vc4: Replace the SSA-style SEL operators with conditional MOVs.

I'm moving away from QIR being SSA (since NIR is doing lots of SSA
optimization for us now) and instead having QIR just be QPU operations
with virtual registers.  By making our SELs be composed of two MOVs, we
could potentially coalesce the registers for the MOV's src and dst and
eliminate the MOV.

total instructions in shared programs: 88448 -> 88028 (-0.47%)
instructions in affected programs:     39845 -> 39425 (-1.05%)
total estimated cycles in shared programs: 246306 -> 245762 (-0.22%)
estimated cycles in affected programs:     162887 -> 162343 (-0.33%)

1 files changed, 78 insertions, 62 deletions

diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index 44e89fe64e9..c24aa19e74e 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -275,7 +275,7 @@ qir_srgb_decode(struct vc4_compile *c, struct qreg srgb)
                                    qir_uniform_f(c, 2.4));
 
         qir_SF(c, qir_FSUB(c, srgb, qir_uniform_f(c, 0.04045)));
-        return qir_SEL_X_Y_NS(c, low, high);
+        return qir_SEL(c, QPU_COND_NS, low, high);
 }
 
 static struct qreg
@@ -475,7 +475,8 @@ ntq_emit_tex(struct vc4_compile *c, nir_tex_instr *instr)
                 struct qreg normalized = ntq_scale_depth_texture(c, tex);
                 struct qreg depth_output;
 
-                struct qreg one = qir_uniform_f(c, 1.0f);
+                struct qreg u0 = qir_uniform_f(c, 0.0f);
+                struct qreg u1 = qir_uniform_f(c, 1.0f);
                 if (c->key->tex[unit].compare_mode) {
                         if (has_proj)
                                 compare = qir_FMUL(c, compare, proj);
@@ -485,31 +486,31 @@ ntq_emit_tex(struct vc4_compile *c, nir_tex_instr *instr)
                                 depth_output = qir_uniform_f(c, 0.0f);
                                 break;
                         case PIPE_FUNC_ALWAYS:
-                                depth_output = one;
+                                depth_output = u1;
                                 break;
                         case PIPE_FUNC_EQUAL:
                                 qir_SF(c, qir_FSUB(c, compare, normalized));
-                                depth_output = qir_SEL_X_0_ZS(c, one);
+                                depth_output = qir_SEL(c, QPU_COND_ZS, u1, u0);
                                 break;
                         case PIPE_FUNC_NOTEQUAL:
                                 qir_SF(c, qir_FSUB(c, compare, normalized));
-                                depth_output = qir_SEL_X_0_ZC(c, one);
+                                depth_output = qir_SEL(c, QPU_COND_ZC, u1, u0);
                                 break;
                         case PIPE_FUNC_GREATER:
                                 qir_SF(c, qir_FSUB(c, compare, normalized));
-                                depth_output = qir_SEL_X_0_NC(c, one);
+                                depth_output = qir_SEL(c, QPU_COND_NC, u1, u0);
                                 break;
                         case PIPE_FUNC_GEQUAL:
                                 qir_SF(c, qir_FSUB(c, normalized, compare));
-                                depth_output = qir_SEL_X_0_NS(c, one);
+                                depth_output = qir_SEL(c, QPU_COND_NS, u1, u0);
                                 break;
                         case PIPE_FUNC_LESS:
                                 qir_SF(c, qir_FSUB(c, compare, normalized));
-                                depth_output = qir_SEL_X_0_NS(c, one);
+                                depth_output = qir_SEL(c, QPU_COND_NS, u1, u0);
                                 break;
                         case PIPE_FUNC_LEQUAL:
                                 qir_SF(c, qir_FSUB(c, normalized, compare));
-                                depth_output = qir_SEL_X_0_NC(c, one);
+                                depth_output = qir_SEL(c, QPU_COND_NC, u1, u0);
                                 break;
                         }
                 } else {
@@ -553,9 +554,8 @@ ntq_ffract(struct vc4_compile *c, struct qreg src)
         struct qreg trunc = qir_ITOF(c, qir_FTOI(c, src));
         struct qreg diff = qir_FSUB(c, src, trunc);
         qir_SF(c, diff);
-        return qir_SEL_X_Y_NS(c,
-                              qir_FADD(c, diff, qir_uniform_f(c, 1.0)),
-                              diff);
+        return qir_SEL(c, QPU_COND_NS,
+                       qir_FADD(c, diff, qir_uniform_f(c, 1.0)), diff);
 }
 
 /**
@@ -572,9 +572,8 @@ ntq_ffloor(struct vc4_compile *c, struct qreg src)
          */
         qir_SF(c, qir_FSUB(c, src, trunc));
 
-        return qir_SEL_X_Y_NS(c,
-                              qir_FSUB(c, trunc, qir_uniform_f(c, 1.0)),
-                              trunc);
+        return qir_SEL(c, QPU_COND_NS,
+                       qir_FSUB(c, trunc, qir_uniform_f(c, 1.0)), trunc);
 }
 
 /**
@@ -591,9 +590,8 @@ ntq_fceil(struct vc4_compile *c, struct qreg src)
          */
         qir_SF(c, qir_FSUB(c, trunc, src));
 
-        return qir_SEL_X_Y_NS(c,
-                              qir_FADD(c, trunc, qir_uniform_f(c, 1.0)),
-                              trunc);
+        return qir_SEL(c, QPU_COND_NS,
+                       qir_FADD(c, trunc, qir_uniform_f(c, 1.0)), trunc);
 }
 
 static struct qreg
@@ -668,10 +666,13 @@ ntq_fcos(struct vc4_compile *c, struct qreg src)
 static struct qreg
 ntq_fsign(struct vc4_compile *c, struct qreg src)
 {
+        struct qreg t = qir_get_temp(c);
+
         qir_SF(c, src);
-        return qir_SEL_X_Y_NC(c,
-                              qir_SEL_X_0_ZC(c, qir_uniform_f(c, 1.0)),
-                              qir_uniform_f(c, -1.0));
+        qir_MOV_dest(c, t, qir_uniform_f(c, 0.0));
+        qir_MOV_dest(c, t, qir_uniform_f(c, 1.0))->cond = QPU_COND_ZC;
+        qir_MOV_dest(c, t, qir_uniform_f(c, -1.0))->cond = QPU_COND_NS;
+        return t;
 }
 
 static void
@@ -888,6 +889,56 @@ ntq_emit_ubfe(struct vc4_compile *c, struct qreg base, struct qreg offset,
         return qir_UNPACK_8_I(c, base, offset_bit / 8);
 }
 
+static struct qreg
+ntq_emit_comparison(struct vc4_compile *c, nir_alu_instr *instr,
+                    struct qreg src0, struct qreg src1)
+{
+        enum qpu_cond cond;
+
+        switch (instr->op) {
+        case nir_op_feq:
+        case nir_op_ieq:
+        case nir_op_seq:
+                cond = QPU_COND_ZS;
+                break;
+        case nir_op_fne:
+        case nir_op_ine:
+        case nir_op_sne:
+                cond = QPU_COND_ZC;
+                break;
+        case nir_op_fge:
+        case nir_op_ige:
+        case nir_op_uge:
+        case nir_op_sge:
+                cond = QPU_COND_NC;
+                break;
+        case nir_op_flt:
+        case nir_op_ilt:
+        case nir_op_slt:
+                cond = QPU_COND_NS;
+                break;
+        default:
+                unreachable("bad ALU op for comparison");
+        }
+
+        if (nir_op_infos[instr->op].input_types[0] == nir_type_float)
+                qir_SF(c, qir_FSUB(c, src0, src1));
+        else
+                qir_SF(c, qir_SUB(c, src0, src1));
+
+        switch (instr->op) {
+        case nir_op_seq:
+        case nir_op_sne:
+        case nir_op_sge:
+        case nir_op_slt:
+                return qir_SEL(c, cond,
+                               qir_uniform_f(c, 1.0), qir_uniform_f(c, 0.0));
+        default:
+                return qir_SEL(c, cond,
+                               qir_uniform_ui(c, ~0), qir_uniform_ui(c, 0.0));
+        }
+}
+
 static void
 ntq_emit_alu(struct vc4_compile *c, nir_alu_instr *instr)
 {
@@ -974,7 +1025,9 @@ ntq_emit_alu(struct vc4_compile *c, nir_alu_instr *instr)
         case nir_op_i2b:
         case nir_op_f2b:
                 qir_SF(c, src[0]);
-                *dest = qir_SEL_X_0_ZC(c, qir_uniform_ui(c, ~0));
+                *dest = qir_SEL(c, QPU_COND_ZC,
+                                qir_uniform_ui(c, ~0),
+                                qir_uniform_ui(c, 0));
                 break;
 
         case nir_op_iadd:
@@ -1016,65 +1069,28 @@ ntq_emit_alu(struct vc4_compile *c, nir_alu_instr *instr)
                 break;
 
         case nir_op_seq:
-                qir_SF(c, qir_FSUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_ZS(c, qir_uniform_f(c, 1.0));
-                break;
         case nir_op_sne:
-                qir_SF(c, qir_FSUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_ZC(c, qir_uniform_f(c, 1.0));
-                break;
         case nir_op_sge:
-                qir_SF(c, qir_FSUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_NC(c, qir_uniform_f(c, 1.0));
-                break;
         case nir_op_slt:
-                qir_SF(c, qir_FSUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_NS(c, qir_uniform_f(c, 1.0));
-                break;
         case nir_op_feq:
-                qir_SF(c, qir_FSUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_ZS(c, qir_uniform_ui(c, ~0));
-                break;
         case nir_op_fne:
-                qir_SF(c, qir_FSUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_ZC(c, qir_uniform_ui(c, ~0));
-                break;
         case nir_op_fge:
-                qir_SF(c, qir_FSUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_NC(c, qir_uniform_ui(c, ~0));
-                break;
         case nir_op_flt:
-                qir_SF(c, qir_FSUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_NS(c, qir_uniform_ui(c, ~0));
-                break;
         case nir_op_ieq:
-                qir_SF(c, qir_SUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_ZS(c, qir_uniform_ui(c, ~0));
-                break;
         case nir_op_ine:
-                qir_SF(c, qir_SUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_ZC(c, qir_uniform_ui(c, ~0));
-                break;
         case nir_op_ige:
-                qir_SF(c, qir_SUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_NC(c, qir_uniform_ui(c, ~0));
-                break;
         case nir_op_uge:
-                qir_SF(c, qir_SUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_CC(c, qir_uniform_ui(c, ~0));
-                break;
         case nir_op_ilt:
-                qir_SF(c, qir_SUB(c, src[0], src[1]));
-                *dest = qir_SEL_X_0_NS(c, qir_uniform_ui(c, ~0));
+                *dest = ntq_emit_comparison(c, instr, src[0], src[1]);
                 break;
 
         case nir_op_bcsel:
                 qir_SF(c, src[0]);
-                *dest = qir_SEL_X_Y_NS(c, src[1], src[2]);
+                *dest = qir_SEL(c, QPU_COND_NS, src[1], src[2]);
                 break;
         case nir_op_fcsel:
                 qir_SF(c, src[0]);
-                *dest = qir_SEL_X_Y_ZC(c, src[1], src[2]);
+                *dest = qir_SEL(c, QPU_COND_ZC, src[1], src[2]);
                 break;
 
         case nir_op_frcp: