x86: add a misc utility header

4 files changed, 60 insertions, 18 deletions

diff --git a/residual_calc.asm b/residual_calc.asm
index 6970fd5..0a85e1d 100644
--- a/residual_calc.asm
+++ b/residual_calc.asm
@@ -18,6 +18,7 @@
 
 %include "config.asm"
 %include "x86inc.asm"
+%include "util.asm"
 
 ; double precision
 %define ELEM_SIZE 8
diff --git a/transfer.c b/transfer.c
index 945470e..bfd0027 100644
--- a/transfer.c
+++ b/transfer.c
@@ -62,11 +62,11 @@ typedef struct GridTransferLagrange {
 } GridTransferLagrange;
 
 #if HAVE_NASM
-void mg2di_transfer_interp2d_line_cont_4_fma3(double *dst, ptrdiff_t dst_len,
+void mg2di_transfer_interp2d_line_cont_4_avx2(double *dst, ptrdiff_t dst_len,
                                               const double *src, ptrdiff_t src_stride,
                                               const ptrdiff_t *idx_x,
                                               const double *fact_x, const double *fact_y);
-void mg2di_transfer_interp2d_line_cont_6_fma3(double *dst, ptrdiff_t dst_len,
+void mg2di_transfer_interp2d_line_cont_6_avx2(double *dst, ptrdiff_t dst_len,
                                               const double *src, ptrdiff_t src_stride,
                                               const ptrdiff_t *idx_x,
                                               const double *fact_x, const double *fact_y);
@@ -177,8 +177,8 @@ static int transfer_lagrange_init(GridTransferContext *ctx)
         priv->stencil = 4;
 
 #if HAVE_NASM
-        if (ctx->cpuflags & MG2DI_CPU_FLAG_FMA3) {
-            priv->transfer2d_cont    = mg2di_transfer_interp2d_line_cont_4_fma3;
+        if (ctx->cpuflags & MG2DI_CPU_FLAG_AVX2) {
+            priv->transfer2d_cont    = mg2di_transfer_interp2d_line_cont_4_avx2;
         }
 #endif
         break;
@@ -189,8 +189,8 @@ static int transfer_lagrange_init(GridTransferContext *ctx)
         priv->transfer2d_generic = interp2d_transfer_line_generic_6;
         priv->stencil = 6;
 #if HAVE_NASM
-        if (ctx->cpuflags & MG2DI_CPU_FLAG_FMA3) {
-            priv->transfer2d_cont    = mg2di_transfer_interp2d_line_cont_6_fma3;
+        if (ctx->cpuflags & MG2DI_CPU_FLAG_AVX2) {
+            priv->transfer2d_cont    = mg2di_transfer_interp2d_line_cont_6_avx2;
         }
 #endif
         break;
diff --git a/transfer_interp.asm b/transfer_interp.asm
index e711b8f..d02e188 100644
--- a/transfer_interp.asm
+++ b/transfer_interp.asm
@@ -18,10 +18,11 @@
 
 %include "config.asm"
 %include "x86inc.asm"
+%include "util.asm"
 
 SECTION .text
 
-INIT_YMM fma3
+INIT_YMM avx2
 cglobal transfer_interp2d_line_cont_4, 7, 8, 6, dst, dst_len, src, src_stride, idx_x, fact_x, fact_y,\
                                                 idx_x_val
     shl src_strideq, 3
@@ -35,10 +36,10 @@ cglobal transfer_interp2d_line_cont_4, 7, 8, 6, dst, dst_len, src, src_stride, i
     %define offsetq dst_lenq
 
     movu m0, [fact_yq]
-    vpermq  m1, m0, 01010101b                           ; fact y + 1 -> m1
-    vpermq  m2, m0, 10101010b                           ; fact y + 2 -> m2
-    vpermq  m3, m0, 11111111b                           ; fact y + 3 -> m3
-    vpermq  m0, m0, 00000000b                           ; fact y + 0 -> m0
+    SPLATPD  m1, m0, 1                           ; fact y + 1 -> m1
+    SPLATPD  m2, m0, 2                           ; fact y + 2 -> m2
+    SPLATPD  m3, m0, 3                           ; fact y + 3 -> m3
+    SPLATPD  m0, m0, 0                           ; fact y + 0 -> m0
 
 .loop:
     mov idx_x_valq, [idx_xq + offsetq]
@@ -73,7 +74,7 @@ cglobal transfer_interp2d_line_cont_4, 7, 8, 6, dst, dst_len, src, src_stride, i
 
     RET
 
-INIT_YMM fma3
+INIT_YMM avx2
 cglobal transfer_interp2d_line_cont_6, 7, 9, 11, dst, dst_len, src, src_stride, idx_x, fact_x, fact_y,\
                                                  idx_x_val, offset6
     shl src_strideq, 3
@@ -90,13 +91,13 @@ cglobal transfer_interp2d_line_cont_6, 7, 9, 11, dst, dst_len, src, src_stride,
     add offset6q, offset6q
 
     movu m0, [fact_yq]
-    vpermq  m1, m0, 01010101b                           ; fact y + 1 -> m1
-    vpermq  m2, m0, 10101010b                           ; fact y + 2 -> m2
-    vpermq  m3, m0, 11111111b                           ; fact y + 3 -> m3
-    vpermq  m0, m0, 00000000b                           ; fact y + 0 -> m0
+    SPLATPD  m1, m0, 1                           ; fact y + 1 -> m1
+    SPLATPD  m2, m0, 2                           ; fact y + 2 -> m2
+    SPLATPD  m3, m0, 3                           ; fact y + 3 -> m3
+    SPLATPD  m0, m0, 0                           ; fact y + 0 -> m0
     movu xm4, [fact_yq + 8 * 4]
-    vpermq m5, m4, 01010101b
-    vpermq m4, m4, 0
+    SPLATPD m5, m4, 1
+    SPLATPD m4, m4, 0
 
 .loop:
     mov idx_x_valq, [idx_xq + offsetq]
diff --git a/util.asm b/util.asm
new file mode 100644
index 0000000..846ea3b
--- /dev/null
+++ b/util.asm
@@ -0,0 +1,40 @@
+;
+; Various ASM utility macros
+; Copyright 2019 Anton Khirnov <anton@khirnov.net>
+;
+; This program is free software: you can redistribute it and/or modify
+; it under the terms of the GNU General Public License as published by
+; the Free Software Foundation, either version 3 of the License, or
+; (at your option) any later version.
+;
+; This program is distributed in the hope that it will be useful,
+; but WITHOUT ANY WARRANTY; without even the implied warranty of
+; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+; GNU General Public License for more details.
+;
+; You should have received a copy of the GNU General Public License
+; along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+%define _IMM8SPLAT2B(x) (((x) << 6) | ((x) << 4) | ((x) << 2) | (x))
+%define _IMM4SPLAT1B(x) (((x) << 3) | ((x) << 2) | ((x) << 1) | (x))
+
+; splat packed double - copy the specified double into all positions in the
+; destination
+
+; %1 destination register
+; %2 source register (may be same as dst)
+; %3 index of the element in the source to splat
+%macro SPLATPD 3
+    %if %3 > 3
+        %error Invalid selector %3
+    %endif
+
+    %if mmsize == 32 && cpuflag(avx2)
+        vpermpd %1, %2, _IMM8SPLAT2B(%3)
+    %elif mmsize == 32 && cpuflag(avx)
+        shufpd %1, %2, %2, _IMM4SPLAT1B(%3 & 1)
+        vperm2f128 %1, %1, ((%3 & 2) >> 1) * 0x11
+    %else
+        %error %? not supported with cpuname
+    %endif
+%endmacro