v3d: emit correct lowering for logic ops with integer render targets

Reviewed-by: Eric Anholt <eric@anholt.net>

diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index 4f12110ded198bc9d7a8ee3ad67e594bb50c96de..de80ab3f9e8bb029e4587cad8cec0c6a46f962e6 100644 (file)
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -1686,8 +1686,11 @@ vir_emit_tlb_color_read(struct v3d_compile *c, nir_intrinsic_instr *instr)
                 if (swap_rb)
                         num_components = MAX2(num_components, 3);
 
-                bool is_int_format = (c->fs_key->int_color_rb & (1 << rt)) ||
-                                     (c->fs_key->uint_color_rb & (1 << rt));
+                nir_variable *var = c->output_color_var[rt];
+                enum glsl_base_type type = glsl_get_base_type(var->type);
+
+                bool is_int_format = type == GLSL_TYPE_INT ||
+                                     type == GLSL_TYPE_UINT;
 
                 bool is_32b_tlb_format = is_int_format ||
                                          (c->fs_key->f32_color_rb & (1 << rt));

diff --git a/src/broadcom/compiler/v3d_nir_lower_logic_ops.c b/src/broadcom/compiler/v3d_nir_lower_logic_ops.c
index 849e554e8b7323a5b58dd70d519d1cacda3313b2..6da23df97317b1f5d297521d6fcb86f949d48527 100644 (file)
--- a/src/broadcom/compiler/v3d_nir_lower_logic_ops.c
+++ b/src/broadcom/compiler/v3d_nir_lower_logic_ops.c
@@ -164,17 +164,32 @@ v3d_nir_get_tlb_color(nir_builder *b, int rt, int sample)
 }
 
 static nir_ssa_def *
-v3d_nir_emit_logic_op(struct v3d_compile *c, nir_builder *b,
-                      nir_ssa_def *src, int rt, int sample)
+v3d_emit_logic_op_raw(struct v3d_compile *c, nir_builder *b,
+                      nir_ssa_def **src_chans, nir_ssa_def **dst_chans,
+                      int rt, int sample)
 {
-        nir_ssa_def *dst = v3d_nir_get_tlb_color(b, rt, sample);
+        const uint8_t *fmt_swz = v3d_get_format_swizzle_for_rt(c, rt);
 
-        nir_ssa_def *src_chans[4], *dst_chans[4];
-        for (unsigned i = 0; i < 4; i++) {
-                src_chans[i] = nir_channel(b, src, i);
-                dst_chans[i] = nir_channel(b, dst, i);
+        nir_ssa_def *op_res[4];
+        for (int i = 0; i < 4; i++) {
+                nir_ssa_def *src = src_chans[i];
+                nir_ssa_def *dst =
+                        v3d_nir_get_swizzled_channel(b, dst_chans, fmt_swz[i]);
+                op_res[i] = v3d_logicop(b, c->fs_key->logicop_func, src, dst);
         }
 
+        nir_ssa_def *r[4];
+        for (int i = 0; i < 4; i++)
+                r[i] = v3d_nir_get_swizzled_channel(b, op_res, fmt_swz[i]);
+
+        return nir_vec4(b, r[0], r[1], r[2], r[3]);
+}
+
+static nir_ssa_def *
+v3d_emit_logic_op_unorm(struct v3d_compile *c, nir_builder *b,
+                        nir_ssa_def **src_chans, nir_ssa_def **dst_chans,
+                        int rt, int sample)
+{
         const uint8_t src_swz[4] = { 0, 1, 2, 3 };
         nir_ssa_def *packed_src =
                 v3d_nir_swizzle_and_pack(b, src_chans, src_swz);
@@ -189,6 +204,26 @@ v3d_nir_emit_logic_op(struct v3d_compile *c, nir_builder *b,
         return v3d_nir_unpack_and_swizzle(b, packed_result, fmt_swz);
 }
 
+static nir_ssa_def *
+v3d_nir_emit_logic_op(struct v3d_compile *c, nir_builder *b,
+                      nir_ssa_def *src, int rt, int sample)
+{
+        nir_ssa_def *dst = v3d_nir_get_tlb_color(b, rt, sample);
+
+        nir_ssa_def *src_chans[4], *dst_chans[4];
+        for (unsigned i = 0; i < 4; i++) {
+                src_chans[i] = nir_channel(b, src, i);
+                dst_chans[i] = nir_channel(b, dst, i);
+        }
+
+        if (util_format_is_unorm(c->fs_key->color_fmt[rt].format)) {
+                return v3d_emit_logic_op_unorm(c, b, src_chans, dst_chans,
+                                               rt, 0);
+        } else {
+                return v3d_emit_logic_op_raw(c, b, src_chans, dst_chans, rt, 0);
+        }
+}
+
 static void
 v3d_nir_lower_logic_op_instr(struct v3d_compile *c,
                              nir_builder *b,