aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorAnton Khirnov <anton@khirnov.net>2024-04-15 14:06:04 +0200
committerAnton Khirnov <anton@khirnov.net>2024-04-15 14:06:04 +0200
commit982d71cb08f6ccf564c0558c659ae2756bb39ba1 (patch)
tree650c33e4ed1b8c433e551dd6a240a6c87fd00964
parentb5de5a6977c1c0d0b1a76b084582d8b7d758a9ad (diff)
residual_calc: rename stride to u_stride
Make it explicit that it only applies to u, as other arrays are not indexed beyond curent line.
-rw-r--r--residual_calc.asm22
-rw-r--r--residual_calc.c20
2 files changed, 21 insertions, 21 deletions
diff --git a/residual_calc.asm b/residual_calc.asm
index 5eea31c..9cd530d 100644
--- a/residual_calc.asm
+++ b/residual_calc.asm
@@ -166,7 +166,7 @@ SECTION .text
mov diff_coeffs20q, [diff_coeffsq + OFF_DIFF_COEFF_20]
; setup the data pointers and the loop counter
- shl strideq, 3
+ shl u_strideq, 3
shl linesizeq, 3
add dstq, linesizeq
add uq, linesizeq
@@ -182,14 +182,14 @@ SECTION .text
%define offsetq linesizeq
; setup pointers to the line above and below
- lea u_upq, [uq + strideq]
+ lea u_upq, [uq + u_strideq]
mov u_downq, uq
- sub u_downq, strideq
+ sub u_downq, u_strideq
%if stencil == 2
- lea u_up2q, [uq + 2 * strideq]
- neg strideq
- add strideq, u_downq
- %define u_down2q strideq ; reuse the stride register for the u[y-2] line
+ lea u_up2q, [uq + 2 * u_strideq]
+ neg u_strideq
+ add u_strideq, u_downq
+ %define u_down2q u_strideq ; reuse the stride register for the u[y-2] line
movu m15, [const30]
movu m14, [const8]
@@ -267,18 +267,18 @@ SECTION .text
%endmacro
INIT_YMM fma3
-cglobal residual_calc_line_s1, 7, 14, 14, linesize, dst, res_max, stride, u, rhs, diff_coeffs,\
+cglobal residual_calc_line_s1, 7, 14, 14, linesize, dst, res_max, u_stride, u, rhs, diff_coeffs,\
diff_coeffs00, diff_coeffs01, diff_coeffs10, diff_coeffs11, diff_coeffs02, u_down, u_up
RESIDUAL_CALC 1, 0
-cglobal residual_add_line_s1, 7, 14, 14, linesize, dst, res_max, stride, u, rhs, diff_coeffs,\
+cglobal residual_add_line_s1, 7, 14, 14, linesize, dst, res_max, u_stride, u, rhs, diff_coeffs,\
diff_coeffs00, diff_coeffs01, diff_coeffs10, diff_coeffs11, diff_coeffs02, u_down, u_up
RESIDUAL_CALC 1, 1
INIT_YMM fma3
-cglobal residual_calc_line_s2, 7, 15, 16, linesize, dst, res_max, stride, u, rhs, diff_coeffs,\
+cglobal residual_calc_line_s2, 7, 15, 16, linesize, dst, res_max, u_stride, u, rhs, diff_coeffs,\
diff_coeffs00, diff_coeffs01, diff_coeffs10, diff_coeffs11, diff_coeffs02, u_down, u_up, u_up2
RESIDUAL_CALC 2, 0
-cglobal residual_add_line_s2, 7, 15, 16, linesize, dst, res_max, stride, u, rhs, diff_coeffs,\
+cglobal residual_add_line_s2, 7, 15, 16, linesize, dst, res_max, u_stride, u, rhs, diff_coeffs,\
diff_coeffs00, diff_coeffs01, diff_coeffs10, diff_coeffs11, diff_coeffs02, u_down, u_up, u_up2
RESIDUAL_CALC 2, 1
diff --git a/residual_calc.c b/residual_calc.c
index a4fd265..c06c966 100644
--- a/residual_calc.c
+++ b/residual_calc.c
@@ -32,11 +32,11 @@
#include "residual_calc.h"
typedef void ResidualLineCalc(size_t linesize, double *dst, double *dst_max,
- ptrdiff_t stride, const double *u, const double *rhs,
+ ptrdiff_t u_stride, const double *u, const double *rhs,
const double * const diff_coeffs[MG2D_DIFF_COEFF_NB],
double res_mult);
typedef void ResidualLineAdd (size_t linesize, double *dst, double *dst_max,
- ptrdiff_t stride, const double *u, const double *rhs,
+ ptrdiff_t u_stride, const double *u, const double *rhs,
const double * const diff_coeffs[MG2D_DIFF_COEFF_NB],
double res_mult, double u_mult);
@@ -141,7 +141,7 @@ derivatives_calc_s2(double *dst, const double *u, ptrdiff_t stride)
}
static void residual_calc_line_s1_c(size_t linesize, double *dst, double *dst_max,
- ptrdiff_t stride, const double *u, const double *rhs,
+ ptrdiff_t u_stride, const double *u, const double *rhs,
const double * const diff_coeffs[MG2D_DIFF_COEFF_NB],
double res_mult)
{
@@ -150,7 +150,7 @@ static void residual_calc_line_s1_c(size_t linesize, double *dst, double *dst_ma
double u_vals[MG2D_DIFF_COEFF_NB];
double res;
- derivatives_calc_s1(u_vals, u + i, stride);
+ derivatives_calc_s1(u_vals, u + i, u_stride);
res = -rhs[i];
for (int j = 0; j < ARRAY_ELEMS(u_vals); j++)
@@ -165,7 +165,7 @@ static void residual_calc_line_s1_c(size_t linesize, double *dst, double *dst_ma
}
static void residual_add_line_s1_c(size_t linesize, double *dst, double *dst_max,
- ptrdiff_t stride, const double *u, const double *rhs,
+ ptrdiff_t u_stride, const double *u, const double *rhs,
const double * const diff_coeffs[MG2D_DIFF_COEFF_NB],
double res_mult, double u_mult)
{
@@ -174,7 +174,7 @@ static void residual_add_line_s1_c(size_t linesize, double *dst, double *dst_max
double u_vals[MG2D_DIFF_COEFF_NB];
double res;
- derivatives_calc_s1(u_vals, u + i, stride);
+ derivatives_calc_s1(u_vals, u + i, u_stride);
res = -rhs[i];
for (int j = 0; j < ARRAY_ELEMS(u_vals); j++)
@@ -189,7 +189,7 @@ static void residual_add_line_s1_c(size_t linesize, double *dst, double *dst_max
}
static void residual_calc_line_s2_c(size_t linesize, double *dst, double *dst_max,
- ptrdiff_t stride, const double *u, const double *rhs,
+ ptrdiff_t u_stride, const double *u, const double *rhs,
const double * const diff_coeffs[MG2D_DIFF_COEFF_NB],
double res_mult)
{
@@ -198,7 +198,7 @@ static void residual_calc_line_s2_c(size_t linesize, double *dst, double *dst_ma
double u_vals[MG2D_DIFF_COEFF_NB];
double res;
- derivatives_calc_s2(u_vals, u + i, stride);
+ derivatives_calc_s2(u_vals, u + i, u_stride);
res = -rhs[i];
for (int j = 0; j < ARRAY_ELEMS(u_vals); j++)
@@ -213,7 +213,7 @@ static void residual_calc_line_s2_c(size_t linesize, double *dst, double *dst_ma
}
static void residual_add_line_s2_c(size_t linesize, double *dst, double *dst_max,
- ptrdiff_t stride, const double *u, const double *rhs,
+ ptrdiff_t u_stride, const double *u, const double *rhs,
const double * const diff_coeffs[MG2D_DIFF_COEFF_NB],
double res_mult, double u_mult)
{
@@ -222,7 +222,7 @@ static void residual_add_line_s2_c(size_t linesize, double *dst, double *dst_max
double u_vals[MG2D_DIFF_COEFF_NB];
double res;
- derivatives_calc_s2(u_vals, u + i, stride);
+ derivatives_calc_s2(u_vals, u + i, u_stride);
res = -rhs[i];
for (int j = 0; j < ARRAY_ELEMS(u_vals); j++)