[Mesa-dev] [PATCH 3/3] i965/vs: Fix comparisons with uint negation.

[Eric Anholt]

The condmod instruction ends up generating garbage condition codes,
because apparently the comparison happens on the accumulator value (33
bits for UD), not the truncated value that would be written.

Fixes vs-op-neg-*
---
 src/mesa/drivers/dri/i965/brw_vec4.h               |    1 +
 .../drivers/dri/i965/brw_vec4_copy_propagation.cpp |    9 ++++++++
 src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp     |   22 ++++++++++++++++++++
 3 files changed, 32 insertions(+), 0 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_vec4.h b/src/mesa/drivers/dri/i965/brw_vec4.h
index 5f44268..a94b765 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4.h
+++ b/src/mesa/drivers/dri/i965/brw_vec4.h
@@ -510,6 +510,7 @@ public:
 				int base_offset);
 
    GLboolean try_emit_sat(ir_expression *ir);
+   void resolve_ud_negate(src_reg *reg);
 
    bool process_move_condition(ir_rvalue *ir);
 
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_copy_propagation.cpp b/src/mesa/drivers/dri/i965/brw_vec4_copy_propagation.cpp
index a5f6f93..c3a9dee 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_copy_propagation.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_copy_propagation.cpp
@@ -205,6 +205,15 @@ try_copy_propagation(struct intel_context *intel,
    if (intel->gen >= 6 && inst->is_math())
       return false;
 
+   /* We can't copy-propagate a UD negation into a condmod
+    * instruction, because the condmod ends up looking at the 33-bit
+    * signed accumulator value instead of the 32-bit value we wanted
+    */
+   if (inst->conditional_mod &&
+       value.negate &&
+       value.type == BRW_REGISTER_TYPE_UD)
+      return false;
+
    /* Don't report progress if this is a noop. */
    if (value.equals(&inst->src[arg]))
       return false;
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
index ad8b433..d364d45 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
@@ -182,6 +182,9 @@ vec4_visitor::IF(src_reg src0, src_reg src1, uint32_t condition)
 
    vec4_instruction *inst;
 
+   resolve_ud_negate(&src0);
+   resolve_ud_negate(&src1);
+
    inst = new(mem_ctx) vec4_instruction(this, BRW_OPCODE_IF, dst_null_d(),
 					src0, src1);
    inst->conditional_mod = condition;
@@ -209,6 +212,9 @@ vec4_visitor::CMP(dst_reg dst, src_reg src0, src_reg src1, uint32_t condition)
 	 dst.fixed_hw_reg.type = dst.type;
    }
 
+   resolve_ud_negate(&src0);
+   resolve_ud_negate(&src1);
+
    inst = new(mem_ctx) vec4_instruction(this, BRW_OPCODE_CMP, dst, src0, src1);
    inst->conditional_mod = condition;
 
@@ -630,6 +636,8 @@ vec4_visitor::emit_bool_to_cond_code(ir_rvalue *ir, uint32_t *predicate)
       for (unsigned int i = 0; i < expr->get_num_operands(); i++) {
 	 expr->operands[i]->accept(this);
 	 op[i] = this->result;
+
+	 resolve_ud_negate(&op[i]);
       }
 
       switch (expr->operation) {
@@ -705,6 +713,8 @@ vec4_visitor::emit_bool_to_cond_code(ir_rvalue *ir, uint32_t *predicate)
 
    ir->accept(this);
 
+   resolve_ud_negate(&this->result);
+
    if (intel->gen >= 6) {
       vec4_instruction *inst = emit(AND(dst_null_d(),
 					this->result, src_reg(1)));
@@ -2296,6 +2306,18 @@ vec4_visitor::move_uniform_array_access_to_pull_constants()
    split_uniform_registers();
 }
 
+void
+vec4_visitor::resolve_ud_negate(src_reg *reg)
+{
+   if (reg->type != BRW_REGISTER_TYPE_UD ||
+       !reg->negate)
+      return;
+
+   src_reg temp = src_reg(this, glsl_type::uvec4_type);
+   emit(BRW_OPCODE_MOV, dst_reg(temp), *reg);
+   *reg = temp;
+}
+
 vec4_visitor::vec4_visitor(struct brw_vs_compile *c,
 			   struct gl_shader_program *prog,
 			   struct brw_shader *shader)
-- 
1.7.6.3