COMPMID-881: Updated arm_gemm to the lastest

Change-Id: Iba2664f33320e79bd15ca9c1399e65e4cc165be6
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/125265
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

1 files changed, 120 insertions, 119 deletions

diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_trans/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_trans/generic.cpp
index 3309baff3a..8fa403bf02 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_trans/generic.cpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_trans/generic.cpp
@@ -50,12 +50,12 @@
 
 namespace arm_gemm
 {
-void a64_sgemv_trans(const float *Astart, const float *Xstart, float *Ystart, float alpha, int lda, int M, int N)
+void a64_sgemv_trans(const float *Astart, const float *Xstart, float *Ystart, float beta, int lda, int M, int N)
 {
     const float *a_ptr_base = Astart;
     float       *y_ptr      = Ystart;
 
-    register const float32x4_t va asm("v1") = vdupq_n_f32(alpha);
+    register const float32x4_t vb asm("v1") = vdupq_n_f32(beta);
 
     int firstpfd = FIRST_PFD;
     if(firstpfd > M)
@@ -344,6 +344,7 @@ void a64_sgemv_trans(const float *Astart, const float *Xstart, float *Ystart, fl
             "fmla    v17.4s, v5.4s, v0.4s\n"
             "ldr    q5, [%[a_ptr], #0xf0]\n"
             "fmla    v18.4s, v6.4s, v0.4s\n"
+
             "ldr    q6, [%[a_ptr], #0x100]\n"
             "fmla    v19.4s, v7.4s, v0.4s\n"
             "ldr    q7, [%[a_ptr], #0x110]\n"
@@ -372,75 +373,75 @@ void a64_sgemv_trans(const float *Astart, const float *Xstart, float *Ystart, fl
             "fmla    v31.4s, v7.4s, v0.4s\n"
             "ldr    q7,  [%[y_ptr], #0x50]\n"
 
-            "fmla    v2.4s, v8.4s, %[va].4s\n"
-            "ldr    q8, [%[y_ptr], #0x60]\n"
-            "fmla    v3.4s, v9.4s, %[va].4s\n"
-            "ldr    q9, [%[y_ptr], #0x70]\n"
-            "fmla    v4.4s, v10.4s, %[va].4s\n"
-            "ldr    q10, [%[y_ptr], #0x80]\n"
-            "fmla    v5.4s, v11.4s, %[va].4s\n"
-            "ldr    q11, [%[y_ptr], #0x90]\n"
-            "fmla    v6.4s, v12.4s, %[va].4s\n"
-            "ldr    q12, [%[y_ptr], #0xa0]\n"
-            "str    q2, [%[y_ptr], #0x00]\n"
-            "fmla    v7.4s, v13.4s, %[va].4s\n"
-            "ldr    q13, [%[y_ptr], #0xb0]\n"
-            "str    q3, [%[y_ptr], #0x10]\n"
-            "fmla    v8.4s, v14.4s, %[va].4s\n"
-            "ldr    q14, [%[y_ptr], #0xc0]\n"
-            "str    q4, [%[y_ptr], #0x20]\n"
-            "fmla    v9.4s, v15.4s, %[va].4s\n"
-            "ldr    q15, [%[y_ptr], #0xd0]\n"
-            "str    q5, [%[y_ptr], #0x30]\n"
-            "fmla    v10.4s, v16.4s, %[va].4s\n"
-            "ldr    q16, [%[y_ptr], #0xe0]\n"
-            "str    q6, [%[y_ptr], #0x40]\n"
-            "fmla    v11.4s, v17.4s, %[va].4s\n"
-            "ldr    q17, [%[y_ptr], #0xf0]\n"
-            "str    q7, [%[y_ptr], #0x50]\n"
-            "fmla    v12.4s, v18.4s, %[va].4s\n"
-            "ldr    q18, [%[y_ptr], #0x100]\n"
-            "str    q8, [%[y_ptr], #0x60]\n"
-            "fmla    v13.4s, v19.4s, %[va].4s\n"
-            "ldr    q19, [%[y_ptr], #0x110]\n"
-            "str    q9, [%[y_ptr], #0x70]\n"
-            "fmla    v14.4s, v20.4s, %[va].4s\n"
-            "ldr    q20, [%[y_ptr], #0x120]\n"
-            "str    q10, [%[y_ptr], #0x80]\n"
-            "fmla    v15.4s, v21.4s, %[va].4s\n"
-            "ldr    q21, [%[y_ptr], #0x130]\n"
-            "str    q11, [%[y_ptr], #0x90]\n"
-            "fmla    v16.4s, v22.4s, %[va].4s\n"
-            "ldr    q22, [%[y_ptr], #0x140]\n"
-            "str    q12, [%[y_ptr], #0xa0]\n"
-            "fmla    v17.4s, v23.4s, %[va].4s\n"
-            "ldr    q23, [%[y_ptr], #0x150]\n"
-            "str    q13, [%[y_ptr], #0xb0]\n"
-            "fmla    v18.4s, v24.4s, %[va].4s\n"
-            "ldr    q24, [%[y_ptr], #0x160]\n"
-            "str    q14, [%[y_ptr], #0xc0]\n"
-            "fmla    v19.4s, v25.4s, %[va].4s\n"
-            "ldr    q25, [%[y_ptr], #0x170]\n"
-            "str    q15, [%[y_ptr], #0xd0]\n"
-            "fmla    v20.4s, v26.4s, %[va].4s\n"
-            "str    q16, [%[y_ptr], #0xe0]\n"
-            "fmla    v21.4s, v27.4s, %[va].4s\n"
-            "str    q17, [%[y_ptr], #0xf0]\n"
-            "fmla    v22.4s, v28.4s, %[va].4s\n"
-            "str    q18, [%[y_ptr], #0x100]\n"
-            "fmla    v23.4s, v29.4s, %[va].4s\n"
-            "str    q19, [%[y_ptr], #0x110]\n"
-            "fmla    v24.4s, v30.4s, %[va].4s\n"
-            "str    q20, [%[y_ptr], #0x120]\n"
-            "fmla    v25.4s, v31.4s, %[va].4s\n"
-            "str    q21, [%[y_ptr], #0x130]\n"
-
-            "stp    q22, q23, [%[y_ptr], #0x140]\n"
-            "stp    q24, q25, [%[y_ptr], #0x160]\n"
+            "fmla    v8.4s, v2.4s, %[vb].4s\n"
+            "ldr    q2, [%[y_ptr], #0x60]\n"
+            "fmla    v9.4s, v3.4s, %[vb].4s\n"
+            "ldr    q3, [%[y_ptr], #0x70]\n"
+            "fmla    v10.4s, v4.4s, %[vb].4s\n"
+            "ldr    q4, [%[y_ptr], #0x80]\n"
+            "fmla    v11.4s, v5.4s, %[vb].4s\n"
+            "ldr    q5, [%[y_ptr], #0x90]\n"
+            "fmla    v12.4s, v6.4s, %[vb].4s\n"
+            "ldr    q6, [%[y_ptr], #0xa0]\n"
+            "str    q8, [%[y_ptr], #0x00]\n"
+            "fmla    v13.4s, v7.4s, %[vb].4s\n"
+            "ldr    q7, [%[y_ptr], #0xb0]\n"
+            "str    q9, [%[y_ptr], #0x10]\n"
+            "fmla    v14.4s, v2.4s, %[vb].4s\n"
+            "ldr    q2, [%[y_ptr], #0xc0]\n"
+            "str    q10, [%[y_ptr], #0x20]\n"
+            "fmla    v15.4s, v3.4s, %[vb].4s\n"
+            "ldr    q3, [%[y_ptr], #0xd0]\n"
+            "str    q11, [%[y_ptr], #0x30]\n"
+            "fmla    v16.4s, v4.4s, %[vb].4s\n"
+            "ldr    q4, [%[y_ptr], #0xe0]\n"
+            "str    q12, [%[y_ptr], #0x40]\n"
+            "fmla    v17.4s, v5.4s, %[vb].4s\n"
+            "ldr    q5, [%[y_ptr], #0xf0]\n"
+            "str    q13, [%[y_ptr], #0x50]\n"
+            "fmla    v18.4s, v6.4s, %[vb].4s\n"
+            "ldr    q6, [%[y_ptr], #0x100]\n"
+            "str    q14, [%[y_ptr], #0x60]\n"
+            "fmla    v19.4s, v7.4s, %[vb].4s\n"
+            "ldr    q7, [%[y_ptr], #0x110]\n"
+            "str    q15, [%[y_ptr], #0x70]\n"
+            "fmla    v20.4s, v2.4s, %[vb].4s\n"
+            "ldr    q2, [%[y_ptr], #0x120]\n"
+            "str    q16, [%[y_ptr], #0x80]\n"
+            "fmla    v21.4s, v3.4s, %[vb].4s\n"
+            "ldr    q3, [%[y_ptr], #0x130]\n"
+            "str    q17, [%[y_ptr], #0x90]\n"
+            "fmla    v22.4s, v4.4s, %[vb].4s\n"
+            "ldr    q4, [%[y_ptr], #0x140]\n"
+            "str    q18, [%[y_ptr], #0xa0]\n"
+            "fmla    v23.4s, v5.4s, %[vb].4s\n"
+            "ldr    q5, [%[y_ptr], #0x150]\n"
+            "str    q19, [%[y_ptr], #0xb0]\n"
+            "fmla    v24.4s, v6.4s, %[vb].4s\n"
+            "ldr    q6, [%[y_ptr], #0x160]\n"
+            "str    q20, [%[y_ptr], #0xc0]\n"
+            "fmla    v25.4s, v7.4s, %[vb].4s\n"
+            "ldr    q7, [%[y_ptr], #0x170]\n"
+            "str    q21, [%[y_ptr], #0xd0]\n"
+            "fmla    v26.4s, v2.4s, %[vb].4s\n"
+            "str    q22, [%[y_ptr], #0xe0]\n"
+            "fmla    v27.4s, v3.4s, %[vb].4s\n"
+            "str    q23, [%[y_ptr], #0xf0]\n"
+            "fmla    v28.4s, v4.4s, %[vb].4s\n"
+            "str    q24, [%[y_ptr], #0x100]\n"
+            "fmla    v29.4s, v5.4s, %[vb].4s\n"
+            "str    q25, [%[y_ptr], #0x110]\n"
+            "fmla    v30.4s, v6.4s, %[vb].4s\n"
+            "str    q26, [%[y_ptr], #0x120]\n"
+            "fmla    v31.4s, v7.4s, %[vb].4s\n"
+            "str    q27, [%[y_ptr], #0x130]\n"
+
+            "stp    q28, q29, [%[y_ptr], #0x140]\n"
+            "stp    q30, q31, [%[y_ptr], #0x160]\n"
             "add    %[y_ptr], %[y_ptr], #0x180\n"
 
             : [a_ptr] "+r"(a_ptr), [x_ptr] "+r"(x_ptr), [y_ptr] "+r"(y_ptr), [k] "+r"(k), [pf_ptr] "+r"(pf_ptr), [firstpf_ptr] "+r"(firstpf_ptr)
-            : [jump] "r"(jump), [va] "w"(va), [pf_limit] "r"(pf_limit)
+            : [jump] "r"(jump), [vb] "w"(vb), [pf_limit] "r"(pf_limit)
             : "w0", "v0", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13",
             "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
             "v27", "v28", "v29", "v30", "v31", "cc");
@@ -747,161 +748,161 @@ void a64_sgemv_trans(const float *Astart, const float *Xstart, float *Ystart, fl
             // Vector 0
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v8.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v8.4s, v7.4s, %[vb].4s\n"
+            "str    q8, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 1
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v9.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v9.4s, v7.4s, %[vb].4s\n"
+            "str    q9, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 2
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v10.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v10.4s, v7.4s, %[vb].4s\n"
+            "str    q10, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 3
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v11.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v11.4s, v7.4s, %[vb].4s\n"
+            "str    q11, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 4
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v12.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v12.4s, v7.4s, %[vb].4s\n"
+            "str    q12, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 5
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v13.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v13.4s, v7.4s, %[vb].4s\n"
+            "str    q13, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 6
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v14.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v14.4s, v7.4s, %[vb].4s\n"
+            "str    q14, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 7
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v15.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v15.4s, v7.4s, %[vb].4s\n"
+            "str    q15, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 8
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v16.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v16.4s, v7.4s, %[vb].4s\n"
+            "str    q16, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 9
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v17.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v17.4s, v7.4s, %[vb].4s\n"
+            "str    q17, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 10
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v18.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v18.4s, v7.4s, %[vb].4s\n"
+            "str    q18, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 11
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v19.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v19.4s, v7.4s, %[vb].4s\n"
+            "str    q19, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 12
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v20.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v20.4s, v7.4s, %[vb].4s\n"
+            "str    q20, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 13
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v21.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v21.4s, v7.4s, %[vb].4s\n"
+            "str    q21, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 14
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v22.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v22.4s, v7.4s, %[vb].4s\n"
+            "str    q22, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 15
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v23.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v23.4s, v7.4s, %[vb].4s\n"
+            "str    q23, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 16
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v24.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v24.4s, v7.4s, %[vb].4s\n"
+            "str    q24, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 17
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v25.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v25.4s, v7.4s, %[vb].4s\n"
+            "str    q25, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 18
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v26.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v26.4s, v7.4s, %[vb].4s\n"
+            "str    q26, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 19
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v27.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v27.4s, v7.4s, %[vb].4s\n"
+            "str    q27, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 20
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v28.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v28.4s, v7.4s, %[vb].4s\n"
+            "str    q28, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 21
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v29.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v29.4s, v7.4s, %[vb].4s\n"
+            "str    q29, [%[y_ptr]], #0x10\n"
             "beq    12f\n"
             // Vector 22
             "subs    %w[vecs], %w[vecs], #1\n"
             "ldr    q7, [%[y_ptr]]\n"
-            "fmla    v7.4s, v30.4s, %[va].4s\n"
-            "str    q7, [%[y_ptr]], #0x10\n"
+            "fmla    v30.4s, v7.4s, %[vb].4s\n"
+            "str    q30, [%[y_ptr]], #0x10\n"
 
             // Odd 2
             "12:\n"
             "cbz    %[odd2_aptr], 13f\n"
             "ldr    d7, [%[y_ptr]]\n"
-            "fmla    v7.2s, v6.2s, %[va].2s\n"
-            "str    d7, [%[y_ptr]], #0x8\n"
+            "fmla    v6.2s, v7.2s, %[vb].2s\n"
+            "str    d6, [%[y_ptr]], #0x8\n"
 
             // Odd 1
             "13:\n"
             "cbz    %[odd1_aptr], 14f\n"
             "ldr    s7, [%[y_ptr]]\n"
-            "fmla    v7.2s, v5.2s, %[va].2s\n"
-            "str    s7, [%[y_ptr]]\n"
+            "fmla    v5.2s, v7.2s, %[vb].2s\n"
+            "str    s5, [%[y_ptr]]\n"
 
             "14:\n"
             : [a_ptr] "+r"(a_ptr), [x_ptr] "+r"(x_ptr), [y_ptr] "+r"(y_ptr), [k] "+r"(k),
             [pf_ptr] "+r"(pf_ptr), [firstpf_ptr] "+r"(firstpf_ptr),
             [odd1_aptr] "+r"(odd1_aptr), [odd2_aptr] "+r"(odd2_aptr),
             [dopf] "+r"(dopf), [vecs] "+r"(vecs)
-            : [jump] "r"(jump), [va] "w"(va), [pf_limit] "r"(pf_limit), [numvecs] "r"(numvecs)
+            : [jump] "r"(jump), [vb] "w"(vb), [pf_limit] "r"(pf_limit), [numvecs] "r"(numvecs)
             : "w0", "v0", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13",
             "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
             "v27", "v28", "v29", "v30", "v31", "cc");