brw/cmod: Enable limited cmod propagation for BFN

cmod propagation needs more work. Since the result type is always UD,
BRW_CONDITION_G should be able to substitute for NZ. Either that or
users of the condition could be rewritten to use an inverted condition.

v2: Add a couple more unit tests. Suggested by Matt.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37186>

src/intel/compiler/brw/brw_from_nir.cpp

@@ -4231,7 +4231,7 @@ brw_from_nir_emit_fs_intrinsic(nir_to_brw_state &ntb,
 
             cmp = (brw_inst *) s.instructions.get_tail();
             if (cmp->conditional_mod == BRW_CONDITIONAL_NONE) {
-               if (cmp->can_do_cmod())
+               if (cmp->can_do_cmod(BRW_CONDITIONAL_Z))
                   cmp->conditional_mod = BRW_CONDITIONAL_Z;
                else
                   cmp = NULL;

src/intel/compiler/brw/brw_inst.cpp

@@ -370,7 +370,7 @@ brw_inst::can_do_source_mods(const struct intel_device_info *devinfo) const
 }
 
 bool
-brw_inst::can_do_cmod() const
+brw_inst::can_do_cmod(enum brw_conditional_mod cmod) const
 {
    switch (opcode) {
    case BRW_OPCODE_ADD:
@@ -406,6 +406,15 @@ brw_inst::can_do_cmod() const
    case BRW_OPCODE_SUBB:
    case BRW_OPCODE_XOR:
       break;
+
+   case BRW_OPCODE_BFN:
+      if (cmod != BRW_CONDITIONAL_Z &&
+          cmod != BRW_CONDITIONAL_G &&
+          cmod != BRW_CONDITIONAL_L)
+         return false;
+
+      break;
+
    default:
       return false;
    }

src/intel/compiler/brw/brw_inst.h

@@ -90,7 +90,7 @@ struct brw_inst : brw_exec_node {
    unsigned components_read(unsigned i) const;
    unsigned size_read(const struct intel_device_info *devinfo, int arg) const;
    bool can_do_source_mods(const struct intel_device_info *devinfo) const;
-   bool can_do_cmod() const;
+   bool can_do_cmod(enum brw_conditional_mod cmod) const;
    bool can_change_types() const;
    bool has_source_and_destination_hazard() const;
 

src/intel/compiler/brw/brw_opt_cmod_propagation.cpp

@@ -210,7 +210,7 @@ cmod_propagate_cmp_to_add(const intel_device_info *devinfo, brw_inst *inst)
          }
 
          /* Otherwise, try propagating the conditional. */
-         if (scan_inst->can_do_cmod() &&
+         if (scan_inst->can_do_cmod(cond) &&
              ((!read_flag && scan_inst->conditional_mod == BRW_CONDITIONAL_NONE) ||
               scan_inst->conditional_mod == cond)) {
             scan_inst->conditional_mod = cond;
@@ -464,7 +464,8 @@ opt_cmod_propagation_local(const intel_device_info *devinfo, bblock_t *block)
                   inst->remove();
                   progress = true;
                   break;
-               } else if (!read_flag && scan_inst->can_do_cmod()) {
+               } else if (!read_flag &&
+                          scan_inst->can_do_cmod(inst->conditional_mod)) {
                   scan_inst->conditional_mod = inst->conditional_mod;
                   scan_inst->flag_subreg = inst->flag_subreg;
                   inst->remove();
@@ -506,6 +507,16 @@ opt_cmod_propagation_local(const intel_device_info *devinfo, bblock_t *block)
                inst->src[0].negate ? brw_swap_cmod(inst->conditional_mod)
                                    : inst->conditional_mod;
 
+            /* BFN supports a limited subset of possible conditions. It
+             * notably lacks NZ. However, if the destination is unsigned, G
+             * and NZ are equivalent.
+             */
+            if (scan_inst->opcode == BRW_OPCODE_BFN &&
+                brw_type_is_uint(scan_inst->dst.type) &&
+                cond == BRW_CONDITIONAL_NZ) {
+               cond = BRW_CONDITIONAL_G;
+            }
+
             /* From the Kaby Lake PRM Vol. 7 "Assigning Conditional Flags":
              *
              *    * Note that the [post condition signal] bits generated at
@@ -524,7 +535,7 @@ opt_cmod_propagation_local(const intel_device_info *devinfo, bblock_t *block)
              */
 
             /* Otherwise, try propagating the conditional. */
-            if (scan_inst->can_do_cmod() &&
+            if (scan_inst->can_do_cmod(cond) &&
                 ((!read_flag && scan_inst->conditional_mod == BRW_CONDITIONAL_NONE) ||
                  scan_inst->conditional_mod == cond)) {
                scan_inst->conditional_mod = cond;

src/intel/compiler/brw/test_opt_cmod_propagation.cpp

@@ -2467,3 +2467,151 @@ TEST_F(cmod_propagation_test, Boolean_size_conversion)
 
    EXPECT_SHADERS_MATCH(bld, exp);
 }
+
+TEST_F(cmod_propagation_test, BFN_CMP_Z)
+{
+   // BFN doesn't exist on earlier platforms.
+   set_gfx_verx10(125);
+
+   brw_builder bld = make_shader();
+   brw_builder exp = make_shader();
+
+   brw_reg dest1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src0 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src2 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg zero  = brw_imm_ud(0);
+
+   bld.BFN(dest1, src0, src1, src2, 0xfe);
+   bld.CMP(bld.null_reg_d(), dest1, zero, BRW_CONDITIONAL_Z);
+
+   EXPECT_PROGRESS(brw_opt_cmod_propagation, bld);
+
+   exp.BFN(dest1, src0, src1, src2, 0xfe)
+      ->conditional_mod = BRW_CONDITIONAL_Z;
+
+   EXPECT_SHADERS_MATCH(bld, exp);
+}
+
+TEST_F(cmod_propagation_test, BFN_Z_CMP_Z)
+{
+   // BFN doesn't exist on earlier platforms.
+   set_gfx_verx10(125);
+
+   brw_builder bld = make_shader();
+   brw_builder exp = make_shader();
+
+   brw_reg dest1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src0 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src2 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg zero  = brw_imm_ud(0);
+
+   bld.BFN(dest1, src0, src1, src2, 0xfe)
+      ->conditional_mod = BRW_CONDITIONAL_Z;
+   bld.CMP(bld.null_reg_d(), dest1, zero, BRW_CONDITIONAL_Z);
+
+   EXPECT_PROGRESS(brw_opt_cmod_propagation, bld);
+
+   exp.BFN(dest1, src0, src1, src2, 0xfe)
+      ->conditional_mod = BRW_CONDITIONAL_Z;
+
+   EXPECT_SHADERS_MATCH(bld, exp);
+}
+
+TEST_F(cmod_propagation_test, BFN_CMP_NZ)
+{
+   // BFN doesn't exist on earlier platforms.
+   set_gfx_verx10(125);
+
+   brw_builder bld = make_shader();
+   brw_builder exp = make_shader();
+
+   brw_reg dest1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src0 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src2 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg zero  = brw_imm_ud(0);
+
+   bld.BFN(dest1, src0, src1, src2, 0xfe);
+   bld.CMP(bld.null_reg_d(), dest1, zero, BRW_CONDITIONAL_NZ);
+
+   EXPECT_PROGRESS(brw_opt_cmod_propagation, bld);
+
+   exp.BFN(dest1, src0, src1, src2, 0xfe)
+      ->conditional_mod = BRW_CONDITIONAL_G;
+
+   EXPECT_SHADERS_MATCH(bld, exp);
+}
+
+TEST_F(cmod_propagation_test, BFN_CMP_D_NZ)
+{
+   // BFN doesn't exist on earlier platforms.
+   set_gfx_verx10(125);
+
+   brw_builder bld = make_shader();
+   brw_builder exp = make_shader();
+
+   brw_reg dest1 = vgrf(bld, exp, BRW_TYPE_D);
+   brw_reg src0 = vgrf(bld, exp, BRW_TYPE_D);
+   brw_reg src1 = vgrf(bld, exp, BRW_TYPE_D);
+   brw_reg src2 = vgrf(bld, exp, BRW_TYPE_D);
+   brw_reg zero  = brw_imm_ud(0);
+
+   bld.BFN(dest1, src0, src1, src2, 0xfe);
+   bld.CMP(bld.null_reg_d(), dest1, zero, BRW_CONDITIONAL_NZ);
+
+   EXPECT_NO_PROGRESS(brw_opt_cmod_propagation, bld);
+}
+
+TEST_F(cmod_propagation_test, BFN_G_CMP_NZ)
+{
+   // BFN doesn't exist on earlier platforms.
+   set_gfx_verx10(125);
+
+   brw_builder bld = make_shader();
+   brw_builder exp = make_shader();
+
+   brw_reg dest1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src0 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src2 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg zero  = brw_imm_ud(0);
+
+   bld.BFN(dest1, src0, src1, src2, 0xfe)
+      ->conditional_mod = BRW_CONDITIONAL_G;
+   bld.CMP(bld.null_reg_d(), dest1, zero, BRW_CONDITIONAL_NZ);
+
+   EXPECT_PROGRESS(brw_opt_cmod_propagation, bld);
+
+   exp.BFN(dest1, src0, src1, src2, 0xfe)
+      ->conditional_mod = BRW_CONDITIONAL_G;
+
+   EXPECT_SHADERS_MATCH(bld, exp);
+}
+
+TEST_F(cmod_propagation_test, BFN_G_CMP_D_G)
+{
+   // BFN doesn't exist on earlier platforms.
+   set_gfx_verx10(125);
+
+   brw_builder bld = make_shader();
+   brw_builder exp = make_shader();
+
+   brw_reg dest1 = vgrf(bld, exp, BRW_TYPE_D);
+   brw_reg src0 = vgrf(bld, exp, BRW_TYPE_D);
+   brw_reg src1 = vgrf(bld, exp, BRW_TYPE_UD);
+   brw_reg src2 = vgrf(bld, exp, BRW_TYPE_D);
+   brw_reg zero  = brw_imm_ud(0);
+
+   bld.BFN(dest1, src0, src1, src2, 0xfe)
+      ->conditional_mod = BRW_CONDITIONAL_G;
+   bld.CMP(bld.null_reg_d(), dest1, zero, BRW_CONDITIONAL_G);
+
+   EXPECT_PROGRESS(brw_opt_cmod_propagation, bld);
+
+   exp.BFN(dest1, src0, src1, src2, 0xfe)
+      ->conditional_mod = BRW_CONDITIONAL_G;
+
+   EXPECT_SHADERS_MATCH(bld, exp);
+}