Sam's simd debugging

dinatraykova · Feb 21, 2024 · 2a36db1 · 2a36db1
1 parent 446aae8
commit 2a36db1
Show file tree

Hide file tree

Showing 9 changed files with 25 additions and 24 deletions.
diff --git a/Examples/Fluid_Kerr/PerfectFluidLevel.hpp b/Examples/Fluid_Kerr/PerfectFluidLevel.hpp
@@ -31,23 +31,23 @@ class PerfectFluidLevel : public GRAMRLevel
     typedef PerfectFluid<EoS> PerfectFluidEoS;
 
     //! Things to do at the end of the advance step, after RK4 calculation
-    virtual void specificAdvance();
+    virtual void specificAdvance() override;
 
     //! Initialize data for the field and metric variables
-    virtual void initialData();
+    virtual void initialData() override;
 
 #ifdef CH_USE_HDF5
     //! routines to do before outputting plot file
-    virtual void prePlotLevel();
+    virtual void prePlotLevel() override;
 #endif
 
     //! RHS routines used at each RK4 step
     virtual void specificEvalRHS(GRLevelData &a_soln, GRLevelData &a_rhs,
-                                 const double a_time);
+                                 const double a_time) override;
 
     //! Things to do in UpdateODE step, after soln + rhs update
     virtual void specificUpdateODE(GRLevelData &a_soln,
-                                   const GRLevelData &a_rhs, Real a_dt);
+                                   const GRLevelData &a_rhs, Real a_dt) override;
 
     /// Things to do before tagging cells (i.e. filling ghosts)
     virtual void preTagCells() override;

diff --git a/Examples/Fluid_Kerr/PositiveDensity.hpp b/Examples/Fluid_Kerr/PositiveDensity.hpp
@@ -35,9 +35,9 @@ class PositiveDensity
         vi[2] = current_cell.load_vars(c_vi3);
 
         auto make_zero = simd_compare_lt(D, m_min_D);
-        D = simd_conditional(make_zero, D, m_min_D);
+        D = simd_conditional(make_zero, m_min_D, D);
         // tau = simd_conditional(make_zero, tau, 1e-4);
-        FOR(i) vi[i] = simd_conditional(make_zero, vi[i], m_min_v);
+        FOR(i) vi[i] = simd_conditional(make_zero, m_min_v, vi[i]);
 
         // current_cell.store_vars(D, c_D);
         //  current_cell.store_vars(rho, c_rho);

diff --git a/Examples/Fluid_Kerr/PrimitiveRecovery.hpp b/Examples/Fluid_Kerr/PrimitiveRecovery.hpp
@@ -10,9 +10,11 @@
 #include "CCZ4Geometry.hpp"
 #include "Cell.hpp"
 #include "DefaultEoS.hpp"
+#include "simd.hpp"
 #include "Tensor.hpp"
 #include "TensorAlgebra.hpp"
 #include "UserVariables.hpp"
+#include "UsingNamespace.H"
 #include "VarsTools.hpp"
 
 // template <class eos_t = DefaultEoS>
@@ -58,12 +60,10 @@ class PrimitiveRecovery
 
         int i = 0;
 
-	auto cond = simd_all_false(simd_compare_lt(diff, tolerance));
-        //while (simd_all_false(simd_compare_lt(diff, tolerance)))
-        while (i<20)
-	// while (simd_compare_lt(tolerance, diff))
+	data_t empty;
+	while (!simd_all_false(simd_compare_lt(tolerance, diff), empty))
         {
-            i++;
+	    i++;
             Wa = sqrt(pow(xa, 2.) / (pow(xa, 2.) - r));
 
             vars.rho = pow(vars.chi, 1.5) * vars.D / Wa;
@@ -74,7 +74,7 @@ class PrimitiveRecovery
             xn = Wa * (1. + vars.eps + P_over_rho);
             diff = abs(xn - xa);
             xa = xn;
-            if (i >= 20)
+            if (i >= 100)
                 break;
         }
     }

diff --git a/Source/simd/arm/neon.hpp b/Source/simd/arm/neon.hpp
@@ -95,7 +95,7 @@ template <> struct simd<double> : public simd_base<double>
         return vmaxq_f64(a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         uint64x2_t high_bits = vshrq_n_u64(input, 63);
         //Lanes are indexed like bits (big-endian)
@@ -178,7 +178,7 @@ template <> struct simd<float> : public simd_base<float>
         return vmaxq_f32(a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         uint32x4_t high_bits = vshrq_n_u32(input, 31);
         //Lanes are indexed like bits (big-endian)

diff --git a/Source/simd/arm/sve.hpp b/Source/simd/arm/sve.hpp
@@ -122,7 +122,7 @@ template <> struct simd<double> : public simd_base<double>
         return svmax_f64_z(svptrue_b64(), a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         //Would be nice to implement a movemask here, for extensions to lane-specific info
         return !svptest_any(svptrue_b64(), cond);
@@ -212,7 +212,7 @@ template <> struct simd<float> : public simd_base<float>
         return svmax_f32_z(svptrue_b32(), a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         //Would be nice to implement a movemask here, for extensions to lane-specific info
         return !svptest_any(svptrue_b32(), cond);

diff --git a/Source/simd/simd.hpp b/Source/simd/simd.hpp
@@ -175,7 +175,8 @@ template <typename t> ALWAYS_INLINE t simd_max(const t &a, const t &b)
     return (a > b) ? a : b;
 }
 
-template <typename t> ALWAYS_INLINE bool simd_all_false(const bool cond)
+// Here, t is some simd variable
+template <typename t> ALWAYS_INLINE bool simd_all_false(const bool cond, const t &b)
 {
     return !cond;
 }

diff --git a/Source/simd/x64/avx.hpp b/Source/simd/x64/avx.hpp
@@ -109,7 +109,7 @@ template <> struct simd<double> : public simd_base<double>
         return _mm256_max_pd(a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         int mask = _mm256_movemask_pd(cond);
         return mask == 0 ? true : false;
@@ -199,7 +199,7 @@ template <> struct simd<float> : public simd_base<float>
         return _mm256_max_ps(a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         int mask = _mm256_movemask_ps(cond);
         return mask == 0 ? true : false;

diff --git a/Source/simd/x64/avx512.hpp b/Source/simd/x64/avx512.hpp
@@ -103,7 +103,7 @@ template <> struct simd<double> : public simd_base<double>
         return _mm512_max_pd(a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         int mask = static_cast<int>(cond);
         return mask == 0 ? true : false;
@@ -204,7 +204,7 @@ template <> struct simd<float> : public simd_base<float>
         return _mm512_max_ps(a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         int mask = static_cast<int>(cond);
         return mask == 0 ? true : false;

diff --git a/Source/simd/x64/sse.hpp b/Source/simd/x64/sse.hpp
@@ -121,7 +121,7 @@ template <> struct simd<double> : public simd_base<double>
         return _mm_max_pd(a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         int mask = _mm_movemask_pd(cond);
         return mask == 0 ? true : false;
@@ -221,7 +221,7 @@ template <> struct simd<float> : public simd_base<float>
         return _mm_max_ps(a, b);
     }
 
-    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond)
+    friend ALWAYS_INLINE bool simd_all_false(const mask_t cond, const simd &b)
     {
         int mask = _mm_movemask_ps(cond);
         return mask == 0 ? true : false;