[FFmpeg-devel] [PATCH 1/2] avcodec/lpc: zero the middle odd sample in the output

[FFmpeg-devel] [PATCH 1/2] avcodec/lpc: zero the middle odd sample in the output

[James Almer]

Signed-off-by: James Almer <jamrial@gmail.com>
---
 libavcodec/lpc.c       |  1 +
 libavcodec/x86/lpc.asm | 10 +++++++---
 2 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/libavcodec/lpc.c b/libavcodec/lpc.c
index 8603bb9709..dc6a3060ce 100644
--- a/libavcodec/lpc.c
+++ b/libavcodec/lpc.c
@@ -53,6 +53,7 @@ static void lpc_apply_welch_window_c(const int32_t *data, ptrdiff_t len,
             w_data[i] = data[i] * w;
             w_data[len-1-i] = data[len-1-i] * w;
         }
+        w_data[n2] = 0.0;
         return;
     }
 
diff --git a/libavcodec/x86/lpc.asm b/libavcodec/x86/lpc.asm
index ad74f1d8ac..47b31aa147 100644
--- a/libavcodec/x86/lpc.asm
+++ b/libavcodec/x86/lpc.asm
@@ -37,7 +37,7 @@ SECTION .text
 %macro APPLY_WELCH_FN 0
 cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
     cmp lenq, 0
-    je .end
+    je .end_e
     cmp lenq, 2
     je .two
     cmp lenq, 1
@@ -104,7 +104,7 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
 
     add lend, (mmsize/4 - 1)
     cmp lend, 0
-    je .end
+    je .end_o
     sub lenq, (mmsize/4 - 1)
 
 .scalar_o:
@@ -135,6 +135,10 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
 
     sub lenq, 2
     jg .loop_o_scalar
+
+.end_o:
+    xorps xm3, xm3
+    movlpd [outq + off1q*2], xm3
     RET
 
 .even:
@@ -233,7 +237,7 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
 .one:
     xorpd xm0, xm0
     movhpd [outq], xm0
-.end:
+.end_e:
     RET
 %endmacro
 
-- 
2.37.3

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH 2/2] tests/checkasm/lpc: randomize buffer length

[James Almer]

Simplifies the test, while trying more values and preventing pointlessly
running benchmarks in a loop.

Signed-off-by: James Almer <jamrial@gmail.com>
---
 tests/checkasm/lpc.c | 19 +++----------------
 1 file changed, 3 insertions(+), 16 deletions(-)

diff --git a/tests/checkasm/lpc.c b/tests/checkasm/lpc.c
index da5364def0..8528fd6e20 100644
--- a/tests/checkasm/lpc.c
+++ b/tests/checkasm/lpc.c
@@ -54,31 +54,18 @@ static void test_window(int len)
 void checkasm_check_lpc(void)
 {
     LPCContext ctx;
+    int len = rnd() % 5000;
     ff_lpc_init(&ctx, 32, 16, FF_LPC_TYPE_DEFAULT);
 
     if (check_func(ctx.lpc_apply_welch_window, "apply_welch_window_even")) {
-        for (int i = 0; i < 64; i += 2)
-            test_window(i);
+        test_window(len & ~1);
     }
     report("apply_welch_window_even");
 
     if (check_func(ctx.lpc_apply_welch_window, "apply_welch_window_odd")) {
-        for (int i = 1; i < 64; i += 2)
-            test_window(i);
+        test_window(len | 1);
     }
     report("apply_welch_window_odd");
 
-    if (check_func(ctx.lpc_apply_welch_window, "apply_welch_window_2560"))
-        test_window(2560);
-    report("apply_welch_window_2560");
-
-    if (check_func(ctx.lpc_apply_welch_window, "apply_welch_window_4096"))
-        test_window(4096);
-    report("apply_welch_window_4096");
-
-    if (check_func(ctx.lpc_apply_welch_window, "apply_welch_window_4097"))
-        test_window(4097);
-    report("apply_welch_window_4097");
-
     ff_lpc_end(&ctx);
 }
-- 
2.37.3

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH 3/3] x86/lpc: use fused negative multiply-add instructions where useful

[James Almer]

Signed-off-by: James Almer <jamrial@gmail.com>
---
 libavcodec/x86/lpc.asm | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/libavcodec/x86/lpc.asm b/libavcodec/x86/lpc.asm
index 61a5796e5d..a585c17ef5 100644
--- a/libavcodec/x86/lpc.asm
+++ b/libavcodec/x86/lpc.asm
@@ -79,11 +79,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
 
 .loop_o:
     movapd m1, m6
-    mulpd m2, m0, m0
-    subpd m1, m2
 %if cpuflag(avx2)
+    fnmaddpd m1, m0, m0, m1
     vpermpd m2, m1, q0123
 %else
+    mulpd m2, m0, m0
+    subpd m1, m2
     shufpd m2, m1, m1, 01b
 %endif
 
@@ -116,8 +117,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
 
 .loop_o_scalar:
     movapd xm1, xm6
+%if cpuflag(avx2)
+    fnmaddpd xm1, xm0, xm0, xm1
+%else
     mulpd xm2, xm0, xm0
     subpd xm1, xm2
+%endif
 
     cvtdq2pd xm3, [dataq + off1q]
     cvtdq2pd xm4, [dataq + off2q]
@@ -174,8 +179,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
 
 .loop_e:
     movapd m1, m6
+%if cpuflag(avx2)
+    fnmaddpd m1, m0, m0, m1
+%else
     mulpd m2, m0, m0
     subpd m1, m2
+%endif
 %if cpuflag(avx2)
     vpermpd m2, m1, q0123
 %else
@@ -210,8 +219,12 @@ cglobal lpc_apply_welch_window, 3, 5, 8, data, len, out, off1, off2
 
 .loop_e_scalar:
     movapd xm1, xm6
+%if cpuflag(avx2)
+    fnmaddpd xm1, xm0, xm0, xm1
+%else
     mulpd xm2, xm0, xm0
     subpd xm1, xm2
+%endif
 
     cvtdq2pd xm3, [dataq + off1q]
     cvtdq2pd xm4, [dataq + off2q]
-- 
2.37.3

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".