intel: Don't propagate conditional modifiers if a UD source is negated

This fixes a bug uncovered by my NIR integer division by constant
optimization series.

Fixes: 19f9cb72c8b "i965/fs: Add pass to propagate conditional..."
Fixes: 627f94b72e0 "i965/vec4: adding vec4_cmod_propagation..."
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp
index 3f7f2b4c984c593e050c7f366bd3c6ef2cf67215..23a25fedca57c20ab84dec3ca27cf10557259db6 100644 (file)
--- a/src/intel/compiler/brw_fs.cpp
+++ b/src/intel/compiler/brw_fs.cpp
@@ -395,6 +395,25 @@ fs_inst::can_do_source_mods(const struct gen_device_info *devinfo)
    return true;
 }
 
+bool
+fs_inst::can_do_cmod()
+{
+   if (!backend_instruction::can_do_cmod())
+      return false;
+
+   /* The accumulator result appears to get used for the conditional modifier
+    * generation.  When negating a UD value, there is a 33rd bit generated for
+    * the sign in the accumulator value, so now you can't check, for example,
+    * equality with a 32-bit value.  See piglit fs-op-neg-uvec4.
+    */
+   for (unsigned i = 0; i < sources; i++) {
+      if (type_is_unsigned_int(src[i].type) && src[i].negate)
+         return false;
+   }
+
+   return true;
+}
+
 bool
 fs_inst::can_change_types() const
 {

diff --git a/src/intel/compiler/brw_ir_fs.h b/src/intel/compiler/brw_ir_fs.h
index 92dad269a349d223d302402cfeb46ba6bc2ff579..07e7224e0f83a6a9e74b8167f645d9028d6f061f 100644 (file)
--- a/src/intel/compiler/brw_ir_fs.h
+++ b/src/intel/compiler/brw_ir_fs.h
@@ -354,6 +354,7 @@ public:
    unsigned components_read(unsigned i) const;
    unsigned size_read(int arg) const;
    bool can_do_source_mods(const struct gen_device_info *devinfo);
+   bool can_do_cmod();
    bool can_change_types() const;
    bool has_source_and_destination_hazard() const;
 

diff --git a/src/intel/compiler/brw_ir_vec4.h b/src/intel/compiler/brw_ir_vec4.h
index e401d8b4d1613d97d915b782f2f67b42d45f382a..65b1e4f3b53506ec39762783065b288efee42c45 100644 (file)
--- a/src/intel/compiler/brw_ir_vec4.h
+++ b/src/intel/compiler/brw_ir_vec4.h
@@ -291,6 +291,7 @@ public:
                       int swizzle, int swizzle_mask);
    void reswizzle(int dst_writemask, int swizzle);
    bool can_do_source_mods(const struct gen_device_info *devinfo);
+   bool can_do_cmod();
    bool can_do_writemask(const struct gen_device_info *devinfo);
    bool can_change_types() const;
    bool has_source_and_destination_hazard() const;

diff --git a/src/intel/compiler/brw_reg.h b/src/intel/compiler/brw_reg.h
index ac12ab3d2dd7f3e6ebbbc96f51a5d44cdb8d55e8..46d66198a1da1edd5f7aa2c6cd68b32ae67cc5ce 100644 (file)
--- a/src/intel/compiler/brw_reg.h
+++ b/src/intel/compiler/brw_reg.h
@@ -376,6 +376,15 @@ brw_int_type(unsigned sz, bool is_signed)
    }
 }
 
+static inline bool
+type_is_unsigned_int(enum brw_reg_type tp)
+{
+   return tp == BRW_REGISTER_TYPE_UB ||
+          tp == BRW_REGISTER_TYPE_UW ||
+          tp == BRW_REGISTER_TYPE_UD ||
+          tp == BRW_REGISTER_TYPE_UQ;
+}
+
 /**
  * Construct a brw_reg.
  * \param file      one of the BRW_x_REGISTER_FILE values

diff --git a/src/intel/compiler/brw_vec4.cpp b/src/intel/compiler/brw_vec4.cpp
index 5a86f30634adee5504910505393cc8b7cac65ed2..74a4d09fc79ea4358671e6c2f8a4afd7b99ad936 100644 (file)
--- a/src/intel/compiler/brw_vec4.cpp
+++ b/src/intel/compiler/brw_vec4.cpp
@@ -257,6 +257,26 @@ vec4_instruction::can_do_source_mods(const struct gen_device_info *devinfo)
    return true;
 }
 
+bool
+vec4_instruction::can_do_cmod()
+{
+   if (!backend_instruction::can_do_cmod())
+      return false;
+
+   /* The accumulator result appears to get used for the conditional modifier
+    * generation.  When negating a UD value, there is a 33rd bit generated for
+    * the sign in the accumulator value, so now you can't check, for example,
+    * equality with a 32-bit value.  See piglit fs-op-neg-uvec4.
+    */
+   for (unsigned i = 0; i < 3; i++) {
+      if (src[i].file != BAD_FILE &&
+          type_is_unsigned_int(src[i].type) && src[i].negate)
+         return false;
+   }
+
+   return true;
+}
+
 bool
 vec4_instruction::can_do_writemask(const struct gen_device_info *devinfo)
 {