nir: make fclamp_pos_mali and fsat_signed_mali opcodes generic

V3D can use these too.

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31480>

src/asahi/lib/agx_nir_lower_tilebuffer.c

@@ -96,7 +96,7 @@ store_tilebuffer(nir_builder *b, struct agx_tilebuffer_layout *tib,
 }
 
 static nir_def *
-nir_fsat_signed(nir_builder *b, nir_def *x)
+nir_build_fsat_signed(nir_builder *b, nir_def *x)
 {
    return nir_fclamp(b, x, nir_imm_floatN_t(b, -1.0, x->bit_size),
                      nir_imm_floatN_t(b, +1.0, x->bit_size));
@@ -108,7 +108,7 @@ nir_fsat_to_format(nir_builder *b, nir_def *x, enum pipe_format format)
    if (util_format_is_unorm(format))
       return nir_fsat(b, x);
    else if (util_format_is_snorm(format))
-      return nir_fsat_signed(b, x);
+      return nir_build_fsat_signed(b, x);
    else
       return x;
 }

src/compiler/nir/nir_lower_blend.c

@@ -133,7 +133,7 @@ nir_blend_factor_value(
 }
 
 static nir_def *
-nir_fsat_signed(nir_builder *b, nir_def *x)
+nir_build_fsat_signed(nir_builder *b, nir_def *x)
 {
    return nir_fclamp(b, x, nir_imm_floatN_t(b, -1.0, x->bit_size),
                      nir_imm_floatN_t(b, +1.0, x->bit_size));
@@ -145,7 +145,7 @@ nir_fsat_to_format(nir_builder *b, nir_def *x, enum pipe_format format)
    if (util_format_is_unorm(format))
       return nir_fsat(b, x);
    else if (util_format_is_snorm(format))
-      return nir_fsat_signed(b, x);
+      return nir_build_fsat_signed(b, x);
    else
       return x;
 }

src/compiler/nir/nir_opcodes.py

@@ -1504,9 +1504,9 @@ unop("pack_2x16_to_unorm_2x10_v3d", tuint32, "pack_2x16_to_unorm_2x10(src0)")
 # and one 10 bit unorm
 unop("pack_2x16_to_unorm_10_2_v3d", tuint32, "pack_2x16_to_unorm_10_2(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)"))
+# These opcodes are used used by Mali and V3D
+unop("fsat_signed", tfloat, ("fmin(fmax(src0, -1.0), 1.0)"))
+unop("fclamp_pos", tfloat, ("fmax(src0, 0.0)"))
 
 opcode("b32fcsel_mdg", 0, tuint, [0, 0, 0],
        [tbool32, tfloat, tfloat], False, selection, "src0 ? src1 : src2",

src/gallium/drivers/lima/ir/lima_nir_algebraic.py

@@ -39,7 +39,7 @@ lower_ftrunc = [
 
 # PP fuse clamp_positive. Shared with Midgard/Bifrost
 ppir_algebraic_late = [
-    (('fmax', 'a', 0.0), ('fclamp_pos_mali', 'a')),
+    (('fmax', 'a', 0.0), ('fclamp_pos', 'a')),
 ]
 
 def main():

src/gallium/drivers/lima/ir/pp/nir.c

@@ -151,7 +151,7 @@ static int nir_to_ppir_opcodes[nir_num_opcodes] = {
    [nir_op_inot] = ppir_op_not,
    [nir_op_ftrunc] = ppir_op_trunc,
    [nir_op_fsat] = ppir_op_sat,
-   [nir_op_fclamp_pos_mali] = ppir_op_clamp_pos,
+   [nir_op_fclamp_pos] = ppir_op_clamp_pos,
 };
 
 static bool ppir_emit_alu(ppir_block *block, nir_instr *ni)

src/panfrost/compiler/bifrost_compile.c

@@ -2643,13 +2643,13 @@ bi_emit_alu(bi_builder *b, nir_alu_instr *instr)
       break;
    }
 
-   case nir_op_fsat_signed_mali: {
+   case nir_op_fsat_signed: {
       bi_instr *I = bi_fclamp_to(b, sz, dst, s0);
       I->clamp = BI_CLAMP_CLAMP_M1_1;
       break;
    }
 
-   case nir_op_fclamp_pos_mali: {
+   case nir_op_fclamp_pos: {
       bi_instr *I = bi_fclamp_to(b, sz, dst, s0);
       I->clamp = BI_CLAMP_CLAMP_0_INF;
       break;

src/panfrost/compiler/bifrost_nir_algebraic.py

@@ -52,9 +52,9 @@ algebraic_late = [
     (('fmul', a, 2.0), ('fadd', a, a)),
 
     # Fuse Mali-specific clamps
-    (('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)),
+    (('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)),
 
     (('b32csel', 'b@32', ('iadd', 'a@32', 1), a), ('iadd', a, ('b2i32', b))),
 

src/panfrost/midgard/midgard_compile.c

@@ -832,8 +832,8 @@ 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_mali, fmov);
-      ALU_CASE(fclamp_pos_mali, fmov);
+      ALU_CASE(fsat_signed, fmov);
+      ALU_CASE(fclamp_pos, 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
@@ -934,9 +934,9 @@ 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_mali) {
+   } else if (instr->op == nir_op_fsat_signed) {
       outmod = midgard_outmod_clamp_m1_1;
-   } else if (instr->op == nir_op_fclamp_pos_mali) {
+   } else if (instr->op == nir_op_fclamp_pos) {
       outmod = midgard_outmod_clamp_0_inf;
    }
 

src/panfrost/midgard/midgard_nir_algebraic.py

@@ -51,11 +51,11 @@ 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_mali', a)),
-    (('~fmax', ('fmin', a, 1.0), -1.0), ('fsat_signed_mali', a)),
+    (('~fmin', ('fmax', a, -1.0), 1.0), ('fsat_signed', a)),
+    (('~fmax', ('fmin', a, 1.0), -1.0), ('fsat_signed', a)),
 
     # Fuse clamp_positive. This should probably be shared with Utgard/bifrost
-    (('fmax', a, 0.0), ('fclamp_pos_mali', a)),
+    (('fmax', a, 0.0), ('fclamp_pos', a)),
 
     (('ishl', 'a@16', b), ('u2u16', ('ishl', ('u2u32', a), b))),
     (('ishr', 'a@16', b), ('i2i16', ('ishr', ('i2i32', a), b))),

src/panfrost/util/pan_lower_framebuffer.c

@@ -202,7 +202,7 @@ static nir_def *
 pan_fsat(nir_builder *b, nir_def *v, bool is_signed)
 {
    if (is_signed)
-      return nir_fsat_signed_mali(b, v);
+      return nir_fsat_signed(b, v);
    else
       return nir_fsat(b, v);
 }