X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fllvmpipe%2Flp_test_arit.c;h=184e50089f793346f401418daf68d507133e9ab1;hb=0e30c6b8a7e84211bb417362ec73f24ef134ae34;hp=f0e43e0f9ccd2993d0690519fd52cb82db18b70c;hpb=0a1d49504de4d34b003625ee7c901667afa43dea;p=mesa.git

diff --git a/src/gallium/drivers/llvmpipe/lp_test_arit.c b/src/gallium/drivers/llvmpipe/lp_test_arit.c
index f0e43e0f9cc..184e50089f7 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_arit.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_arit.c
@@ -32,8 +32,11 @@
 
 #include "util/u_pointer.h"
 #include "util/u_memory.h"
+#include "util/u_math.h"
+#include "util/u_cpu_detect.h"
 
 #include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_debug.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_arit.h"
 
@@ -51,7 +54,7 @@ write_tsv_header(FILE *fp)
 }
 
 
-typedef float (*unary_func_t)(float);
+typedef void (*unary_func_t)(float *out, const float *in);
 
 
 /**
@@ -79,10 +82,56 @@ struct unary_test_t
     */
    const float *values;
    unsigned num_values;
+
+   /*
+    * Required precision in bits.
+    */
+   double precision;
+};
+
+
+static float negf(float x)
+{
+   return -x;
+}
+
+
+static float sgnf(float x)
+{
+   if (x > 0.0f) {
+      return 1.0f;
+   }
+   if (x < 0.0f) {
+      return -1.0f;
+   }
+   return 0.0f;
+}
+
+
+const float sgn_values[] = {
+   -INFINITY,
+   -60,
+   -4,
+   -2,
+   -1,
+   -1e-007,
+   0,
+   1e-007,
+   0.01,
+   0.1,
+   0.9,
+   0.99,
+   1,
+   2,
+   4,
+   60,
+   INFINITY,
+   NAN
 };
 
 
 const float exp2_values[] = {
+   -INFINITY,
    -60,
    -4,
    -2,
@@ -90,10 +139,16 @@ const float exp2_values[] = {
    -1e-007,
    0,
    1e-007,
+   0.01,
+   0.1,
+   0.9,
+   0.99,
    1, 
    2, 
    4, 
-   60
+   60,
+   INFINITY,
+   NAN
 };
 
 
@@ -105,44 +160,132 @@ const float log2_values[] = {
     */
    1.4012984643248171e-45,
 #endif
+   -INFINITY,
+   0,
    1e-007,
+   0.1,
    0.5,
+   0.99,
    1,
+   1.01,
+   1.1,
+   1.9,
+   1.99,
    2,
    4,
    100000,
-   1e+018
+   1e+018,
+   INFINITY,
+   NAN
+};
+
+
+static float rcpf(float x)
+{
+   return 1.0/x;
+}
+
+
+const float rcp_values[] = {
+   -0.0, 0.0,
+   -1.0, 1.0,
+   -1e-007, 1e-007,
+   -4.0, 4.0,
+   -1e+035, -100000,
+   100000, 1e+035,
+   5.88e-39f, // denormal
+#if (__STDC_VERSION__ >= 199901L)
+   INFINITY, -INFINITY,
+#endif
 };
 
 
 static float rsqrtf(float x)
 {
-   return 1.0/sqrt(x);
+   return 1.0/(float)sqrt(x);
 }
 
 
 const float rsqrt_values[] = {
-   -1, -1e-007,
-   1e-007, 1,
-   -4, -1,
-   1, 4,
-   -1e+035, -100000,
+   // http://msdn.microsoft.com/en-us/library/windows/desktop/bb147346.aspx
+   0.0, // must yield infinity
+   1.0, // must yield 1.0
+   1e-007, 4.0,
    100000, 1e+035,
+   5.88e-39f, // denormal
+#if (__STDC_VERSION__ >= 199901L)
+   INFINITY,
+#endif
 };
 
 
 const float sincos_values[] = {
+   -INFINITY,
    -5*M_PI/4,
    -4*M_PI/4,
    -4*M_PI/4,
    -3*M_PI/4,
    -2*M_PI/4,
    -1*M_PI/4,
-    1*M_PI/4,
-    2*M_PI/4,
-    3*M_PI/4,
-    4*M_PI/4,
-    5*M_PI/4,
+   1*M_PI/4,
+   2*M_PI/4,
+   3*M_PI/4,
+   4*M_PI/4,
+   5*M_PI/4,
+   INFINITY,
+   NAN
+};
+
+const float round_values[] = {
+      -10.0, -1, 0.0, 12.0,
+      -1.49, -0.25, 1.25, 2.51,
+      -0.99, -0.01, 0.01, 0.99,
+      -1.5, -0.5, 0.5, 1.5,
+      1.401298464324817e-45f, // smallest denormal
+      -1.401298464324817e-45f,
+      1.62981451e-08f,
+      -1.62981451e-08f,
+      1.62981451e15f, // large number not representable as 32bit int
+      -1.62981451e15f,
+      FLT_EPSILON,
+      -FLT_EPSILON,
+      1.0f - 0.5f*FLT_EPSILON,
+      -1.0f + FLT_EPSILON,
+      FLT_MAX,
+      -FLT_MAX
+};
+
+static float fractf(float x)
+{
+   x -= floorf(x);
+   if (x >= 1.0f) {
+      // clamp to the largest number smaller than one
+      x = 1.0f - 0.5f*FLT_EPSILON;
+   }
+   return x;
+}
+
+
+const float fract_values[] = {
+   // http://en.wikipedia.org/wiki/IEEE_754-1985#Examples
+   0.0f,
+   -0.0f,
+   1.0f,
+   -1.0f,
+   0.5f,
+   -0.5f,
+   1.401298464324817e-45f, // smallest denormal
+   -1.401298464324817e-45f,
+   5.88e-39f, // middle denormal
+   1.18e-38f, // largest denormal
+   -1.18e-38f,
+   -1.62981451e-08f,
+   FLT_EPSILON,
+   -FLT_EPSILON,
+   1.0f - 0.5f*FLT_EPSILON,
+   -1.0f + FLT_EPSILON,
+   FLT_MAX,
+   -FLT_MAX
 };
 
 
@@ -150,14 +293,25 @@ const float sincos_values[] = {
  * Unary test cases.
  */
 
-static const struct unary_test_t unary_tests[] = {
-   {"exp2", &lp_build_exp2, &exp2f, exp2_values, Elements(exp2_values)},
-   {"log2", &lp_build_log2, &log2f, log2_values, Elements(log2_values)},
-   {"exp", &lp_build_exp, &expf, exp2_values, Elements(exp2_values)},
-   {"log", &lp_build_log, &logf, log2_values, Elements(log2_values)},
-   {"rsqrt", &lp_build_rsqrt, &rsqrtf, rsqrt_values, Elements(rsqrt_values)},
-   {"sin", &lp_build_sin, &sinf, sincos_values, Elements(sincos_values)},
-   {"cos", &lp_build_cos, &cosf, sincos_values, Elements(sincos_values)},
+static const struct unary_test_t
+unary_tests[] = {
+   {"abs", &lp_build_abs, &fabsf, sgn_values, ARRAY_SIZE(sgn_values), 20.0 },
+   {"neg", &lp_build_negate, &negf, sgn_values, ARRAY_SIZE(sgn_values), 20.0 },
+   {"sgn", &lp_build_sgn, &sgnf, sgn_values, ARRAY_SIZE(sgn_values), 20.0 },
+   {"exp2", &lp_build_exp2, &exp2f, exp2_values, ARRAY_SIZE(exp2_values), 18.0 },
+   {"log2", &lp_build_log2_safe, &log2f, log2_values, ARRAY_SIZE(log2_values), 20.0 },
+   {"exp", &lp_build_exp, &expf, exp2_values, ARRAY_SIZE(exp2_values), 18.0 },
+   {"log", &lp_build_log_safe, &logf, log2_values, ARRAY_SIZE(log2_values), 20.0 },
+   {"rcp", &lp_build_rcp, &rcpf, rcp_values, ARRAY_SIZE(rcp_values), 20.0 },
+   {"rsqrt", &lp_build_rsqrt, &rsqrtf, rsqrt_values, ARRAY_SIZE(rsqrt_values), 20.0 },
+   {"sin", &lp_build_sin, &sinf, sincos_values, ARRAY_SIZE(sincos_values), 20.0 },
+   {"cos", &lp_build_cos, &cosf, sincos_values, ARRAY_SIZE(sincos_values), 20.0 },
+   {"sgn", &lp_build_sgn, &sgnf, sgn_values, ARRAY_SIZE(sgn_values), 20.0 },
+   {"round", &lp_build_round, &nearbyintf, round_values, ARRAY_SIZE(round_values), 24.0 },
+   {"trunc", &lp_build_trunc, &truncf, round_values, ARRAY_SIZE(round_values), 24.0 },
+   {"floor", &lp_build_floor, &floorf, round_values, ARRAY_SIZE(round_values), 24.0 },
+   {"ceil", &lp_build_ceil, &ceilf, round_values, ARRAY_SIZE(round_values), 24.0 },
+   {"fract", &lp_build_fract_safe, &fractf, fract_values, ARRAY_SIZE(fract_values), 24.0 },
 };
 
 
@@ -166,108 +320,197 @@ static const struct unary_test_t unary_tests[] = {
  */
 static LLVMValueRef
 build_unary_test_func(struct gallivm_state *gallivm,
-                      LLVMModuleRef module,
-                      LLVMContextRef context,
-                      const struct unary_test_t *test)
+                      const struct unary_test_t *test,
+                      unsigned length,
+                      const char *test_name)
 {
-   LLVMTypeRef i32t = LLVMInt32TypeInContext(context);
-   LLVMTypeRef f32t = LLVMFloatTypeInContext(context);
-   LLVMTypeRef v4f32t = LLVMVectorType(f32t, 4);
-   LLVMTypeRef args[1] = { f32t };
-   LLVMValueRef func = LLVMAddFunction(module, test->name, LLVMFunctionType(f32t, args, Elements(args), 0));
-   LLVMValueRef arg1 = LLVMGetParam(func, 0);
+   struct lp_type type = lp_type_float_vec(32, length * 32);
+   LLVMContextRef context = gallivm->context;
+   LLVMModuleRef module = gallivm->module;
+   LLVMTypeRef vf32t = lp_build_vec_type(gallivm, type);
+   LLVMTypeRef args[2] = { LLVMPointerType(vf32t, 0), LLVMPointerType(vf32t, 0) };
+   LLVMValueRef func = LLVMAddFunction(module, test_name,
+                                       LLVMFunctionType(LLVMVoidTypeInContext(context),
+                                                        args, ARRAY_SIZE(args), 0));
+   LLVMValueRef arg0 = LLVMGetParam(func, 0);
+   LLVMValueRef arg1 = LLVMGetParam(func, 1);
    LLVMBuilderRef builder = gallivm->builder;
    LLVMBasicBlockRef block = LLVMAppendBasicBlockInContext(context, func, "entry");
-   LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
    LLVMValueRef ret;
 
    struct lp_build_context bld;
 
-   lp_build_context_init(&bld, gallivm, lp_float32_vec4_type());
+   lp_build_context_init(&bld, gallivm, type);
 
    LLVMSetFunctionCallConv(func, LLVMCCallConv);
 
    LLVMPositionBuilderAtEnd(builder, block);
    
-   /* scalar to vector */
-   arg1 = LLVMBuildInsertElement(builder, LLVMGetUndef(v4f32t), arg1, index0, "");
+   arg1 = LLVMBuildLoad(builder, arg1, "");
 
    ret = test->builder(&bld, arg1);
    
-   /* vector to scalar */
-   ret = LLVMBuildExtractElement(builder, ret, index0, "");
+   LLVMBuildStore(builder, ret, arg0);
+
+   LLVMBuildRetVoid(builder);
+
+   gallivm_verify_function(gallivm, func);
 
-   LLVMBuildRet(builder, ret);
    return func;
 }
 
 
+/*
+ * Flush denorms to zero.
+ */
+static float
+flush_denorm_to_zero(float val)
+{
+   /*
+    * If we have a denorm manually set it to (+-)0.
+    * This is because the reference may or may not do the right thing
+    * otherwise because we want the result according to treating all
+    * denormals as zero (FTZ/DAZ). Not using fpclassify because
+    * a) some compilers are stuck at c89 (msvc)
+    * b) not sure it reliably works with non-standard ftz/daz mode
+    * And, right now we only disable denorms with jited code on x86/sse
+    * (albeit this should be classified as a bug) so to get results which
+    * match we must only flush them to zero here in that case too.
+    */
+   union fi fi_val;
+
+   fi_val.f = val;
+
+#if defined(PIPE_ARCH_SSE)
+   if (util_cpu_caps.has_sse) {
+      if ((fi_val.ui & 0x7f800000) == 0) {
+         fi_val.ui &= 0xff800000;
+      }
+   }
+#endif
+
+   return fi_val.f;
+}
+
 /*
  * Test one LLVM unary arithmetic builder function.
  */
 static boolean
-test_unary(struct gallivm_state *gallivm, unsigned verbose, FILE *fp, const struct unary_test_t *test)
+test_unary(unsigned verbose, FILE *fp, const struct unary_test_t *test, unsigned length)
 {
-   LLVMModuleRef module = gallivm->module;
+   char test_name[128];
+   snprintf(test_name, sizeof test_name, "%s.v%u", test->name, length);
+   LLVMContextRef context;
+   struct gallivm_state *gallivm;
    LLVMValueRef test_func;
-   LLVMExecutionEngineRef engine = gallivm->engine;
-   LLVMContextRef context = gallivm->context;
-   char *error = NULL;
    unary_func_t test_func_jit;
    boolean success = TRUE;
-   int i;
+   int i, j;
+   float *in, *out;
 
-   test_func = build_unary_test_func(gallivm, module, context, test);
+   in = align_malloc(length * 4, length * 4);
+   out = align_malloc(length * 4, length * 4);
 
-   if (LLVMVerifyModule(module, LLVMPrintMessageAction, &error)) {
-      printf("LLVMVerifyModule: %s\n", error);
-      LLVMDumpModule(module);
-      abort();
+   /* random NaNs or 0s could wreak havoc */
+   for (i = 0; i < length; i++) {
+      in[i] = 1.0;
    }
-   LLVMDisposeMessage(error);
 
-   test_func_jit = (unary_func_t) pointer_to_func(LLVMGetPointerToGlobal(engine, test_func));
+   context = LLVMContextCreate();
+   gallivm = gallivm_create("test_module", context);
 
-   for (i = 0; i < test->num_values; ++i) {
-      float value = test->values[i];
-      float ref = test->ref(value);
-      float src = test_func_jit(value);
+   test_func = build_unary_test_func(gallivm, test, length, test_name);
 
-      double error = fabs(src - ref);
-      double precision = error ? -log2(error/fabs(ref)) : FLT_MANT_DIG;
+   gallivm_compile_module(gallivm);
 
-      bool pass = precision >= 20.0;
+   test_func_jit = (unary_func_t) gallivm_jit_function(gallivm, test_func);
 
-      if (isnan(ref)) {
-         continue;
-      }
+   gallivm_free_ir(gallivm);
+
+   for (j = 0; j < (test->num_values + length - 1) / length; j++) {
+      int num_vals = ((j + 1) * length <= test->num_values) ? length :
+                                                              test->num_values % length;
 
-      if (!pass || verbose) {
-         printf("%s(%.9g): ref = %.9g, src = %.9g, precision = %f bits, %s\n",
-               test->name, value, ref, src, precision,
-               pass ? "PASS" : "FAIL");
+      for (i = 0; i < num_vals; ++i) {
+         in[i] = test->values[i+j*length];
       }
 
-      if (!pass) {
-         success = FALSE;
+      test_func_jit(out, in);
+      for (i = 0; i < num_vals; ++i) {
+         float testval, ref;
+         double error, precision;
+         boolean expected_pass = TRUE;
+         bool pass;
+
+         testval = flush_denorm_to_zero(in[i]);
+         ref = flush_denorm_to_zero(test->ref(testval));
+
+         if (util_inf_sign(ref) && util_inf_sign(out[i]) == util_inf_sign(ref)) {
+            error = 0;
+         } else {
+            error = fabs(out[i] - ref);
+         }
+         precision = error ? -log2(error/fabs(ref)) : FLT_MANT_DIG;
+
+         pass = precision >= test->precision;
+
+         if (isnan(ref)) {
+            continue;
+         }
+
+         if (!util_cpu_caps.has_neon &&
+             test->ref == &nearbyintf && length == 2 &&
+             ref != roundf(testval)) {
+            /* FIXME: The generic (non SSE) path in lp_build_iround, which is
+             * always taken for length==2 regardless of native round support,
+             * does not round to even. */
+            expected_pass = FALSE;
+         }
+
+         if (test->ref == &expf && util_inf_sign(testval) == -1) {
+            /* XXX: 64bits MSVCRT's expf(-inf) returns -inf instead of 0 */
+#if defined(_MSC_VER) && defined(_WIN64)
+            expected_pass = FALSE;
+#endif
+         }
+
+         if (pass != expected_pass || verbose) {
+            printf("%s(%.9g): ref = %.9g, out = %.9g, precision = %f bits, %s%s\n",
+                  test_name, in[i], ref, out[i], precision,
+                  pass ? "PASS" : "FAIL",
+                  !expected_pass ? (pass ? " (unexpected)" : " (expected)" ): "");
+            fflush(stdout);
+         }
+
+         if (pass != expected_pass) {
+            success = FALSE;
+         }
       }
    }
 
-   LLVMFreeMachineCodeForFunction(engine, test_func);
+   gallivm_destroy(gallivm);
+   LLVMContextDispose(context);
+
+   align_free(in);
+   align_free(out);
 
    return success;
 }
 
 
 boolean
-test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
+test_all(unsigned verbose, FILE *fp)
 {
    boolean success = TRUE;
    int i;
 
-   for (i = 0; i < Elements(unary_tests); ++i) {
-      if (!test_unary(gallivm, verbose, fp, &unary_tests[i])) {
-         success = FALSE;
+   for (i = 0; i < ARRAY_SIZE(unary_tests); ++i) {
+      unsigned max_length = lp_native_vector_width / 32;
+      unsigned length;
+      for (length = 1; length <= max_length; length *= 2) {
+         if (!test_unary(verbose, fp, &unary_tests[i], length)) {
+            success = FALSE;
+         }
       }
    }
 
@@ -276,19 +519,19 @@ test_all(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
 
 
 boolean
-test_some(struct gallivm_state *gallivm, unsigned verbose, FILE *fp,
+test_some(unsigned verbose, FILE *fp,
           unsigned long n)
 {
    /*
     * Not randomly generated test cases, so test all.
     */
 
-   return test_all(gallivm, verbose, fp);
+   return test_all(verbose, fp);
 }
 
 
 boolean
-test_single(struct gallivm_state *gallivm, unsigned verbose, FILE *fp)
+test_single(unsigned verbose, FILE *fp)
 {
    return TRUE;
 }