From 12ff721ec1cb52a3328c3c88d81040485fa62c64 Mon Sep 17 00:00:00 2001
From: Vsevolod Livinskiy <vsevolod.livinskij@frtk.ru>
Date: Thu, 14 May 2015 16:13:08 +0300
Subject: [PATCH] Add roundings for varying double in knc.h and knl.h

---
 examples/intrinsics/knc.h | 21 +++++++++++++++++++++
 examples/intrinsics/knl.h | 18 ++++++++++++++++++
 tests/test-149.ispc       | 12 ++++++++++++
 tests/test-150.ispc       | 12 ++++++++++++
 tests/test-151.ispc       | 12 ++++++++++++
 tests/test-152.ispc       | 12 ++++++++++++
 tests/test-153.ispc       | 12 ++++++++++++
 7 files changed, 99 insertions(+)
 create mode 100644 tests/test-149.ispc
 create mode 100644 tests/test-150.ispc
 create mode 100644 tests/test-151.ispc
 create mode 100644 tests/test-152.ispc
 create mode 100644 tests/test-153.ispc

diff --git a/examples/intrinsics/knc.h b/examples/intrinsics/knc.h
index 3a097ce6..dc2a3a19 100644
--- a/examples/intrinsics/knc.h
+++ b/examples/intrinsics/knc.h
@@ -2675,6 +2675,27 @@ static FORCEINLINE __vec16_f __ceil_varying_float(__vec16_f v) {
   return _mm512_ceil_ps(v);
 }
 
+static FORCEINLINE __vec16_d __round_varying_double(__vec16_d v) {
+  double tmp [16];
+  for (int i = 0; i < 16; i++)
+    tmp [i] = round(v [i]);
+  return __vec16_d (tmp [8],  tmp [9],  tmp [10], tmp [11],
+                    tmp [12], tmp [13], tmp [14], tmp [15],
+                    tmp [0],  tmp [1],  tmp [2],  tmp [3],
+                    tmp [4],  tmp [5],  tmp [6],  tmp [7]);
+}
+
+static FORCEINLINE __vec16_d __floor_varying_double(__vec16_d v) {
+  __m512d tmp1 = _mm512_floor_pd(v.v1);
+  __m512d tmp2 = _mm512_floor_pd(v.v2);
+  return __vec16_d (tmp1, tmp2);
+}
+
+static FORCEINLINE __vec16_d __ceil_varying_double(__vec16_d v) {
+  __m512d tmp1 = _mm512_ceil_pd(v.v1);
+  __m512d tmp2 = _mm512_ceil_pd(v.v2);
+  return __vec16_d (tmp1, tmp2);
+}
 // min/max
 
 static FORCEINLINE float __min_uniform_float(float a, float b) { return (a<b) ? a : b; }
diff --git a/examples/intrinsics/knl.h b/examples/intrinsics/knl.h
index 6c3602b0..c0ed80ea 100644
--- a/examples/intrinsics/knl.h
+++ b/examples/intrinsics/knl.h
@@ -2536,6 +2536,24 @@ static FORCEINLINE __vec16_f __ceil_varying_float(__vec16_f v) {
   return _mm512_ceil_ps(v);
 }
 
+static FORCEINLINE __vec16_d __round_varying_double(__vec16_d v) {
+  __m512d tmp1 =_mm512_cvtepi32_pd(_mm512_cvtpd_epi32(v.v_lo));
+  __m512d tmp2 =_mm512_cvtepi32_pd(_mm512_cvtpd_epi32(v.v_hi));
+  return __vec16_d (tmp1, tmp2);
+}
+
+static FORCEINLINE __vec16_d __floor_varying_double(__vec16_d v) {
+  __m512d tmp1 = _mm512_floor_pd(v.v_lo);
+  __m512d tmp2 = _mm512_floor_pd(v.v_hi);
+  return __vec16_d (tmp1, tmp2);
+}
+
+static FORCEINLINE __vec16_d __ceil_varying_double(__vec16_d v) {
+  __m512d tmp1 = _mm512_ceil_pd(v.v_lo);
+  __m512d tmp2 = _mm512_ceil_pd(v.v_hi);
+  return __vec16_d (tmp1, tmp2);
+}
+
 // min/max
 
 static FORCEINLINE float __min_uniform_float(float a, float b) { return (a<b) ? a : b; }
diff --git a/tests/test-149.ispc b/tests/test-149.ispc
new file mode 100644
index 00000000..e3997678
--- /dev/null
+++ b/tests/test-149.ispc
@@ -0,0 +1,12 @@
+
+export uniform int width() { return programCount; }
+
+
+export void f_f(uniform float RET[], uniform float aFOO[]) {
+    double a = aFOO[programIndex]; 
+    RET[programIndex] = round(a+.49999); 
+}
+
+export void result(uniform float RET[]) {
+    RET[programIndex] = 1 + (double) programIndex;
+}
diff --git a/tests/test-150.ispc b/tests/test-150.ispc
new file mode 100644
index 00000000..97c95a30
--- /dev/null
+++ b/tests/test-150.ispc
@@ -0,0 +1,12 @@
+
+export uniform int width() { return programCount; }
+
+
+export void f_f(uniform float RET[], uniform float aFOO[]) {
+    double a = aFOO[programIndex]; 
+    RET[programIndex] = round(a+(0.50001d));
+}
+
+export void result(uniform float RET[]) {
+    RET[programIndex] = programIndex + 2;
+}
diff --git a/tests/test-151.ispc b/tests/test-151.ispc
new file mode 100644
index 00000000..488ecf5a
--- /dev/null
+++ b/tests/test-151.ispc
@@ -0,0 +1,12 @@
+
+export uniform int width() { return programCount; }
+
+
+export void f_f(uniform float RET[], uniform float aFOO[]) {
+    double a = aFOO[programIndex]; 
+    RET[programIndex] = floor(a+.99999); 
+}
+
+export void result(uniform float RET[]) {
+    RET[programIndex] = 1 +(double) programIndex;
+}
diff --git a/tests/test-152.ispc b/tests/test-152.ispc
new file mode 100644
index 00000000..3aaf27da
--- /dev/null
+++ b/tests/test-152.ispc
@@ -0,0 +1,12 @@
+
+export uniform int width() { return programCount; }
+
+
+export void f_f(uniform float RET[], uniform float aFOO[]) {
+    double a = aFOO[programIndex]; 
+    RET[programIndex] = ceil(a-1e-5); 
+}
+
+export void result(uniform float RET[]) {
+    RET[programIndex] = 1 + (double) programIndex;
+}
diff --git a/tests/test-153.ispc b/tests/test-153.ispc
new file mode 100644
index 00000000..7672427b
--- /dev/null
+++ b/tests/test-153.ispc
@@ -0,0 +1,12 @@
+
+export uniform int width() { return programCount; }
+
+
+export void f_f(uniform float RET[], uniform float aFOO[]) {
+    double a = aFOO[programIndex]; 
+    RET[programIndex] = ceil(-a+1e-5); 
+}
+
+export void result(uniform float RET[]) {
+    RET[programIndex] = - (double) programIndex;
+}