erasure_code: add optimization implementation

Signed-off-by: Shuo Lv <[email protected]>

Author: lvshuo2016

configure.ac

@@ -70,9 +70,10 @@ case "${CPU}" in
 			[AC_DEFINE([HAVE_RVV], [0], [Disable RVV instructions])
 			AM_CONDITIONAL([HAVE_RVV], [false]) rvv=no]
 		)
+		AC_MSG_RESULT([$zicbop])
 		if test "x$rvv" = "xyes"; then
-		  CFLAGS+=" -march=rv64gcv"
-		  CCASFLAGS+=" -march=rv64gcv"
+			CFLAGS+=" -march=rv64gcv"
+			CCASFLAGS+=" -march=rv64gcv"
 		fi
 		AC_MSG_RESULT([$rvv])
 		;;

erasure_code/riscv64/gf_2vect_dot_prod_rvv.S

@@ -75,8 +75,9 @@ gf_2vect_dot_prod_rvv:
     blt x_len, t6, .return_fail
 
     vsetvli a5, x0, e8, m1  /* Set vector length to maximum */
-
     li x_pos, 0
+    slli x_vec, x_vec, 3
+
     ld x_dest1, 0(x_dest)
     ld x_dest2, 8(x_dest)
 
@@ -92,15 +93,12 @@ gf_2vect_dot_prod_rvv:
 
     /* gf_tbl base = (x_tbl + dest_idx * x_vec * 32) */
     mv x_tbl1, x_tbl           /* reset x_tbl1 */
-    slli t6, x_vec, 5
+    slli t6, x_vec, 2
     add x_tbl2, x_tbl1, t6     /* reset x_tbl2 */
 
 /* Loop 2: x_vec, number of source vectors (ie. data blocks) */
 .Llooprvv_vl_vects:
     /* load src data */
-    slli a6, x_vec_i, 3
-    add a6,x_src,a6
-    ld x_ptr, 0(a6)
     add x_ptr,x_ptr,x_pos
 
     vle8.v v_src, (x_ptr)      /* load from: src base + pos offset */
@@ -120,6 +118,11 @@ gf_2vect_dot_prod_rvv:
     vle8.v v_gft2_hi, (x_tbl2)
     addi x_tbl2, x_tbl2, 16
 
+    /* calc for next */
+    addi x_vec_i, x_vec_i, 8   /* move x_vec_i to next */
+    add a6,x_src,x_vec_i
+    ld x_ptr, 0(a6)
+
     /* dest 1 */
     /* table indexing, ie. gf(2^8) multiplication */
     vrgather.vv v26, v_gft1_lo, v_src_lo
@@ -134,8 +137,6 @@ gf_2vect_dot_prod_rvv:
     vxor.vv v_dest2, v_dest2, v26
     vxor.vv v_dest2, v_dest2, v27
 
-    /* calc for next */
-    addi x_vec_i, x_vec_i, 1   /* move x_vec_i to next */
     blt x_vec_i, x_vec, .Llooprvv_vl_vects
 /* end of Loop 2 */
 

erasure_code/riscv64/gf_3vect_dot_prod_rvv.S

@@ -85,9 +85,10 @@ gf_3vect_dot_prod_rvv:
     sd s1, 8(sp)
 
     vsetvli a7, x0, e8, m1  /* Set vector length to maximum */
-
     li x_pos, 0
-    slli t_offset, x_vec, 5
+    slli x_vec, x_vec, 3
+
+    slli t_offset, x_vec, 2
     ld x_dest1, 0(x_dest)
     ld x_dest2, 8(x_dest)
     ld x_dest3, 16(x_dest)
@@ -101,20 +102,19 @@ gf_3vect_dot_prod_rvv:
     vmv.v.i v_dest2, 0
     vmv.v.i v_dest3, 0
 
+    /* Loop 2: x_vec, number of source vectors (ie. data blocks) */
+    li x_vec_i, 0
+    /* load source pointer */
+    ld x_ptr, 0(x_src)
+
     /* Reset table pointers */
     mv x_tbl1, x_tbl
     add x_tbl2, x_tbl1, t_offset
     add x_tbl3, x_tbl2, t_offset
 
-    /* Loop 2: x_vec, number of source vectors (ie. data blocks) */
-    li x_vec_i, 0
 .Lloop_rvv_vl_vects:
-    /* Load source data */
-    slli t0, x_vec_i, 3
-    add t0,x_src,t0
-    ld x_ptr, 0(t0)
+    /* load source data */
     add x_ptr,x_ptr,x_pos
-
     vle8.v v_src, (x_ptr)
 
     /* Split 4-bit lo; 4-bit hi */
@@ -131,14 +131,18 @@ gf_3vect_dot_prod_rvv:
     vle8.v v_gft2_hi, (x_tbl2)
     addi x_tbl2, x_tbl2, 16
 
+    /* Move to next source vector */
+    addi x_vec_i, x_vec_i, 8
+    add t0,x_src,x_vec_i
+    ld x_ptr, 0(t0)
 
     /* Load next gf_table's */
     vle8.v v_gft3_lo, (x_tbl3)
     addi x_tbl3, x_tbl3, 16
     vle8.v v_gft3_hi, (x_tbl3)
     addi x_tbl3, x_tbl3, 16
 
-/* dest 1 */
+    /* dest 1 */
     vrgather.vv v26, v_gft1_lo, v_src_lo
     vrgather.vv v27, v_gft1_hi, v_src_hi
     vxor.vv v_dest1, v_dest1, v26
@@ -156,9 +160,6 @@ gf_3vect_dot_prod_rvv:
     vxor.vv v_dest3, v_dest3, v26
     vxor.vv v_dest3, v_dest3, v27
 
-    /* Move to next source vector */
-    addi x_vec_i, x_vec_i, 1
-
     /* Check if we have processed all vectors */
     blt x_vec_i, x_vec, .Lloop_rvv_vl_vects
 

erasure_code/riscv64/gf_4vect_dot_prod_rvv.S

@@ -91,10 +91,13 @@ gf_4vect_dot_prod_rvv:
     sd s2, 16(sp)
     sd s3, 24(sp)
 
-    vsetvli t0, x0, e8, m1  /* Set vector length to maximum */
-
+  #  vsetvli t0, x0, e8, m1  /* Set vector length to maximum */
+    vsetvli t0, x0, e8, m1
     li x_pos, 0
-    slli t_offset, x_vec, 5
+
+    slli x_vec, x_vec, 3
+    slli t_offset, x_vec, 2
+
     ld x_dest1, 0(x_dest)
     ld x_dest2, 8(x_dest)
     ld x_dest3, 16(x_dest)
@@ -111,19 +114,20 @@ gf_4vect_dot_prod_rvv:
     vmv.v.i v_dest3, 0
     vmv.v.i v_dest4, 0
 
+    /*  x_vec, number of source vectors (ie. data blocks) */
+    li x_vec_i, 0
+
+    /* load source pointer */
+    ld x_ptr, 0(x_src)
+
     /* Reset table pointers */
     mv x_tbl1, x_tbl
     add x_tbl2, x_tbl1, t_offset
     add x_tbl3, x_tbl2, t_offset
     add x_tbl4, x_tbl3, t_offset
 
-    /* Loop 2: x_vec, number of source vectors (ie. data blocks) */
-    li x_vec_i, 0
 .Lloop_rvv_vl_vects:
     /* Load source data */
-    slli a6, x_vec_i, 3
-    add a6,x_src,a6
-    ld x_ptr, 0(a6)
     add x_ptr,x_ptr,x_pos
 
     vle8.v v_src, (x_ptr)
@@ -142,6 +146,10 @@ gf_4vect_dot_prod_rvv:
     vle8.v v_gft2_hi, (x_tbl2)
     addi x_tbl2, x_tbl2, 16
 
+    /* Move to next source vector */
+    addi x_vec_i, x_vec_i, 8
+    add a6,x_src,x_vec_i
+    ld x_ptr, 0(a6)
 
     /* Load next gf_table's */
     vle8.v v_gft3_lo, (x_tbl3)
@@ -178,9 +186,6 @@ gf_4vect_dot_prod_rvv:
     vxor.vv v_dest4, v_dest4, v26
     vxor.vv v_dest4, v_dest4, v27
 
-    /* Move to next source vector */
-    addi x_vec_i, x_vec_i, 1
-
     /* Check if we have processed all vectors */
     blt x_vec_i, x_vec, .Lloop_rvv_vl_vects
 
@@ -198,7 +203,7 @@ gf_4vect_dot_prod_rvv:
     j .Lloop_rvv_vl
 
 .return_pass:
-/* restore callee-saved registers */
+    /* restore callee-saved registers */
     ld s0, 0(sp)
     ld s1, 8(sp)
     ld s2, 16(sp)

erasure_code/riscv64/gf_5vect_dot_prod_rvv.S

@@ -96,7 +96,6 @@ gf_5vect_dot_prod_rvv:
     sd s5, 40(sp)
 
     vsetvli a5, x0, e8, m1
-
     /* Initialize position */
     li x_pos, 0
 

erasure_code/riscv64/gf_6vect_dot_prod_rvv.S

@@ -107,6 +107,8 @@ gf_6vect_dot_prod_rvv:
     /* initialize position */
     li x_pos, 0
 
+    slli x_vec, x_vec, 3
+
     /* load destination pointers */
     ld x_dest1, 0(x14)  #  a4 is also x14
     ld x_dest2, 8(x_dest)
@@ -136,7 +138,7 @@ gf_6vect_dot_prod_rvv:
     /* initialize table pointers */
     /* gf_tbl base = (x_tbl + dest_idx * x_vec * 32) */
     mv x_tbl1, x_tbl
-    slli t0, x_vec, 5
+    slli t0, x_vec, 2
     add x_tbl2, x_tbl1, t0
     add x_tbl3, x_tbl2, t0
     add x_tbl4, x_tbl3, t0
@@ -145,11 +147,7 @@ gf_6vect_dot_prod_rvv:
 
 .Llooprvv_vl_vects:
     /* load source data */
-    slli a6, x_vec_i, 3
-    add a6,x_src,a6
-    ld x_ptr, 0(a6)
     add x_ptr,x_ptr,x_pos
-
     vle8.v v_src, (x_ptr)
 
 
@@ -168,6 +166,11 @@ gf_6vect_dot_prod_rvv:
     vle8.v v_gft2_hi, (x_tbl2)
     addi x_tbl2, x_tbl2, 16
 
+    /* load next source pointer */
+    addi x_vec_i, x_vec_i,8
+    add a6,x_src,x_vec_i
+    ld x_ptr, 0(a6)
+
     vle8.v v_gft3_lo, (x_tbl3)
     addi x_tbl3, x_tbl3, 16
     vle8.v v_gft3_hi, (x_tbl3)
@@ -188,7 +191,6 @@ gf_6vect_dot_prod_rvv:
     vle8.v v_gft6_hi, (x_tbl6)
     addi x_tbl6, x_tbl6, 16
 
-
     /* dest 1 */
     vrgather.vv v26, v_gft1_lo, v_src_lo
     vrgather.vv v27, v_gft1_hi, v_src_hi
@@ -225,10 +227,6 @@ gf_6vect_dot_prod_rvv:
     vxor.vv v_dest6, v_dest6, v26
     vxor.vv v_dest6, v_dest6, v27
 
-
-    /* load next source pointer */
-    addi x_vec_i, x_vec_i,1
-
     /* check if we have processed all vectors */
     blt x_vec_i, x_vec, .Llooprvv_vl_vects
 

erasure_code/riscv64/gf_7vect_dot_prod_rvv.S

@@ -113,7 +113,6 @@ gf_7vect_dot_prod_rvv:
     sd s8, 64(sp)
 
     vsetvli t0, x0, e8, m1
-
     /* initialize position */
     li x_pos, 0
 
@@ -206,7 +205,6 @@ gf_7vect_dot_prod_rvv:
     vle8.v v_gft7_hi, (x_tbl7)
     addi x_tbl7, x_tbl7, 16
 
-
     /* dest 1 */
     vrgather.vv v26, v_gft1_lo, v_src_lo
     vrgather.vv v27, v_gft1_hi, v_src_hi

erasure_code/riscv64/gf_vect_dot_prod_rvv.S

@@ -63,7 +63,6 @@ gf_vect_dot_prod_rvv:
     blt a0, t4, .return_fail
 
     vsetvli t5, zero, e8, m1  # Set vector length to maximum
-
     # Initialize pos = 0
     li t2, 0