freedreno/a3xx/compiler: ensure min # of cycles after bary instr

The results of a bary.f do not appear to be immediatley available, but
there is no explicit sync bit.  Instead the compiler must just ensure
that there are a minimum number of instructions following the bary
before use of the result of the bary.  We aren't clever enough for that
so just throw in some nop's.

Signed-off-by: Rob Clark <robclark@freedesktop.org>

diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_compiler.c b/src/gallium/drivers/freedreno/a3xx/fd3_compiler.c
index cf3925a84b2352727d8b91fbf2e059492c2e7270..eabe21cb7e98cc56fb28291d1f29557999478d9a 100644 (file)
--- a/src/gallium/drivers/freedreno/a3xx/fd3_compiler.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_compiler.c
@@ -1072,12 +1072,13 @@ static const struct instr_translater translaters[TGSI_OPCODE_LAST] = {
        INSTR(END,          instr_cat0, .opc = OPC_END),
 };
 
-static void
+static int
 decl_in(struct fd3_compile_context *ctx, struct tgsi_full_declaration *decl)
 {
        struct fd3_shader_stateobj *so = ctx->so;
        unsigned base = ctx->base_reg[TGSI_FILE_INPUT];
        unsigned i, flags = 0;
+       int nop = 0;
 
        if (ctx->so->half_precision)
                flags |= IR3_REG_HALF;
@@ -1115,8 +1116,12 @@ decl_in(struct fd3_compile_context *ctx, struct tgsi_full_declaration *decl)
 
                        /* input base (always r0.x): */
                        ir3_reg_create(instr, regid(0,0), 0);
+
+                       nop = 6;
                }
        }
+
+       return nop;
 }
 
 static void
@@ -1174,6 +1179,7 @@ static void
 compile_instructions(struct fd3_compile_context *ctx)
 {
        struct ir3_shader *ir = ctx->ir;
+       int nop = 0;
 
        while (!tgsi_parse_end_of_tokens(&ctx->parser)) {
                tgsi_parse_token(&ctx->parser);
@@ -1185,7 +1191,7 @@ compile_instructions(struct fd3_compile_context *ctx)
                        if (decl->Declaration.File == TGSI_FILE_OUTPUT) {
                                decl_out(ctx, decl);
                        } else if (decl->Declaration.File == TGSI_FILE_INPUT) {
-                               decl_in(ctx, decl);
+                               nop = decl_in(ctx, decl);
                        } else if (decl->Declaration.File == TGSI_FILE_SAMPLER) {
                                decl_samp(ctx, decl);
                        }
@@ -1208,6 +1214,11 @@ compile_instructions(struct fd3_compile_context *ctx)
                        unsigned opc = inst->Instruction.Opcode;
                        const struct instr_translater *t = &translaters[opc];
 
+                       if (nop) {
+                               ir3_instr_create(ctx->ir, 0, OPC_NOP)->repeat = nop - 1;
+                               nop = 0;
+                       }
+
                        if (t->fxn) {
                                t->fxn(t, ctx, inst);
                                ctx->num_internal_temps = 0;