diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/a55.cpp | 52 | ||||
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/generic.cpp | 10 |
2 files changed, 31 insertions, 31 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/a55.cpp index 184cfaf95c..e8b7db21bd 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/a55.cpp @@ -839,14 +839,14 @@ void a64_hybrid_fp32_mla_6x16_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "mov v12.16b, v8.16b\n" - "mov v16.16b, v8.16b\n" - "mov v13.16b, v9.16b\n" - "mov v17.16b, v9.16b\n" - "mov v14.16b, v10.16b\n" - "mov v18.16b, v10.16b\n" "ldr q11, [x16, #0x30]\n" + "mov v13.16b, v9.16b\n" "add x16, x16, #0x40\n" + "mov v14.16b, v10.16b\n" "mov v15.16b, v11.16b\n" + "mov v16.16b, v8.16b\n" + "mov v17.16b, v9.16b\n" + "mov v18.16b, v10.16b\n" "mov v19.16b, v11.16b\n" "b 80f\n" "69:" // Height 3: no bias @@ -1364,18 +1364,18 @@ void a64_hybrid_fp32_mla_6x16_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "mov v12.16b, v8.16b\n" - "mov v16.16b, v8.16b\n" + "ldr q11, [x16, #0x30]\n" "mov v13.16b, v9.16b\n" - "mov v17.16b, v9.16b\n" + "add x16, x16, #0x40\n" "mov v14.16b, v10.16b\n" + "mov v15.16b, v11.16b\n" + "mov v16.16b, v8.16b\n" + "mov v17.16b, v9.16b\n" "mov v18.16b, v10.16b\n" + "mov v19.16b, v11.16b\n" "mov v20.16b, v8.16b\n" "mov v21.16b, v9.16b\n" "mov v22.16b, v10.16b\n" - "ldr q11, [x16, #0x30]\n" - "add x16, x16, #0x40\n" - "mov v15.16b, v11.16b\n" - "mov v19.16b, v11.16b\n" "mov v23.16b, v11.16b\n" "b 113f\n" "102:" // Height 4: no bias @@ -1996,22 +1996,22 @@ void a64_hybrid_fp32_mla_6x16_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "mov v12.16b, v8.16b\n" - "mov v16.16b, v8.16b\n" + "ldr q11, [x16, #0x30]\n" "mov v13.16b, v9.16b\n" - "mov v17.16b, v9.16b\n" + "add x16, x16, #0x40\n" "mov v14.16b, v10.16b\n" + "mov v15.16b, v11.16b\n" + "mov v16.16b, v8.16b\n" + "mov v17.16b, v9.16b\n" "mov v18.16b, v10.16b\n" + "mov v19.16b, v11.16b\n" "mov v20.16b, v8.16b\n" "mov v21.16b, v9.16b\n" "mov v22.16b, v10.16b\n" + "mov v23.16b, v11.16b\n" "mov v24.16b, v8.16b\n" "mov v25.16b, v9.16b\n" "mov v26.16b, v10.16b\n" - "ldr q11, [x16, #0x30]\n" - "add x16, x16, #0x40\n" - "mov v15.16b, v11.16b\n" - "mov v19.16b, v11.16b\n" - "mov v23.16b, v11.16b\n" "mov v27.16b, v11.16b\n" "b 146f\n" "135:" // Height 5: no bias @@ -2738,26 +2738,26 @@ void a64_hybrid_fp32_mla_6x16_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "mov v12.16b, v8.16b\n" - "mov v16.16b, v8.16b\n" + "ldr q11, [x16, #0x30]\n" "mov v13.16b, v9.16b\n" - "mov v17.16b, v9.16b\n" + "add x16, x16, #0x40\n" "mov v14.16b, v10.16b\n" + "mov v15.16b, v11.16b\n" + "mov v16.16b, v8.16b\n" + "mov v17.16b, v9.16b\n" "mov v18.16b, v10.16b\n" + "mov v19.16b, v11.16b\n" "mov v20.16b, v8.16b\n" "mov v21.16b, v9.16b\n" "mov v22.16b, v10.16b\n" + "mov v23.16b, v11.16b\n" "mov v24.16b, v8.16b\n" "mov v25.16b, v9.16b\n" "mov v26.16b, v10.16b\n" + "mov v27.16b, v11.16b\n" "mov v28.16b, v8.16b\n" "mov v29.16b, v9.16b\n" "mov v30.16b, v10.16b\n" - "ldr q11, [x16, #0x30]\n" - "add x16, x16, #0x40\n" - "mov v15.16b, v11.16b\n" - "mov v19.16b, v11.16b\n" - "mov v23.16b, v11.16b\n" - "mov v27.16b, v11.16b\n" "mov v31.16b, v11.16b\n" "b 179f\n" "168:" // Height 6: no bias diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/generic.cpp index f5504b44d4..28e9be4cb7 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_6x16/generic.cpp @@ -1893,8 +1893,8 @@ void a64_hybrid_fp32_mla_6x16 ( "ld1 { v22.4s }, [x22], #0x10\n" "ld1 { v26.4s }, [x21], #0x10\n" "tbz x11, #1, 136f\n" - "mov x19, #0x38\n" "ldr d11, [x28], #0x8\n" + "mov x19, #0x38\n" "ldr d15, [x24], #0x8\n" "ldr d19, [x23], #0x8\n" "ldr d23, [x22], #0x8\n" @@ -1947,8 +1947,8 @@ void a64_hybrid_fp32_mla_6x16 ( "ld1 { v20.4s }, [x22], #0x10\n" "ld1 { v24.4s }, [x21], #0x10\n" "tbz x11, #1, 140f\n" - "mov x19, #0x18\n" "ldr d9, [x28], #0x8\n" + "mov x19, #0x18\n" "ldr d13, [x24], #0x8\n" "ldr d17, [x23], #0x8\n" "ldr d21, [x22], #0x8\n" @@ -2586,12 +2586,12 @@ void a64_hybrid_fp32_mla_6x16 ( "ld1 { v16.4s }, [x23], #0x10\n" "ld1 { v20.4s }, [x22], #0x10\n" "ld1 { v24.4s }, [x21], #0x10\n" - "ld1 { v28.4s }, [x20], #0x10\n" "ld1 { v9.4s }, [x28], #0x10\n" "ld1 { v13.4s }, [x24], #0x10\n" "ld1 { v17.4s }, [x23], #0x10\n" "ld1 { v21.4s }, [x22], #0x10\n" "ld1 { v25.4s }, [x21], #0x10\n" + "ld1 { v28.4s }, [x20], #0x10\n" "ld1 { v29.4s }, [x20], #0x10\n" "tbz x11, #2, 170f\n" "ld1 { v10.4s }, [x28], #0x10\n" @@ -2601,8 +2601,8 @@ void a64_hybrid_fp32_mla_6x16 ( "ld1 { v26.4s }, [x21], #0x10\n" "ld1 { v30.4s }, [x20], #0x10\n" "tbz x11, #1, 169f\n" - "mov x19, #0x38\n" "ldr d11, [x28], #0x8\n" + "mov x19, #0x38\n" "ldr d15, [x24], #0x8\n" "ldr d19, [x23], #0x8\n" "ldr d23, [x22], #0x8\n" @@ -2662,8 +2662,8 @@ void a64_hybrid_fp32_mla_6x16 ( "ld1 { v24.4s }, [x21], #0x10\n" "ld1 { v28.4s }, [x20], #0x10\n" "tbz x11, #1, 173f\n" - "mov x19, #0x18\n" "ldr d9, [x28], #0x8\n" + "mov x19, #0x18\n" "ldr d13, [x24], #0x8\n" "ldr d17, [x23], #0x8\n" "ldr d21, [x22], #0x8\n" |