nir,panfrost: Suffix fsat_signed and fclamp_pos with _mali

Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11463>

src/compiler/nir/nir_opcodes.py

@@ -217,8 +217,6 @@ unop("isign", tint, "(src0 == 0) ? 0 : ((src0 > 0) ? 1 : -1)")
 unop("iabs", tint, "(src0 < 0) ? -src0 : src0")
 unop("fabs", tfloat, "fabs(src0)")
 unop("fsat", tfloat, ("fmin(fmax(src0, 0.0), 1.0)"))
-unop("fsat_signed", tfloat, ("fmin(fmax(src0, -1.0), 1.0)"))
-unop("fclamp_pos", tfloat, ("fmax(src0, 0.0)"))
 unop("frcp", tfloat, "bit_size == 64 ? 1.0 / src0 : 1.0f / src0")
 unop("frsq", tfloat, "bit_size == 64 ? 1.0 / sqrt(src0) : 1.0f / sqrtf(src0)")
 unop("fsqrt", tfloat, "bit_size == 64 ? sqrt(src0) : sqrtf(src0)")
@@ -1288,6 +1286,10 @@ binop("umul24_relaxed", tuint32, _2src_commutative + associative, "src0 * src1")
 unop_convert("fisnormal", tbool1, tfloat, "isnormal(src0)")
 unop_convert("fisfinite", tbool1, tfloat, "isfinite(src0)")
 
+# Mali-specific opcodes
+unop("fsat_signed_mali", tfloat, ("fmin(fmax(src0, -1.0), 1.0)"))
+unop("fclamp_pos_mali", tfloat, ("fmax(src0, 0.0)"))
+
 # DXIL specific double [un]pack
 # DXIL doesn't support generic [un]pack instructions, so we want those
 # lowered to bit ops. HLSL doesn't support 64bit bitcasts to/from

src/panfrost/bifrost/bifrost_compile.c

@@ -1785,13 +1785,13 @@ bi_emit_alu(bi_builder *b, nir_alu_instr *instr)
                 break;
         }
 
-        case nir_op_fsat_signed: {
+        case nir_op_fsat_signed_mali: {
                 bi_instr *I = bi_fadd_to(b, sz, dst, s0, bi_negzero(), BI_ROUND_NONE);
                 I->clamp = BI_CLAMP_CLAMP_M1_1;
                 break;
         }
 
-        case nir_op_fclamp_pos: {
+        case nir_op_fclamp_pos_mali: {
                 bi_instr *I = bi_fadd_to(b, sz, dst, s0, bi_negzero(), BI_ROUND_NONE);
                 I->clamp = BI_CLAMP_CLAMP_0_INF;
                 break;

src/panfrost/bifrost/bifrost_nir_algebraic.py

@@ -33,9 +33,9 @@ algebraic_late = [
     (('fmul', a, 2.0), ('fadd', a, a)),
 
     # Fuse Mali-specific clamps
-    (('fmin', ('fmax', a, -1.0), 1.0), ('fsat_signed', a)),
-    (('fmax', ('fmin', a, 1.0), -1.0), ('fsat_signed', a)),
-    (('fmax', a, 0.0), ('fclamp_pos', a)),
+    (('fmin', ('fmax', a, -1.0), 1.0), ('fsat_signed_mali', a)),
+    (('fmax', ('fmin', a, 1.0), -1.0), ('fsat_signed_mali', a)),
+    (('fmax', a, 0.0), ('fclamp_pos_mali', a)),
 ]
 
 

src/panfrost/midgard/midgard_compile.c

@@ -585,9 +585,9 @@ mir_accept_dest_mod(compiler_context *ctx, nir_dest **dest, nir_op op)
 static unsigned
 mir_determine_float_outmod(compiler_context *ctx, nir_dest **dest, unsigned prior_outmod)
 {
-        bool clamp_0_inf = mir_accept_dest_mod(ctx, dest, nir_op_fclamp_pos);
+        bool clamp_0_inf = mir_accept_dest_mod(ctx, dest, nir_op_fclamp_pos_mali);
         bool clamp_0_1 = mir_accept_dest_mod(ctx, dest, nir_op_fsat);
-        bool clamp_m1_1 = mir_accept_dest_mod(ctx, dest, nir_op_fsat_signed);
+        bool clamp_m1_1 = mir_accept_dest_mod(ctx, dest, nir_op_fsat_signed_mali);
         bool prior = (prior_outmod != midgard_outmod_none);
         int count = (int) prior + (int) clamp_0_inf + (int) clamp_0_1 + (int) clamp_m1_1;
 
@@ -659,7 +659,7 @@ mir_is_bcsel_float(nir_alu_instr *instr)
         };
 
         nir_op floatdestmods[] = {
-                nir_op_fsat, nir_op_fsat_signed, nir_op_fclamp_pos,
+                nir_op_fsat, nir_op_fsat_signed_mali, nir_op_fclamp_pos_mali,
                 nir_op_f2f16, nir_op_f2f32
         };
 
@@ -847,8 +847,8 @@ emit_alu(compiler_context *ctx, nir_alu_instr *instr)
                 ALU_CASE(fabs, fmov);
                 ALU_CASE(fneg, fmov);
                 ALU_CASE(fsat, fmov);
-                ALU_CASE(fsat_signed, fmov);
-                ALU_CASE(fclamp_pos, fmov);
+                ALU_CASE(fsat_signed_mali, fmov);
+                ALU_CASE(fclamp_pos_mali, fmov);
 
         /* For size conversion, we use a move. Ideally though we would squash
          * these ops together; maybe that has to happen after in NIR as part of
@@ -939,9 +939,9 @@ emit_alu(compiler_context *ctx, nir_alu_instr *instr)
                 outmod = midgard_outmod_keeplo;
         } else if (instr->op == nir_op_fsat) {
                 outmod = midgard_outmod_clamp_0_1;
-        } else if (instr->op == nir_op_fsat_signed) {
+        } else if (instr->op == nir_op_fsat_signed_mali) {
                 outmod = midgard_outmod_clamp_m1_1;
-        } else if (instr->op == nir_op_fclamp_pos) {
+        } else if (instr->op == nir_op_fclamp_pos_mali) {
                 outmod = midgard_outmod_clamp_0_inf;
         }
 

src/panfrost/midgard/midgard_nir_algebraic.py

@@ -53,11 +53,11 @@ algebraic_late = [
     (('b32csel', a, 0, 'b@32'), ('iand', ('inot', a), b)),
 
     # Fuse sat_signed. This should probably be shared with Bifrost
-    (('~fmin', ('fmax', a, -1.0), 1.0), ('fsat_signed', a)),
-    (('~fmax', ('fmin', a, 1.0), -1.0), ('fsat_signed', a)),
+    (('~fmin', ('fmax', a, -1.0), 1.0), ('fsat_signed_mali', a)),
+    (('~fmax', ('fmin', a, 1.0), -1.0), ('fsat_signed_mali', a)),
 
     # Fuse clamp_positive. This should probably be shared with Utgard/bifrost
-    (('fmax', a, 0.0), ('fclamp_pos', a)),
+    (('fmax', a, 0.0), ('fclamp_pos_mali', a)),
 
     (('ishl', 'a@16', b), ('u2u16', ('ishl', ('u2u32', a), b))),
     (('ishr', 'a@16', b), ('i2i16', ('ishr', ('i2i32', a), b))),