Skip to content

Commit 2cba920

Browse files
committed
Further work to improve indexing code gen.
1 parent 574e34e commit 2cba920

31 files changed

+144
-80
lines changed

Project.toml

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
name = "LoopVectorization"
22
uuid = "bdcacae8-1622-11e9-2a5c-532679323890"
33
authors = ["Chris Elrod <[email protected]>"]
4-
version = "0.8.8"
4+
version = "0.8.9"
55

66
[deps]
77
DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"
@@ -15,10 +15,10 @@ VectorizationBase = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
1515
[compat]
1616
DocStringExtensions = "0.8"
1717
OffsetArrays = "1"
18-
SIMDPirates = "0.8.11"
18+
SIMDPirates = "0.8.12"
1919
SLEEFPirates = "0.5"
2020
UnPack = "0,1"
21-
VectorizationBase = "0.12.14"
21+
VectorizationBase = "0.12.15"
2222
julia = "1.1"
2323

2424
[extras]

benchmark/benchmarkflops.jl

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -61,9 +61,9 @@ function matmul_bench!(br, C, A, B, i)
6161
@assert C Cblas "Fort builtin gemm wrong?"; fill!(C, NaN)
6262
br[10,i] = n_gflop / @belapsed ifgemm_builtin!($C, $A, $B)
6363
@assert C Cblas "ifort builtin gemm wrong?"; fill!(C, NaN)
64-
br[11,i] = n_gflop / @belapsed dgemmopenblas!($C, $A, $B);
64+
br[11,i] = n_gflop / @belapsed gemmopenblas!($C, $A, $B);
6565
@assert C Cblas "OpenBLAS gemm wrong?"
66-
br[12,i] = n_gflop / @belapsed dgemmmkl!($C, $A, $B)
66+
br[12,i] = n_gflop / @belapsed gemmmkl!($C, $A, $B)
6767
@assert C Cblas "MKL gemm wrong?"
6868
# br[12,i] = n_gflop / @belapsed gemmavx!($C, $A, $B)
6969
end

benchmark/loadsharedlibs.jl

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -24,7 +24,7 @@ end
2424
ffile = joinpath(LOOPVECBENCHDIR, "looptests.f90")
2525
if !isfile(LIBFTEST) || mtime(ffile) > mtime(LIBFTEST)
2626
# --param max-unroll-times defaults to ≥8, which is generally excessive
27-
run(`gfortran -Ofast -march=native -funroll-loops -mprefer-vector-width=$(8REGISTER_SIZE) -shared -fPIC $ffile -o $LIBFTEST`)
27+
run(`gfortran -Ofast -march=native -funroll-loops -mprefer-vector-width=$(8REGISTER_SIZE) -fvariable-expansion-in-unroller --param max-variable-expansions-in-unroller=4 -shared -fPIC $ffile -o $LIBFTEST`)
2828
# run(`gfortran -Ofast -march=native -funroll-loops -floop-nest-optimize -mprefer-vector-width=$(8REGISTER_SIZE) -shared -fPIC $ffile -o $LIBFTEST`)
2929
end
3030
if !isfile(LIBIFTEST) || mtime(ffile) > mtime(LIBIFTEST)

docs/src/assets/bench_AmulB_v1.png

4.98 KB
Loading

docs/src/assets/bench_AmulBt_v1.png

15.7 KB
Loading

docs/src/assets/bench_Amulvb_v1.png

9 Bytes
Loading

docs/src/assets/bench_AplusAt_v1.png

164 KB
Loading

docs/src/assets/bench_AtmulB_v1.png

3.85 KB
Loading

docs/src/assets/bench_AtmulBt_v1.png

28.8 KB
Loading

docs/src/assets/bench_Atmulvb_v1.png

-29.7 KB
Loading

0 commit comments

Comments
 (0)