diff options
Diffstat (limited to 'src/core/NEON/kernels')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/mergeresults.hpp | 2 | ||||
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp | 96 |
2 files changed, 49 insertions, 49 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/mergeresults.hpp b/src/core/NEON/kernels/arm_gemm/mergeresults.hpp index 2ab01d680c..4a6da3d21c 100644 --- a/src/core/NEON/kernels/arm_gemm/mergeresults.hpp +++ b/src/core/NEON/kernels/arm_gemm/mergeresults.hpp @@ -61,7 +61,7 @@ inline void MergeResults(Tout *out, const Tin *in, int ldc, int y0, int ymax, in { Tout &p = out[(ybase + row) * ldc + xbase + col]; - p = (p * alpha) + (beta * in[row * width + col]); + p = (p * beta) + (alpha * in[row * width + col]); } } diff --git a/src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp b/src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp index dc247aad37..79dd1f07e3 100644 --- a/src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp +++ b/src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp @@ -123,140 +123,140 @@ inline void MergeResults<12, 8>(int32_t *out, const int32_t *in, const int ldout "ldr q3, [%x[outptr0]]\n" "ldr q4, [%x[outptr0], #0x10]\n" "ldr q5, [%x[outptr0], #0x20]\n" - "mul v3.4s, v3.4s, %[alpha_value].4s\n" + "mul v3.4s, v3.4s, %[beta_value].4s\n" "ldr q6, [%x[inptr]]\n" - "mul v4.4s, v4.4s, %[alpha_value].4s\n" + "mul v4.4s, v4.4s, %[beta_value].4s\n" "ldr q7, [%x[inptr], #0x10]\n" - "mul v5.4s, v5.4s, %[alpha_value].4s\n" + "mul v5.4s, v5.4s, %[beta_value].4s\n" "ldr q8, [%x[inptr], #0x20]\n" - "mla v3.4s, v6.4s, %[beta_value].4s\n" + "mla v3.4s, v6.4s, %[alpha_value].4s\n" "ldr q0, [%x[outptr1]]\n" - "mla v4.4s, v7.4s, %[beta_value].4s\n" + "mla v4.4s, v7.4s, %[alpha_value].4s\n" "ldr q1, [%x[outptr1], #0x10]\n" - "mla v5.4s, v8.4s, %[beta_value].4s\n" + "mla v5.4s, v8.4s, %[alpha_value].4s\n" "ldr q2, [%x[outptr1], #0x20]\n" // Row 1 ASM_PREFETCH("[%x[outptr2], #192]") - "mul v0.4s, v0.4s, %[alpha_value].4s\n" + "mul v0.4s, v0.4s, %[beta_value].4s\n" "ldr q6, [%x[inptr], #0x30]\n" "str q3, [%x[outptr0]], #0x10\n" - "mul v1.4s, v1.4s, %[alpha_value].4s\n" + "mul v1.4s, v1.4s, %[beta_value].4s\n" "ldr q7, [%x[inptr], #0x40]\n" "str q4, [%x[outptr0]], #0x10\n" - "mul v2.4s, v2.4s, %[alpha_value].4s\n" + "mul v2.4s, v2.4s, %[beta_value].4s\n" "ldr q8, [%x[inptr], #0x50]\n" "str q5, [%x[outptr0]], #0x10\n" - "mla v0.4s, v6.4s, %[beta_value].4s\n" + "mla v0.4s, v6.4s, %[alpha_value].4s\n" "ldr q3, [%x[outptr2]]\n" - "mla v1.4s, v7.4s, %[beta_value].4s\n" + "mla v1.4s, v7.4s, %[alpha_value].4s\n" "ldr q4, [%x[outptr2], #0x10]\n" - "mla v2.4s, v8.4s, %[beta_value].4s\n" + "mla v2.4s, v8.4s, %[alpha_value].4s\n" "ldr q5, [%x[outptr2], #0x20]\n" // Row 2 ASM_PREFETCH("[%x[outptr3], #192]") - "mul v3.4s, v3.4s, %[alpha_value].4s\n" + "mul v3.4s, v3.4s, %[beta_value].4s\n" "ldr q6, [%x[inptr], #0x60]\n" "str q0, [%x[outptr1]], #0x10\n" - "mul v4.4s, v4.4s, %[alpha_value].4s\n" + "mul v4.4s, v4.4s, %[beta_value].4s\n" "ldr q7, [%x[inptr], #0x70]\n" "str q1, [%x[outptr1]], #0x10\n" - "mul v5.4s, v5.4s, %[alpha_value].4s\n" + "mul v5.4s, v5.4s, %[beta_value].4s\n" "ldr q8, [%x[inptr], #0x80]\n" "str q2, [%x[outptr1]], #0x10\n" - "mla v3.4s, v6.4s, %[beta_value].4s\n" + "mla v3.4s, v6.4s, %[alpha_value].4s\n" "ldr q0, [%x[outptr3]]\n" - "mla v4.4s, v7.4s, %[beta_value].4s\n" + "mla v4.4s, v7.4s, %[alpha_value].4s\n" "ldr q1, [%x[outptr3], #0x10]\n" - "mla v5.4s, v8.4s, %[beta_value].4s\n" + "mla v5.4s, v8.4s, %[alpha_value].4s\n" "ldr q2, [%x[outptr3], #0x20]\n" // Row 3 ASM_PREFETCH("[%x[outptr4], #192]") - "mul v0.4s, v0.4s, %[alpha_value].4s\n" + "mul v0.4s, v0.4s, %[beta_value].4s\n" "ldr q6, [%x[inptr], #0x90]\n" "str q3, [%x[outptr2]], #0x10\n" - "mul v1.4s, v1.4s, %[alpha_value].4s\n" + "mul v1.4s, v1.4s, %[beta_value].4s\n" "ldr q7, [%x[inptr], #0xa0]\n" "str q4, [%x[outptr2]], #0x10\n" - "mul v2.4s, v2.4s, %[alpha_value].4s\n" + "mul v2.4s, v2.4s, %[beta_value].4s\n" "ldr q8, [%x[inptr], #0xb0]\n" "str q5, [%x[outptr2]], #0x10\n" - "mla v0.4s, v6.4s, %[beta_value].4s\n" + "mla v0.4s, v6.4s, %[alpha_value].4s\n" "ldr q3, [%x[outptr4]]\n" - "mla v1.4s, v7.4s, %[beta_value].4s\n" + "mla v1.4s, v7.4s, %[alpha_value].4s\n" "ldr q4, [%x[outptr4], #0x10]\n" - "mla v2.4s, v8.4s, %[beta_value].4s\n" + "mla v2.4s, v8.4s, %[alpha_value].4s\n" "ldr q5, [%x[outptr4], #0x20]\n" // Row 4 ASM_PREFETCH("[%x[outptr5], #192]") - "mul v3.4s, v3.4s, %[alpha_value].4s\n" + "mul v3.4s, v3.4s, %[beta_value].4s\n" "ldr q6, [%x[inptr], #0xc0]\n" "str q0, [%x[outptr3]], #0x10\n" - "mul v4.4s, v4.4s, %[alpha_value].4s\n" + "mul v4.4s, v4.4s, %[beta_value].4s\n" "ldr q7, [%x[inptr], #0xd0]\n" "str q1, [%x[outptr3]], #0x10\n" - "mul v5.4s, v5.4s, %[alpha_value].4s\n" + "mul v5.4s, v5.4s, %[beta_value].4s\n" "ldr q8, [%x[inptr], #0xe0]\n" "str q2, [%x[outptr3]], #0x10\n" - "mla v3.4s, v6.4s, %[beta_value].4s\n" + "mla v3.4s, v6.4s, %[alpha_value].4s\n" "ldr q0, [%x[outptr5]]\n" - "mla v4.4s, v7.4s, %[beta_value].4s\n" + "mla v4.4s, v7.4s, %[alpha_value].4s\n" "ldr q1, [%x[outptr5], #0x10]\n" - "mla v5.4s, v8.4s, %[beta_value].4s\n" + "mla v5.4s, v8.4s, %[alpha_value].4s\n" "ldr q2, [%x[outptr5], #0x20]\n" // Row 5 ASM_PREFETCH("[%x[outptr6], #192]") - "mul v0.4s, v0.4s, %[alpha_value].4s\n" + "mul v0.4s, v0.4s, %[beta_value].4s\n" "ldr q6, [%x[inptr], #0xf0]\n" "str q3, [%x[outptr4]], #0x10\n" - "mul v1.4s, v1.4s, %[alpha_value].4s\n" + "mul v1.4s, v1.4s, %[beta_value].4s\n" "ldr q7, [%x[inptr], #0x100]\n" "str q4, [%x[outptr4]], #0x10\n" - "mul v2.4s, v2.4s, %[alpha_value].4s\n" + "mul v2.4s, v2.4s, %[beta_value].4s\n" "ldr q8, [%x[inptr], #0x110]\n" "str q5, [%x[outptr4]], #0x10\n" - "mla v0.4s, v6.4s, %[beta_value].4s\n" + "mla v0.4s, v6.4s, %[alpha_value].4s\n" "ldr q3, [%x[outptr6]]\n" - "mla v1.4s, v7.4s, %[beta_value].4s\n" + "mla v1.4s, v7.4s, %[alpha_value].4s\n" "ldr q4, [%x[outptr6], #0x10]\n" - "mla v2.4s, v8.4s, %[beta_value].4s\n" + "mla v2.4s, v8.4s, %[alpha_value].4s\n" "ldr q5, [%x[outptr6], #0x20]\n" // Row 6 ASM_PREFETCH("[%x[outptr7], #192]") - "mul v3.4s, v3.4s, %[alpha_value].4s\n" + "mul v3.4s, v3.4s, %[beta_value].4s\n" "ldr q6, [%x[inptr], #0x120]\n" "str q0, [%x[outptr5]], #0x10\n" - "mul v4.4s, v4.4s, %[alpha_value].4s\n" + "mul v4.4s, v4.4s, %[beta_value].4s\n" "ldr q7, [%x[inptr], #0x130]\n" "str q1, [%x[outptr5]], #0x10\n" - "mul v5.4s, v5.4s, %[alpha_value].4s\n" + "mul v5.4s, v5.4s, %[beta_value].4s\n" "ldr q8, [%x[inptr], #0x140]\n" "str q2, [%x[outptr5]], #0x10\n" - "mla v3.4s, v6.4s, %[beta_value].4s\n" + "mla v3.4s, v6.4s, %[alpha_value].4s\n" "ldr q0, [%x[outptr7]]\n" - "mla v4.4s, v7.4s, %[beta_value].4s\n" + "mla v4.4s, v7.4s, %[alpha_value].4s\n" "ldr q1, [%x[outptr7], #0x10]\n" - "mla v5.4s, v8.4s, %[beta_value].4s\n" + "mla v5.4s, v8.4s, %[alpha_value].4s\n" "ldr q2, [%x[outptr7], #0x20]\n" // Row 7 - "mul v0.4s, v0.4s, %[alpha_value].4s\n" + "mul v0.4s, v0.4s, %[beta_value].4s\n" "ldr q6, [%x[inptr], #0x150]\n" "str q3, [%x[outptr6]], #0x10\n" - "mul v1.4s, v1.4s, %[alpha_value].4s\n" + "mul v1.4s, v1.4s, %[beta_value].4s\n" "ldr q7, [%x[inptr], #0x160]\n" "str q4, [%x[outptr6]], #0x10\n" - "mul v2.4s, v2.4s, %[alpha_value].4s\n" + "mul v2.4s, v2.4s, %[beta_value].4s\n" "ldr q8, [%x[inptr], #0x170]\n" "str q5, [%x[outptr6]], #0x10\n" - "mla v0.4s, v6.4s, %[beta_value].4s\n" - "mla v1.4s, v7.4s, %[beta_value].4s\n" - "mla v2.4s, v8.4s, %[beta_value].4s\n" + "mla v0.4s, v6.4s, %[alpha_value].4s\n" + "mla v1.4s, v7.4s, %[alpha_value].4s\n" + "mla v2.4s, v8.4s, %[alpha_value].4s\n" "str q0, [%x[outptr7]], #0x10\n" "str q1, [%x[outptr7]], #0x10\n" "str q2, [%x[outptr7]], #0x10\n" |