nir: nir_const_value_negative_equal compares one value at a time

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>

src/compiler/nir/nir.h

@@ -1029,9 +1029,7 @@ nir_ssa_alu_instr_src_components(const nir_alu_instr *instr, unsigned src)
    return instr->dest.dest.ssa.num_components;
 }
 
-bool nir_const_value_negative_equal(const nir_const_value *c1,
+bool nir_const_value_negative_equal(nir_const_value c1, nir_const_value c2,
-                                    const nir_const_value *c2,
-                                    unsigned components,
                                     nir_alu_type full_type);
 
 bool nir_alu_srcs_equal(const nir_alu_instr *alu1, const nir_alu_instr *alu2,

src/compiler/nir/nir_instr_set.c

@@ -302,9 +302,8 @@ get_neg_instr(nir_src s)
 }
 
 bool
-nir_const_value_negative_equal(const nir_const_value *c1,
+nir_const_value_negative_equal(nir_const_value c1,
-                               const nir_const_value *c2,
+                               nir_const_value c2,
-                               unsigned components,
                                nir_alu_type full_type)
 {
    assert(nir_alu_type_get_base_type(full_type) != nir_type_invalid);
@@ -312,66 +311,29 @@ nir_const_value_negative_equal(const nir_const_value *c1,
 
    switch (full_type) {
    case nir_type_float16:
-      for (unsigned i = 0; i < components; i++) {
+      return _mesa_half_to_float(c1.u16) == -_mesa_half_to_float(c2.u16);
-         if (_mesa_half_to_float(c1[i].u16) !=
-             -_mesa_half_to_float(c2[i].u16)) {
-            return false;
-         }
-      }
-
-      return true;
 
    case nir_type_float32:
-      for (unsigned i = 0; i < components; i++) {
+      return c1.f32 == -c2.f32;
-         if (c1[i].f32 != -c2[i].f32)
-            return false;
-      }
-
-      return true;
 
    case nir_type_float64:
-      for (unsigned i = 0; i < components; i++) {
+      return c1.f64 == -c2.f64;
-         if (c1[i].f64 != -c2[i].f64)
-            return false;
-      }
-
-      return true;
 
    case nir_type_int8:
    case nir_type_uint8:
-      for (unsigned i = 0; i < components; i++) {
+      return c1.i8 == -c2.i8;
-         if (c1[i].i8 != -c2[i].i8)
-            return false;
-      }
-
-      return true;
 
    case nir_type_int16:
    case nir_type_uint16:
-      for (unsigned i = 0; i < components; i++) {
+      return c1.i16 == -c2.i16;
-         if (c1[i].i16 != -c2[i].i16)
-            return false;
-      }
-
-      return true;
 
    case nir_type_int32:
    case nir_type_uint32:
-      for (unsigned i = 0; i < components; i++) {
+      return c1.i32 == -c2.i32;
-         if (c1[i].i32 != -c2[i].i32)
-            return false;
-      }
-
-      return true;
 
    case nir_type_int64:
    case nir_type_uint64:
-      for (unsigned i = 0; i < components; i++) {
+      return c1.i64 == -c2.i64;
-         if (c1[i].i64 != -c2[i].i64)
-            return false;
-      }
-
-      return true;
 
    default:
       break;
@@ -424,11 +386,15 @@ nir_alu_srcs_negative_equal(const nir_alu_instr *alu1,
          return false;
 
       /* FINISHME: Apply the swizzle? */
-      return nir_const_value_negative_equal(const1,
+      const unsigned components = nir_ssa_alu_instr_src_components(alu1, src1);
-                                            const2,
+      const nir_alu_type full_type = nir_op_infos[alu1->op].input_types[src1] |
-                                            nir_ssa_alu_instr_src_components(alu1, src1),
+                                     nir_src_bit_size(alu1->src[src1].src);
-                                            nir_op_infos[alu1->op].input_types[src1] |
+      for (unsigned i = 0; i < components; i++) {
-                                            nir_src_bit_size(alu1->src[src1].src));
+         if (!nir_const_value_negative_equal(const1[i], const2[i], full_type))
+            return false;
+      }
+
+      return true;
    }
 
    uint8_t alu1_swizzle[4] = {0};

src/compiler/nir/tests/negative_equal_tests.cpp

@@ -71,15 +71,13 @@ protected:
 TEST_F(const_value_negative_equal_test, float32_zero)
 {
    /* Verify that 0.0 negative-equals 0.0. */
-   EXPECT_TRUE(nir_const_value_negative_equal(c1, c1, NIR_MAX_VEC_COMPONENTS,
+   EXPECT_TRUE(nir_const_value_negative_equal(c1[0], c1[0], nir_type_float32));
-                                              nir_type_float32));
 }
 
 TEST_F(const_value_negative_equal_test, float64_zero)
 {
    /* Verify that 0.0 negative-equals 0.0. */
-   EXPECT_TRUE(nir_const_value_negative_equal(c1, c1, NIR_MAX_VEC_COMPONENTS,
+   EXPECT_TRUE(nir_const_value_negative_equal(c1[0], c1[0], nir_type_float64));
-                                              nir_type_float64));
 }
 
 /* Compare an object with non-zero values to itself.  This should always be
@@ -89,9 +87,7 @@ TEST_F(const_value_negative_equal_test, float64_zero)
 TEST_F(const_value_negative_equal_test, full_type ## _self)             \
 {                                                                       \
    count_sequence(c1, full_type, 1);                                    \
-   EXPECT_FALSE(nir_const_value_negative_equal(c1, c1,                  \
+   EXPECT_FALSE(nir_const_value_negative_equal(c1[0], c1[0], full_type)); \
-                                               NIR_MAX_VEC_COMPONENTS,  \
-                                               full_type));             \
 }
 
 compare_with_self(nir_type_float16)
@@ -113,10 +109,8 @@ compare_with_self(nir_type_uint64)
 TEST_F(const_value_negative_equal_test, full_type ## _trivially_true)   \
 {                                                                       \
    count_sequence(c1, full_type, 1);                                    \
-   negate(c2, c1, full_type, NIR_MAX_VEC_COMPONENTS);                   \
+   negate(c2, c1, full_type, 1);                                        \
-   EXPECT_TRUE(nir_const_value_negative_equal(c1, c2,                   \
+   EXPECT_TRUE(nir_const_value_negative_equal(c1[0], c2[0], full_type)); \
-                                              NIR_MAX_VEC_COMPONENTS,   \
-                                              full_type));              \
 }
 
 compare_with_negation(nir_type_float16)
@@ -132,32 +126,6 @@ compare_with_negation(nir_type_int64)
 compare_with_negation(nir_type_uint64)
 #undef compare_with_negation
 
-/* Compare fewer than the maximum possible components.  All of the components
- * that are compared a negative-equal, but the extra components are not.
- */
-#define compare_fewer_components(full_type)                             \
-TEST_F(const_value_negative_equal_test, full_type ## _fewer_components) \
-{                                                                       \
-   count_sequence(c1, full_type, 1);                                    \
-   negate(c2, c1, full_type, 3);                                        \
-   EXPECT_TRUE(nir_const_value_negative_equal(c1, c2, 3, full_type));   \
-   EXPECT_FALSE(nir_const_value_negative_equal(c1, c2,                  \
-                                               NIR_MAX_VEC_COMPONENTS,  \
-                                               full_type));             \
-}
-
-compare_fewer_components(nir_type_float16)
-compare_fewer_components(nir_type_float32)
-compare_fewer_components(nir_type_float64)
-compare_fewer_components(nir_type_int8)
-compare_fewer_components(nir_type_uint8)
-compare_fewer_components(nir_type_int16)
-compare_fewer_components(nir_type_uint16)
-compare_fewer_components(nir_type_int32)
-compare_fewer_components(nir_type_uint32)
-compare_fewer_components(nir_type_int64)
-compare_fewer_components(nir_type_uint64)
-
 TEST_F(alu_srcs_negative_equal_test, trivial_float)
 {
    nir_ssa_def *two = nir_imm_float(&bld, 2.0f);