Add scaling and clipping to SSE float-int and int-float conversions

git-svn-id: svn+ssh://jackaudio.org/trunk/jack@3053 0c269be4-1314-0410-8aa9-9f06e86f4224

3 files changed, 49 insertions, 16 deletions

diff --git a/configure.ac b/configure.ac
index e81e7ea..e8f523e 100644
--- a/configure.ac
+++ b/configure.ac
@@ -17,7 +17,7 @@ dnl                 changes are made
 dnl ---
 JACK_MAJOR_VERSION=0
 JACK_MINOR_VERSION=115
-JACK_MICRO_VERSION=0
+JACK_MICRO_VERSION=1
 
 dnl ---
 dnl HOWTO: updating the jack protocol version
diff --git a/jack/intsimd.h b/jack/intsimd.h
index 29b3b7c..2377bd1 100644
--- a/jack/intsimd.h
+++ b/jack/intsimd.h
@@ -45,8 +45,8 @@ void x86_3dnow_copyf (float *, const float *, int);
 void x86_3dnow_add2f (float *, const float *, int);
 void x86_sse_copyf (float *, const float *, int);
 void x86_sse_add2f (float *, const float *, int);
-void x86_sse_f2i (int *, const float *, int);
-void x86_sse_i2f (float *, const int *, int);
+void x86_sse_f2i (int *, const float *, int, float);
+void x86_sse_i2f (float *, const int *, int, float);
 
 #endif /* ARCH_X86 */
 
diff --git a/libjack/simd.c b/libjack/simd.c
index def5656..6fabb85 100644
--- a/libjack/simd.c
+++ b/libjack/simd.c
@@ -329,20 +329,40 @@ sse_nonalign:
 	}
 }
 
-void x86_sse_f2i (int *dest, const float *src, int length)
+void x86_sse_f2i (int *dest, const float *src, int length, float scale)
 {
 	int i;
+	float max[4] __attribute__((aligned(16))) =
+		{ -1.0F, -1.0F, -1.0F, -1.0F };
+	float min[4] __attribute__((aligned(16))) =
+		{ 1.0F, 1.0F, 1.0F, 1.0F };
+	float s[4] __attribute__((aligned(16)));
+
+	s[0] = s[1] = s[2] = s[3] = scale;
+	asm volatile (
+		"movaps %0, %%xmm4\n\t" \
+		"movaps %1, %%xmm5\n\t" \
+		"movaps %2, %%xmm6\n\t"
+		:
+		: "m" (*max),
+		  "m" (*min),
+		  "m" (*s)
+		: "xmm4", "xmm5", "xmm6");
 
 	if (__builtin_expect((((long) dest & 0xf) || ((long) src & 0xf)), 0))
 		goto sse_nonalign;
 	for (i = 0; i < length; i += 4)
 	{
 		asm volatile (
-			"cvtps2dq %1, %%xmm0\n\t" \
+			"movaps %1, %%xmm1\n\t" \
+			"maxps %%xmm4, %%xmm1\n\t" \
+			"minps %%xmm5, %%xmm1\n\t" \
+			"mulps %%xmm6, %%xmm1\n\t" \
+			"cvtps2dq %%xmm1, %%xmm0\n\t" \
 			"movdqa %%xmm0, %0\n\t"
 			: "=m" (dest[i])
 			: "m" (src[i])
-			: "xmm0", "memory");
+			: "xmm0", "xmm1", "xmm4", "xmm5", "xmm6", "memory");
 	}
 	return;
 
@@ -350,19 +370,30 @@ sse_nonalign:
 	for (i = 0; i < length; i += 4)
 	{
 		asm volatile (
-			"movups %1, %%xmm0\n\t" \
-			"cvtps2dq %%xmm0, %%xmm1\n\t" \
-			"movdqu %%xmm1, %0\n\t"
+			"movups %1, %%xmm1\n\t" \
+			"maxps %%xmm4, %%xmm1\n\t" \
+			"minps %%xmm5, %%xmm1\n\t" \
+			"mulps %%xmm6, %%xmm1\n\t" \
+			"cvtps2dq %%xmm1, %%xmm0\n\t" \
+			"movdqu %%xmm0, %0\n\t"
 			: "=m" (dest[i])
 			: "m" (src[i])
-			: "xmm0", "xmm1", "memory");
+			: "xmm0", "xmm1", "xmm4", "xmm5", "xmm6", "memory");
 	}
 }
 
 
-void x86_sse_i2f (float *dest, const int *src, int length)
+void x86_sse_i2f (float *dest, const int *src, int length, float scale)
 {
 	int i;
+	float s[4] __attribute__((aligned(16)));
+
+	s[0] = s[1] = s[2] = s[3] = scale;
+	asm volatile (
+		"movaps %0, %%xmm4\n\t"
+		:
+		: "m" (*s)
+		: "xmm4" );
 
 	if (__builtin_expect((((long) dest & 0xf) || ((long) src & 0xf)), 0))
 		goto sse_nonalign; 
@@ -370,10 +401,11 @@ void x86_sse_i2f (float *dest, const int *src, int length)
 	{
 		asm volatile (
 			"cvtdq2ps %1, %%xmm0\n\t" \
+			"mulps %%xmm4, %%xmm0\n\t" \
 			"movaps %%xmm0, %0\n\t"
 			: "=m" (dest[i])
 			: "m" (src[i])
-			: "xmm0", "memory");
+			: "xmm0", "xmm4", "memory");
 	}
 	return;
 
@@ -381,12 +413,13 @@ sse_nonalign:
 	for (i = 0; i < length; i += 4)
 	{
 		asm volatile (
-			"movdqu %1, %%xmm0\n\t" \
-			"cvtdq2ps %%xmm0, %%xmm1\n\t" \
-			"movups %%xmm1, %0\n\t"
+			"movdqu %1, %%xmm1\n\t" \
+			"cvtdq2ps %%xmm1, %%xmm0\n\t" \
+			"mulps %%xmm4, %%xmm0\n\t" \
+			"movups %%xmm0, %0\n\t"
 			: "=m" (dest[i])
 			: "m" (src[i])
-			: "xmm0", "xmm1", "memory");
+			: "xmm0", "xmm1", "xmm4", "memory");
 	}
 }