registers, 19 instructions:
0 r0 = splat 3F800000 (1)
1 r1 = splat 3B808081 (0.0039215689)
2 r2 = splat 3D93DD98 (0.0722)
3 r3 = splat 3F371759 (0.71520001)
4 r4 = splat 3E59B3D0 (0.21259999)
5 r5 = splat 437F0000 (255)
loop:
6 r6 = load8 arg(0)
7 r6 = to_f32 r6
8 r6 = fnma_f32 r6 r1 r0
9 r7 = load8 arg(1)
10 r7 = to_f32 r7
11 r7 = mul_f32 r7 r1
12 r6 = mul_f32 r7 r6
13 r7 = mul_f32 r6 r2
14 r7 = fma_f32 r6 r3 r7
15 r7 = fma_f32 r6 r4 r7
16 r7 = mul_f32 r7 r5
17 r7 = round r7
18 store8 arg(1) r7
A8 over RGBA_8888
39 values (originally 41):
v0 = load8 arg(0)
v1 = to_f32 v0
↑ v2 = splat 3F800000 (1)
↑ v3 = splat 3B808081 (0.0039215689)
v4 = fnma_f32 v1 v3 v2
v5 = load32 arg(1)
v6 = shr_i32 v5 24
v7 = to_f32 v6
v8 = mul_f32 v7 v3
v9 = mul_f32 v1 v3
v10 = fma_f32 v8 v4 v9
↑ v11 = splat 437F0000 (255)
v12 = mul_f32 v10 v11
v13 = shr_i32 v5 16
↑ v14 = splat FF (3.5733111e-43)
v15 = bit_and v14 v13
v16 = round v12
v17 = to_f32 v15
v18 = mul_f32 v17 v3
v19 = mul_f32 v18 v4
v20 = mul_f32 v19 v11
v21 = round v20
v22 = pack v21 v16 8
v23 = shr_i32 v5 8
v24 = bit_and v14 v23
v25 = to_f32 v24
v26 = mul_f32 v25 v3
v27 = mul_f32 v26 v4
v28 = mul_f32 v27 v11
v29 = round v28
v30 = bit_and v14 v5
v31 = to_f32 v30
v32 = mul_f32 v31 v3
v33 = mul_f32 v32 v4
v34 = mul_f32 v33 v11
v35 = round v34
v36 = pack v35 v29 8
v37 = pack v36 v22 16
store32 arg(1) v37
8 registers, 39 instructions:
0 r0 = splat 3F800000 (1)
1 r1 = splat 3B808081 (0.0039215689)
2 r2 = splat 437F0000 (255)
3 r3 = splat FF (3.5733111e-43)
loop:
4 r4 = load8 arg(0)
5 r4 = to_f32 r4
6 r5 = fnma_f32 r4 r1 r0
7 r6 = load32 arg(1)
8 r7 = shr_i32 r6 24
9 r7 = to_f32 r7
10 r7 = mul_f32 r7 r1
11 r4 = mul_f32 r4 r1
12 r4 = fma_f32 r7 r5 r4
13 r4 = mul_f32 r4 r2
14 r7 = shr_i32 r6 16
15 r7 = bit_and r3 r7
16 r4 = round r4
17 r7 = to_f32 r7
18 r7 = mul_f32 r7 r1
19 r7 = mul_f32 r7 r5
20 r7 = mul_f32 r7 r2
21 r7 = round r7
22 r4 = pack r7 r4 8
23 r7 = shr_i32 r6 8
24 r7 = bit_and r3 r7
25 r7 = to_f32 r7
26 r7 = mul_f32 r7 r1
27 r7 = mul_f32 r7 r5
28 r7 = mul_f32 r7 r2
29 r7 = round r7
30 r6 = bit_and r3 r6
31 r6 = to_f32 r6
32 r6 = mul_f32 r6 r1
33 r5 = mul_f32 r6 r5
34 r5 = mul_f32 r5 r2
35 r5 = round r5
36 r7 = pack r5 r7 8
37 r4 = pack r7 r4 16
38 store32 arg(1) r4
G8 over A8
11 values (originally 15):
v0 = load8 arg(1)
v1 = to_f32 v0
↑ v2 = splat 3B808081 (0.0039215689)
v3 = mul_f32 v1 v2
↑ v4 = splat 0 (0)
↑ v5 = splat 3F800000 (1)
v6 = fma_f32 v3 v4 v5
↑ v7 = splat 437F0000 (255)
v8 = mul_f32 v6 v7
v9 = round v8
store8 arg(1) v9
5 registers, 11 instructions:
0 r0 = splat 3B808081 (0.0039215689)
1 r1 = splat 0 (0)
2 r2 = splat 3F800000 (1)
3 r3 = splat 437F0000 (255)
loop:
4 r4 = load8 arg(1)
5 r4 = to_f32 r4
6 r4 = mul_f32 r4 r0
7 r4 = fma_f32 r4 r1 r2
8 r4 = mul_f32 r4 r3
9 r4 = round r4
10 store8 arg(1) r4
G8 over G8
19 values (originally 23):
v0 = load8 arg(1)
v1 = to_f32 v0
↑ v2 = splat 3B808081 (0.0039215689)
v3 = mul_f32 v1 v2
v4 = load8 arg(0)
v5 = to_f32 v4
v6 = mul_f32 v5 v2
↑ v7 = splat 0 (0)
v8 = fma_f32 v3 v7 v6
↑ v9 = splat 3D93DD98 (0.0722)
v10 = mul_f32 v8 v9
↑ v11 = splat 3F371759 (0.71520001)
v12 = fma_f32 v8 v11 v10
↑ v13 = splat 3E59B3D0 (0.21259999)
v14 = fma_f32 v8 v13 v12
↑ v15 = splat 437F0000 (255)
v16 = mul_f32 v14 v15
v17 = round v16
store8 arg(1) v17
8 registers, 19 instructions:
0 r0 = splat 3B808081 (0.0039215689)
1 r1 = splat 0 (0)
2 r2 = splat 3D93DD98 (0.0722)
3 r3 = splat 3F371759 (0.71520001)
4 r4 = splat 3E59B3D0 (0.21259999)
5 r5 = splat 437F0000 (255)
loop:
6 r6 = load8 arg(1)
7 r6 = to_f32 r6
8 r6 = mul_f32 r6 r0
9 r7 = load8 arg(0)
10 r7 = to_f32 r7
11 r7 = mul_f32 r7 r0
12 r7 = fma_f32 r6 r1 r7
13 r6 = mul_f32 r7 r2
14 r6 = fma_f32 r7 r3 r6
15 r6 = fma_f32 r7 r4 r6
16 r6 = mul_f32 r6 r5
17 r6 = round r6
18 store8 arg(1) r6
G8 over RGBA_8888
39 values (originally 43):
v0 = load32 arg(1)
v1 = shr_i32 v0 24
v2 = to_f32 v1
↑ v3 = splat 3B808081 (0.0039215689)
v4 = mul_f32 v2 v3
↑ v5 = splat 0 (0)
↑ v6 = splat 3F800000 (1)
v7 = fma_f32 v4 v5 v6
↑ v8 = splat 437F0000 (255)
v9 = mul_f32 v7 v8
v10 = shr_i32 v0 16
↑ v11 = splat FF (3.5733111e-43)
v12 = bit_and v11 v10
v13 = to_f32 v12
v14 = mul_f32 v13 v3
v15 = load8 arg(0)
v16 = to_f32 v15
v17 = mul_f32 v16 v3
v18 = fma_f32 v14 v5 v17
v19 = round v9
v20 = mul_f32 v18 v8
v21 = round v20
v22 = pack v21 v19 8
v23 = shr_i32 v0 8
v24 = bit_and v11 v23
v25 = to_f32 v24
v26 = mul_f32 v25 v3
v27 = fma_f32 v26 v5 v17
v28 = mul_f32 v27 v8
v29 = round v28
v30 = bit_and v11 v0
v31 = to_f32 v30
v32 = mul_f32 v31 v3
v33 = fma_f32 v32 v5 v17
v34 = mul_f32 v33 v8
v35 = round v34
v36 = pack v35 v29 8
v37 = pack v36 v22 16
store32 arg(1) v37
9 registers, 39 instructions:
0 r0 = splat 3B808081 (0.0039215689)
1 r1 = splat 0 (0)
2 r2 = splat 3F800000 (1)
3 r3 = splat 437F0000 (255)
4 r4 = splat FF (3.5733111e-43)
loop:
5 r5 = load32 arg(1)
6 r6 = shr_i32 r5 24
7 r6 = to_f32 r6
8 r6 = mul_f32 r6 r0
9 r6 = fma_f32 r6 r1 r2
10 r6 = mul_f32 r6 r3
11 r7 = shr_i32 r5 16
12 r7 = bit_and r4 r7
13 r7 = to_f32 r7
14 r7 = mul_f32 r7 r0
15 r8 = load8 arg(0)
16 r8 = to_f32 r8
17 r8 = mul_f32 r8 r0
18 r7 = fma_f32 r7 r1 r8
19 r6 = round r6
20 r7 = mul_f32 r7 r3
21 r7 = round r7
22 r6 = pack r7 r6 8
23 r7 = shr_i32 r5 8
24 r7 = bit_and r4 r7
25 r7 = to_f32 r7
26 r7 = mul_f32 r7 r0
27 r7 = fma_f32 r7 r1 r8
28 r7 = mul_f32 r7 r3
29 r7 = round r7
30 r5 = bit_and r4 r5
31 r5 = to_f32 r5
32 r5 = mul_f32 r5 r0
33 r8 = fma_f32 r5 r1 r8
34 r8 = mul_f32 r8 r3
35 r8 = round r8
36 r7 = pack r8 r7 8
37 r6 = pack r7 r6 16
38 store32 arg(1) r6
RGBA_8888 over A8
15 values (originally 33):
v0 = load32 arg(0)
v1 = shr_i32 v0 24
v2 = to_f32 v1
↑ v3 = splat 3F800000 (1)
↑ v4 = splat 3B808081 (0.0039215689)
v5 = fnma_f32 v2 v4 v3
v6 = load8 arg(1)
v7 = to_f32 v6
v8 = mul_f32 v7 v4
v9 = mul_f32 v2 v4
v10 = fma_f32 v8 v5 v9
↑ v11 = splat 437F0000 (255)
v12 = mul_f32 v10 v11
v13 = round v12
store8 arg(1) v13
6 registers, 15 instructions:
0 r0 = splat 3F800000 (1)
1 r1 = splat 3B808081 (0.0039215689)
2 r2 = splat 437F0000 (255)
loop:
3 r3 = load32 arg(0)
4 r3 = shr_i32 r3 24
5 r3 = to_f32 r3
6 r4 = fnma_f32 r3 r1 r0
7 r5 = load8 arg(1)
8 r5 = to_f32 r5
9 r5 = mul_f32 r5 r1
10 r3 = mul_f32 r3 r1
11 r3 = fma_f32 r5 r4 r3
12 r3 = mul_f32 r3 r2
13 r3 = round r3
14 store8 arg(1) r3
RGBA_8888 over G8
34 values (originally 39):
v0 = load32 arg(0)
v1 = shr_i32 v0 24
v2 = to_f32 v1
↑ v3 = splat 3F800000 (1)
↑ v4 = splat 3B808081 (0.0039215689)
v5 = fnma_f32 v2 v4 v3
v6 = shr_i32 v0 16
↑ v7 = splat FF (3.5733111e-43)
v8 = bit_and v7 v6
v9 = load8 arg(1)
v10 = to_f32 v9
v11 = mul_f32 v10 v4
v12 = to_f32 v8
v13 = mul_f32 v12 v4
v14 = fma_f32 v11 v5 v13
↑ v15 = splat 3D93DD98 (0.0722)
v16 = mul_f32 v14 v15
v17 = shr_i32 v0 8
v18 = bit_and v7 v17
v19 = to_f32 v18
v20 = mul_f32 v19 v4
v21 = fma_f32 v11 v5 v20
↑ v22 = splat 3F371759 (0.71520001)
v23 = fma_f32 v21 v22 v16
v24 = bit_and v7 v0
v25 = to_f32 v24
v26 = mul_f32 v25 v4
v27 = fma_f32 v11 v5 v26
↑ v28 = splat 3E59B3D0 (0.21259999)
v29 = fma_f32 v27 v28 v23
↑ v30 = splat 437F0000 (255)
v31 = mul_f32 v29 v30
v32 = round v31
store8 arg(1) v32
12 registers, 34 instructions:
0 r0 = splat 3F800000 (1)
1 r1 = splat 3B808081 (0.0039215689)
2 r2 = splat FF (3.5733111e-43)
3 r3 = splat 3D93DD98 (0.0722)
4 r4 = splat 3F371759 (0.71520001)
5 r5 = splat 3E59B3D0 (0.21259999)
6 r6 = splat 437F0000 (255)
loop:
7 r7 = load32 arg(0)
8 r8 = shr_i32 r7 24
9 r8 = to_f32 r8
10 r8 = fnma_f32 r8 r1 r0
11 r9 = shr_i32 r7 16
12 r9 = bit_and r2 r9
13 r10 = load8 arg(1)
14 r10 = to_f32 r10
15 r10 = mul_f32 r10 r1
16 r9 = to_f32 r9
17 r9 = mul_f32 r9 r1
18 r9 = fma_f32 r10 r8 r9
19 r9 = mul_f32 r9 r3
20 r11 = shr_i32 r7 8
21 r11 = bit_and r2 r11
22 r11 = to_f32 r11
23 r11 = mul_f32 r11 r1
24 r11 = fma_f32 r10 r8 r11
25 r9 = fma_f32 r11 r4 r9
26 r7 = bit_and r2 r7
27 r7 = to_f32 r7
28 r7 = mul_f32 r7 r1
29 r7 = fma_f32 r10 r8 r7
30 r9 = fma_f32 r7 r5 r9
31 r9 = mul_f32 r9 r6
32 r9 = round r9
33 store8 arg(1) r9
RGBA_8888 over RGBA_8888
51 values (originally 55):
v0 = load32 arg(0)
v1 = shr_i32 v0 24
v2 = to_f32 v1
↑ v3 = splat 3F800000 (1)
↑ v4 = splat 3B808081 (0.0039215689)
v5 = fnma_f32 v2 v4 v3
v6 = load32 arg(1)
v7 = shr_i32 v6 24
v8 = to_f32 v7
v9 = mul_f32 v8 v4
v10 = mul_f32 v2 v4
v11 = fma_f32 v9 v5 v10
↑ v12 = splat 437F0000 (255)
v13 = mul_f32 v11 v12
v14 = shr_i32 v6 16
↑ v15 = splat FF (3.5733111e-43)
v16 = bit_and v15 v14
v17 = to_f32 v16
v18 = mul_f32 v17 v4
v19 = shr_i32 v0 16
v20 = bit_and v15 v19
v21 = to_f32 v20
v22 = mul_f32 v21 v4
v23 = fma_f32 v18 v5 v22
v24 = round v13
v25 = mul_f32 v23 v12
v26 = round v25
v27 = pack v26 v24 8
v28 = shr_i32 v6 8
v29 = bit_and v15 v28
v30 = to_f32 v29
v31 = mul_f32 v30 v4
v32 = shr_i32 v0 8
v33 = bit_and v15 v32
v34 = to_f32 v33
v35 = mul_f32 v34 v4
v36 = fma_f32 v31 v5 v35
v37 = bit_and v15 v6
v38 = to_f32 v37
v39 = mul_f32 v38 v4
v40 = bit_and v15 v0
v41 = to_f32 v40
v42 = mul_f32 v41 v4
v43 = fma_f32 v39 v5 v42
v44 = mul_f32 v36 v12
v45 = round v44
v46 = mul_f32 v43 v12
v47 = round v46
v48 = pack v47 v45 8
v49 = pack v48 v27 16
store32 arg(1) v49
10 registers, 51 instructions:
0 r0 = splat 3F800000 (1)
1 r1 = splat 3B808081 (0.0039215689)
2 r2 = splat 437F0000 (255)
3 r3 = splat FF (3.5733111e-43)
loop:
4 r4 = load32 arg(0)
5 r5 = shr_i32 r4 24
6 r5 = to_f32 r5
7 r6 = fnma_f32 r5 r1 r0
8 r7 = load32 arg(1)
9 r8 = shr_i32 r7 24
10 r8 = to_f32 r8
11 r8 = mul_f32 r8 r1
12 r5 = mul_f32 r5 r1
13 r5 = fma_f32 r8 r6 r5
14 r5 = mul_f32 r5 r2
15 r8 = shr_i32 r7 16
16 r8 = bit_and r3 r8
17 r8 = to_f32 r8
18 r8 = mul_f32 r8 r1
19 r9 = shr_i32 r4 16
20 r9 = bit_and r3 r9
21 r9 = to_f32 r9
22 r9 = mul_f32 r9 r1
23 r9 = fma_f32 r8 r6 r9
24 r5 = round r5
25 r9 = mul_f32 r9 r2
26 r9 = round r9
27 r5 = pack r9 r5 8
28 r9 = shr_i32 r7 8
29 r9 = bit_and r3 r9
30 r9 = to_f32 r9
31 r9 = mul_f32 r9 r1
32 r8 = shr_i32 r4 8
33 r8 = bit_and r3 r8
34 r8 = to_f32 r8
35 r8 = mul_f32 r8 r1
36 r8 = fma_f32 r9 r6 r8
37 r7 = bit_and r3 r7
38 r7 = to_f32 r7
39 r7 = mul_f32 r7 r1
40 r4 = bit_and r3 r4
41 r4 = to_f32 r4
42 r4 = mul_f32 r4 r1
43 r4 = fma_f32 r7 r6 r4
44 r8 = mul_f32 r8 r2
45 r8 = round r8
46 r4 = mul_f32 r4 r2
47 r4 = round r4
48 r8 = pack r4 r8 8
49 r5 = pack r8 r5 16
50 store32 arg(1) r5
I32 (Naive) 8888 over 8888
33 values (originally 33):
v0 = load32 arg(0)
v1 = shr_i32 v0 24
↑ v2 = splat 100 (3.5873241e-43)
v3 = sub_i32 v2 v1
v4 = load32 arg(1)
v5 = shr_i32 v4 16
↑ v6 = splat FF (3.5733111e-43)
v7 = bit_and v6 v5
v8 = mul_i32 v7 v3
v9 = shr_i32 v8 8
v10 = shr_i32 v0 16
v11 = bit_and v6 v10
v12 = add_i32 v11 v9
v13 = shr_i32 v4 24
v14 = mul_i32 v13 v3
v15 = shr_i32 v14 8
v16 = add_i32 v1 v15
v17 = pack v12 v16 8
v18 = shr_i32 v4 8
v19 = bit_and v6 v18
v20 = mul_i32 v19 v3
v21 = shr_i32 v20 8
v22 = shr_i32 v0 8
v23 = bit_and v6 v22
v24 = add_i32 v23 v21
v25 = bit_and v6 v4
v26 = mul_i32 v25 v3
v27 = shr_i32 v26 8
v28 = bit_and v6 v0
v29 = add_i32 v28 v27
v30 = pack v29 v24 8
v31 = pack v30 v17 16
store32 arg(1) v31
8 registers, 33 instructions:
0 r0 = splat 100 (3.5873241e-43)
1 r1 = splat FF (3.5733111e-43)
loop:
2 r2 = load32 arg(0)
3 r3 = shr_i32 r2 24
4 r4 = sub_i32 r0 r3
5 r5 = load32 arg(1)
6 r6 = shr_i32 r5 16
7 r6 = bit_and r1 r6
8 r6 = mul_i32 r6 r4
9 r6 = shr_i32 r6 8
10 r7 = shr_i32 r2 16
11 r7 = bit_and r1 r7
12 r6 = add_i32 r7 r6
13 r7 = shr_i32 r5 24
14 r7 = mul_i32 r7 r4
15 r7 = shr_i32 r7 8
16 r7 = add_i32 r3 r7
17 r7 = pack r6 r7 8
18 r6 = shr_i32 r5 8
19 r6 = bit_and r1 r6
20 r6 = mul_i32 r6 r4
21 r6 = shr_i32 r6 8
22 r3 = shr_i32 r2 8
23 r3 = bit_and r1 r3
24 r6 = add_i32 r3 r6
25 r5 = bit_and r1 r5
26 r4 = mul_i32 r5 r4
27 r4 = shr_i32 r4 8
28 r2 = bit_and r1 r2
29 r4 = add_i32 r2 r4
30 r6 = pack r4 r6 8
31 r7 = pack r6 r7 16
32 store32 arg(1) r7
I32 8888 over 8888
29 values (originally 29):
v0 = load32 arg(0)
v1 = shr_i32 v0 24
↑ v2 = splat 100 (3.5873241e-43)
v3 = sub_i32 v2 v1
v4 = load32 arg(1)
v5 = bytes v4 3
v6 = mul_i16x2 v5 v3
v7 = shr_i32 v6 8
v8 = bytes v0 3
v9 = add_i32 v8 v7
v10 = shr_i32 v4 24
v11 = mul_i16x2 v10 v3
v12 = shr_i32 v11 8
v13 = add_i32 v1 v12
v14 = pack v9 v13 8
v15 = bytes v4 2
v16 = mul_i16x2 v15 v3
v17 = shr_i32 v16 8
v18 = bytes v0 2
v19 = add_i32 v18 v17
↑ v20 = splat FF (3.5733111e-43)
v21 = bit_and v4 v20
v22 = mul_i16x2 v21 v3
v23 = shr_i32 v22 8
v24 = bit_and v0 v20
v25 = add_i32 v24 v23
v26 = pack v25 v19 8
v27 = pack v26 v14 16
store32 arg(1) v27
8 registers, 29 instructions:
0 r0 = splat 100 (3.5873241e-43)
1 r1 = splat FF (3.5733111e-43)
loop:
2 r2 = load32 arg(0)
3 r3 = shr_i32 r2 24
4 r4 = sub_i32 r0 r3
5 r5 = load32 arg(1)
6 r6 = bytes r5 3
7 r6 = mul_i16x2 r6 r4
8 r6 = shr_i32 r6 8
9 r7 = bytes r2 3
10 r6 = add_i32 r7 r6
11 r7 = shr_i32 r5 24
12 r7 = mul_i16x2 r7 r4
13 r7 = shr_i32 r7 8
14 r7 = add_i32 r3 r7
15 r7 = pack r6 r7 8
16 r6 = bytes r5 2
17 r6 = mul_i16x2 r6 r4
18 r6 = shr_i32 r6 8
19 r3 = bytes r2 2
20 r6 = add_i32 r3 r6
21 r5 = bit_and r5 r1
22 r4 = mul_i16x2 r5 r4
23 r4 = shr_i32 r4 8
24 r2 = bit_and r2 r1
25 r4 = add_i32 r2 r4
26 r6 = pack r4 r6 8
27 r7 = pack r6 r7 16
28 store32 arg(1) r7
I32 (SWAR) 8888 over 8888
15 values (originally 15):
v0 = load32 arg(0)
v1 = bytes v0 404
↑ v2 = splat 1000100 (2.3510604e-38)
v3 = sub_i16x2 v2 v1
v4 = load32 arg(1)
v5 = shr_i16x2 v4 8
v6 = mul_i16x2 v5 v3
↑ v7 = splat FF00FF (2.3418409e-38)
v8 = bit_clear v6 v7
v9 = bit_and v4 v7
v10 = mul_i16x2 v9 v3
v11 = shr_i16x2 v10 8
v12 = bit_or v11 v8
v13 = add_i32 v0 v12
store32 arg(1) v13
6 registers, 15 instructions:
0 r0 = splat 1000100 (2.3510604e-38)
1 r1 = splat FF00FF (2.3418409e-38)
loop:
2 r2 = load32 arg(0)
3 r3 = bytes r2 404
4 r3 = sub_i16x2 r0 r3
5 r4 = load32 arg(1)
6 r5 = shr_i16x2 r4 8
7 r5 = mul_i16x2 r5 r3
8 r5 = bit_clear r5 r1
9 r4 = bit_and r4 r1
10 r3 = mul_i16x2 r4 r3
11 r3 = shr_i16x2 r3 8
12 r5 = bit_or r3 r5
13 r5 = add_i32 r2 r5
14 store32 arg(1) r5
6 values (originally 6):
↟ v0 = splat 2 (2.8025969e-45)
↟ v1 = splat 1 (1.4012985e-45)
↑ v2 = add_i32 v1 v0
v3 = load32 arg(0)
v4 = mul_i32 v3 v2
store32 arg(0) v4
2 registers, 6 instructions:
0 r0 = splat 2 (2.8025969e-45)
1 r1 = splat 1 (1.4012985e-45)
2 r0 = add_i32 r1 r0
loop:
3 r1 = load32 arg(0)
4 r1 = mul_i32 r1 r0
5 store32 arg(0) r1
23 values (originally 23):
v0 = load32 arg(1)
v1 = shr_i32 v0 24
v2 = load32 arg(0)
v3 = shr_i32 v2 24
v4 = add_i32 v3 v1
v5 = shr_i32 v0 16
↑ v6 = splat FF (3.5733111e-43)
v7 = bit_and v6 v5
v8 = shr_i32 v2 16
v9 = bit_and v6 v8
v10 = add_i32 v9 v7
v11 = pack v10 v4 8
v12 = shr_i32 v0 8
v13 = bit_and v6 v12
v14 = shr_i32 v2 8
v15 = bit_and v6 v14
v16 = add_i32 v15 v13
v17 = bit_and v6 v0
v18 = bit_and v6 v2
v19 = add_i32 v18 v17
v20 = pack v19 v16 8
v21 = pack v20 v11 16
store32 arg(1) v21
6 registers, 23 instructions:
0 r0 = splat FF (3.5733111e-43)
loop:
1 r1 = load32 arg(1)
2 r2 = shr_i32 r1 24
3 r3 = load32 arg(0)
4 r4 = shr_i32 r3 24
5 r2 = add_i32 r4 r2
6 r4 = shr_i32 r1 16
7 r4 = bit_and r0 r4
8 r5 = shr_i32 r3 16
9 r5 = bit_and r0 r5
10 r4 = add_i32 r5 r4
11 r2 = pack r4 r2 8
12 r4 = shr_i32 r1 8
13 r4 = bit_and r0 r4
14 r5 = shr_i32 r3 8
15 r5 = bit_and r0 r5
16 r4 = add_i32 r5 r4
17 r1 = bit_and r0 r1
18 r3 = bit_and r0 r3
19 r1 = add_i32 r3 r1
20 r4 = pack r1 r4 8
21 r2 = pack r4 r2 16
22 store32 arg(1) r2