diff --git a/examples/opt/aarch64/aarch64_simple0_a55.s b/examples/opt/aarch64/aarch64_simple0_a55.s deleted file mode 100644 index 093bb27d..00000000 --- a/examples/opt/aarch64/aarch64_simple0_a55.s +++ /dev/null @@ -1,73 +0,0 @@ - ldr q13, [x1, #0] // *.................. -// gap // ................... -// gap // ................... -// gap // ................... - ldr q30, [x0, #16] // ..*................ -// gap // ................... -// gap // ................... -// gap // ................... - ldr q8, [x0, #48] // ....*.............. -// gap // ................... -// gap // ................... -// gap // ................... - mul v25.8H, v30.8H, v13.H[0] // .....*............. -// gap // ................... - sqrdmulh v6.8H, v30.8H, v13.H[1] // ......*............ -// gap // ................... - sqrdmulh v10.8H, v8.8H, v13.H[1] // ...........*....... -// gap // ................... - mul v13.8H, v8.8H, v13.H[0] // ..........*........ -// gap // ................... - ldr q8, [x0] // .*................. -// gap // ................... -// gap // ................... -// gap // ................... - mls v25.8H, v6.8H, v1.H[0] // .......*........... -// gap // ................... - ldr q17, [x0, #32] // ...*............... -// gap // ................... -// gap // ................... -// gap // ................... - mls v13.8H, v10.8H, v1.H[0] // ............*...... -// gap // ................... - add v6.8H, v8.8H, v25.8H // .........*......... -// gap // ................... - sub v8.8H, v8.8H, v25.8H // ........*.......... -// gap // ................... -// gap // ................... -// gap // ................... - str q6, [x0], #4*16 // ...............*... -// gap // ................... - sub v4.8H, v17.8H, v13.8H // .............*..... -// gap // ................... - str q8, [x0, #-48] // ................*.. -// gap // ................... - add v8.8H, v17.8H, v13.8H // ..............*.... -// gap // ................... - str q4, [x0, #-16] // ..................* -// gap // ................... -// gap // ................... -// gap // ................... - str q8, [x0, #-32] // .................*. -// gap // ................... - -// original source code -// ldr q0, [x1, #0] // *.................. -// ldr q8, [x0] // .......*........... -// ldr q9, [x0, #1*16] // .*................. -// ldr q10, [x0, #2*16] // .........*......... -// ldr q11, [x0, #3*16] // ..*................ -// mul v24.8h, v9.8h, v0.h[0] // ...*............... -// sqrdmulh v9.8h, v9.8h, v0.h[1] // ....*.............. -// mls v24.8h, v9.8h, v1.h[0] // ........*.......... -// sub v9.8h, v8.8h, v24.8h // ............*...... -// add v8.8h, v8.8h, v24.8h // ...........*....... -// mul v24.8h, v11.8h, v0.h[0] // ......*............ -// sqrdmulh v11.8h, v11.8h, v0.h[1] // .....*............. -// mls v24.8h, v11.8h, v1.h[0] // ..........*........ -// sub v11.8h, v10.8h, v24.8h // ..............*.... -// add v10.8h, v10.8h, v24.8h // ................*.. -// str q8, [x0], #4*16 // .............*..... -// str q9, [x0, #-3*16] // ...............*... -// str q10, [x0, #-2*16] // ..................* -// str q11, [x0, #-1*16] // .................*. diff --git a/tutorial/tutorial-3-macros.py b/tutorial/tutorial-3-macros.py index 1b1cac46..74f0c9a6 100644 --- a/tutorial/tutorial-3-macros.py +++ b/tutorial/tutorial-3-macros.py @@ -20,4 +20,4 @@ slothy.config.constraints.stalls_first_attempt=32 slothy.optimize(start="start", end="end") -slothy.write_source_to_file("../examples/opt/aarch64/aarch64_simple0_macros_a55.s") +slothy.write_source_to_file("../examples/opt/aarch64/aarch64_simple0_macros_opt_a55.s") diff --git a/tutorial/tutorial-3.py b/tutorial/tutorial-3.py index 59b1d46f..6b8e2fc3 100644 --- a/tutorial/tutorial-3.py +++ b/tutorial/tutorial-3.py @@ -20,4 +20,4 @@ slothy.config.constraints.stalls_first_attempt=32 slothy.optimize() -slothy.write_source_to_file("../examples/opt/aarch64/aarch64_simple0_a55.s") \ No newline at end of file +slothy.write_source_to_file("../examples/opt/aarch64/aarch64_simple0_opt_a55.s") \ No newline at end of file diff --git a/tutorial/tutorial-4.py b/tutorial/tutorial-4.py index f3644b13..6858463f 100644 --- a/tutorial/tutorial-4.py +++ b/tutorial/tutorial-4.py @@ -24,4 +24,4 @@ slothy.config.sw_pipelining.optimize_postamble = False slothy.config.with_llvm_mca = True slothy.optimize_loop("start") -slothy.write_source_to_file("../examples/opt/aarch64/aarch64_simple0_loop_a55.s") +slothy.write_source_to_file("../examples/opt/aarch64/aarch64_simple0_loop_opt_a55.s")